diff --git a/Eigen/Array b/Eigen/Array
deleted file mode 100644
index 3d004fb..0000000
--- a/Eigen/Array
+++ /dev/null
@@ -1,11 +0,0 @@
-#ifndef EIGEN_ARRAY_MODULE_H
-#define EIGEN_ARRAY_MODULE_H
-
-// include Core first to handle Eigen2 support macros
-#include "Core"
-
-#ifndef EIGEN2_SUPPORT
-  #error The Eigen/Array header does no longer exist in Eigen3. All that functionality has moved to Eigen/Core.
-#endif
-
-#endif // EIGEN_ARRAY_MODULE_H
diff --git a/Eigen/CMakeLists.txt b/Eigen/CMakeLists.txt
index a92dd6f..9eb502b 100644
--- a/Eigen/CMakeLists.txt
+++ b/Eigen/CMakeLists.txt
@@ -16,4 +16,4 @@
   DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel
   )
 
-add_subdirectory(src)
+install(DIRECTORY src DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel FILES_MATCHING PATTERN "*.h")
diff --git a/Eigen/Cholesky b/Eigen/Cholesky
index f727f5d..1332b54 100644
--- a/Eigen/Cholesky
+++ b/Eigen/Cholesky
@@ -1,7 +1,15 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_CHOLESKY_MODULE_H
 #define EIGEN_CHOLESKY_MODULE_H
 
 #include "Core"
+#include "Jacobi"
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
@@ -10,20 +18,26 @@
   *
   *
   * This module provides two variants of the Cholesky decomposition for selfadjoint (hermitian) matrices.
-  * Those decompositions are accessible via the following MatrixBase methods:
-  *  - MatrixBase::llt(),
+  * Those decompositions are also accessible via the following methods:
+  *  - MatrixBase::llt()
   *  - MatrixBase::ldlt()
+  *  - SelfAdjointView::llt()
+  *  - SelfAdjointView::ldlt()
   *
   * \code
   * #include <Eigen/Cholesky>
   * \endcode
   */
 
-#include "src/misc/Solve.h"
 #include "src/Cholesky/LLT.h"
 #include "src/Cholesky/LDLT.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/Cholesky/LLT_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/Cholesky/LLT_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/Eigen/CholmodSupport b/Eigen/CholmodSupport
index 745b884..bed8924 100644
--- a/Eigen/CholmodSupport
+++ b/Eigen/CholmodSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_CHOLMODSUPPORT_MODULE_H
 #define EIGEN_CHOLMODSUPPORT_MODULE_H
 
@@ -12,7 +19,7 @@
 /** \ingroup Support_modules
   * \defgroup CholmodSupport_Module CholmodSupport module
   *
-  * This module provides an interface to the Cholmod library which is part of the <a href="http://www.cise.ufl.edu/research/sparse/SuiteSparse/">suitesparse</a> package.
+  * This module provides an interface to the Cholmod library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
   * It provides the two following main factorization classes:
   * - class CholmodSupernodalLLT: a supernodal LLT Cholesky factorization.
   * - class CholmodDecomposiiton: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial).
@@ -33,12 +40,8 @@
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/CholmodSupport/CholmodSupport.h"
 
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_CHOLMODSUPPORT_MODULE_H
diff --git a/Eigen/Core b/Eigen/Core
index 509c529..b923b8c 100644
--- a/Eigen/Core
+++ b/Eigen/Core
@@ -14,6 +14,74 @@
 // first thing Eigen does: stop the compiler from committing suicide
 #include "src/Core/util/DisableStupidWarnings.h"
 
+#if defined(__CUDACC__) && !defined(EIGEN_NO_CUDA)
+  #define EIGEN_CUDACC __CUDACC__
+#endif
+
+#if defined(__CUDA_ARCH__) && !defined(EIGEN_NO_CUDA)
+  #define EIGEN_CUDA_ARCH __CUDA_ARCH__
+#endif
+
+#if defined(__CUDACC_VER_MAJOR__) && (__CUDACC_VER_MAJOR__ >= 9)
+#define EIGEN_CUDACC_VER  ((__CUDACC_VER_MAJOR__ * 10000) + (__CUDACC_VER_MINOR__ * 100))
+#elif defined(__CUDACC_VER__)
+#define EIGEN_CUDACC_VER __CUDACC_VER__
+#else
+#define EIGEN_CUDACC_VER 0
+#endif
+
+// Handle NVCC/CUDA/SYCL
+#if defined(__CUDACC__) || defined(__SYCL_DEVICE_ONLY__)
+  // Do not try asserts on CUDA and SYCL!
+  #ifndef EIGEN_NO_DEBUG
+  #define EIGEN_NO_DEBUG
+  #endif
+
+  #ifdef EIGEN_INTERNAL_DEBUGGING
+  #undef EIGEN_INTERNAL_DEBUGGING
+  #endif
+
+  #ifdef EIGEN_EXCEPTIONS
+  #undef EIGEN_EXCEPTIONS
+  #endif
+
+  // All functions callable from CUDA code must be qualified with __device__
+  #ifdef __CUDACC__
+    // Do not try to vectorize on CUDA and SYCL!
+    #ifndef EIGEN_DONT_VECTORIZE
+    #define EIGEN_DONT_VECTORIZE
+    #endif
+
+    #define EIGEN_DEVICE_FUNC __host__ __device__
+    // We need cuda_runtime.h to ensure that that EIGEN_USING_STD_MATH macro
+    // works properly on the device side
+    #include <cuda_runtime.h>
+  #else
+    #define EIGEN_DEVICE_FUNC
+  #endif
+
+#else
+  #define EIGEN_DEVICE_FUNC
+
+#endif
+
+// When compiling CUDA device code with NVCC, pull in math functions from the
+// global namespace.  In host mode, and when device doee with clang, use the
+// std versions.
+#if defined(__CUDA_ARCH__) && defined(__NVCC__)
+  #define EIGEN_USING_STD_MATH(FUNC) using ::FUNC;
+#else
+  #define EIGEN_USING_STD_MATH(FUNC) using std::FUNC;
+#endif
+
+#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(EIGEN_EXCEPTIONS) && !defined(EIGEN_USE_SYCL)
+  #define EIGEN_EXCEPTIONS
+#endif
+
+#ifdef EIGEN_EXCEPTIONS
+  #include <new>
+#endif
+
 // then include this file where all our macros are defined. It's really important to do it first because
 // it's where we do all the alignment settings (platform detection and honoring the user's will if he
 // defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization.
@@ -21,7 +89,7 @@
 
 // Disable the ipa-cp-clone optimization flag with MinGW 6.x or newer (enabled by default with -O3)
 // See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=556 for details.
-#if defined(__MINGW32__) && EIGEN_GNUC_AT_LEAST(4,6)
+#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6)
   #pragma GCC optimize ("-fno-ipa-cp-clone")
 #endif
 
@@ -31,26 +99,26 @@
 // and inclusion of their respective header files
 #include "src/Core/util/MKL_support.h"
 
-// if alignment is disabled, then disable vectorization. Note: EIGEN_ALIGN is the proper check, it takes into
-// account both the user's will (EIGEN_DONT_ALIGN) and our own platform checks
-#if !EIGEN_ALIGN
+// if alignment is disabled, then disable vectorization. Note: EIGEN_MAX_ALIGN_BYTES is the proper check, it takes into
+// account both the user's will (EIGEN_MAX_ALIGN_BYTES,EIGEN_DONT_ALIGN) and our own platform checks
+#if EIGEN_MAX_ALIGN_BYTES==0
   #ifndef EIGEN_DONT_VECTORIZE
     #define EIGEN_DONT_VECTORIZE
   #endif
 #endif
 
-#ifdef _MSC_VER
+#if EIGEN_COMP_MSVC
   #include <malloc.h> // for _aligned_malloc -- need it regardless of whether vectorization is enabled
-  #if (_MSC_VER >= 1500) // 2008 or later
+  #if (EIGEN_COMP_MSVC >= 1500) // 2008 or later
     // Remember that usage of defined() in a #define is undefined by the standard.
     // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP.
-    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64)
+    #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || EIGEN_ARCH_x86_64
       #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER
     #endif
   #endif
 #else
   // Remember that usage of defined() in a #define is undefined by the standard
-  #if (defined __SSE2__) && ( (!defined __GNUC__) || (defined __INTEL_COMPILER) || EIGEN_GNUC_AT_LEAST(4,2) )
+  #if (defined __SSE2__) && ( (!EIGEN_COMP_GNUC) || EIGEN_COMP_ICC || EIGEN_GNUC_AT_LEAST(4,2) )
     #define EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC
   #endif
 #endif
@@ -82,6 +150,31 @@
     #ifdef __SSE4_2__
       #define EIGEN_VECTORIZE_SSE4_2
     #endif
+    #ifdef __AVX__
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_SSE3
+      #define EIGEN_VECTORIZE_SSSE3
+      #define EIGEN_VECTORIZE_SSE4_1
+      #define EIGEN_VECTORIZE_SSE4_2
+    #endif
+    #ifdef __AVX2__
+      #define EIGEN_VECTORIZE_AVX2
+    #endif
+    #ifdef __FMA__
+      #define EIGEN_VECTORIZE_FMA
+    #endif
+    #if defined(__AVX512F__) && defined(EIGEN_ENABLE_AVX512)
+      #define EIGEN_VECTORIZE_AVX512
+      #define EIGEN_VECTORIZE_AVX2
+      #define EIGEN_VECTORIZE_AVX
+      #define EIGEN_VECTORIZE_FMA
+      #ifdef __AVX512DQ__
+        #define EIGEN_VECTORIZE_AVX512DQ
+      #endif
+      #ifdef __AVX512ER__
+        #define EIGEN_VECTORIZE_AVX512ER
+      #endif
+    #endif
 
     // include files
 
@@ -95,9 +188,10 @@
     extern "C" {
       // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly.
       // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus:
-      #if defined(__INTEL_COMPILER) && __INTEL_COMPILER >= 1110
+      #if EIGEN_COMP_ICC >= 1110
         #include <immintrin.h>
       #else
+        #include <mmintrin.h>
         #include <emmintrin.h>
         #include <xmmintrin.h>
         #ifdef  EIGEN_VECTORIZE_SSE3
@@ -112,8 +206,20 @@
         #ifdef EIGEN_VECTORIZE_SSE4_2
         #include <nmmintrin.h>
         #endif
+        #if defined(EIGEN_VECTORIZE_AVX) || defined(EIGEN_VECTORIZE_AVX512)
+        #include <immintrin.h>
+        #endif
       #endif
     } // end extern "C"
+  #elif defined __VSX__
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_VSX
+    #include <altivec.h>
+    // We need to #undef all these ugly tokens defined in <altivec.h>
+    // => use __vector instead of vector
+    #undef bool
+    #undef vector
+    #undef pixel
   #elif defined __ALTIVEC__
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_ALTIVEC
@@ -123,13 +229,35 @@
     #undef bool
     #undef vector
     #undef pixel
-  #elif defined  __ARM_NEON
+  #elif (defined  __ARM_NEON) || (defined __ARM_NEON__)
     #define EIGEN_VECTORIZE
     #define EIGEN_VECTORIZE_NEON
     #include <arm_neon.h>
+  #elif (defined __s390x__ && defined __VEC__)
+    #define EIGEN_VECTORIZE
+    #define EIGEN_VECTORIZE_ZVECTOR
+    #include <vecintrin.h>
   #endif
 #endif
 
+#if defined(__F16C__) && !defined(EIGEN_COMP_CLANG)
+  // We can use the optimized fp16 to float and float to fp16 conversion routines
+  #define EIGEN_HAS_FP16_C
+#endif
+
+#if defined __CUDACC__
+  #define EIGEN_VECTORIZE_CUDA
+  #include <vector_types.h>
+  #if EIGEN_CUDACC_VER >= 70500
+    #define EIGEN_HAS_CUDA_FP16
+  #endif
+#endif
+
+#if defined EIGEN_HAS_CUDA_FP16
+  #include <host_defines.h>
+  #include <cuda_fp16.h>
+#endif
+
 #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE)
   #define EIGEN_HAS_OPENMP
 #endif
@@ -139,7 +267,7 @@
 #endif
 
 // MSVC for windows mobile does not have the errno.h file
-#if !(defined(_MSC_VER) && defined(_WIN32_WCE)) && !defined(__ARMCC_VERSION)
+#if !(EIGEN_COMP_MSVC && EIGEN_OS_WINCE) && !EIGEN_COMP_ARM
 #define EIGEN_HAS_ERRNO
 #endif
 
@@ -159,29 +287,30 @@
 // for min/max:
 #include <algorithm>
 
+// for std::is_nothrow_move_assignable
+#ifdef EIGEN_INCLUDE_TYPE_TRAITS
+#include <type_traits>
+#endif
+
 // for outputting debug info
 #ifdef EIGEN_DEBUG_ASSIGN
 #include <iostream>
 #endif
 
 // required for __cpuid, needs to be included after cmath
-#if defined(_MSC_VER) && (defined(_M_IX86)||defined(_M_X64)) && (!defined(_WIN32_WCE))
+#if EIGEN_COMP_MSVC && EIGEN_ARCH_i386_OR_x86_64 && !EIGEN_OS_WINCE
   #include <intrin.h>
 #endif
 
-#if defined(_CPPUNWIND) || defined(__EXCEPTIONS)
-  #define EIGEN_EXCEPTIONS
-#endif
-
-#ifdef EIGEN_EXCEPTIONS
-  #include <new>
-#endif
-
 /** \brief Namespace containing all symbols from the %Eigen library. */
 namespace Eigen {
 
 inline static const char *SimdInstructionSetsInUse(void) {
-#if defined(EIGEN_VECTORIZE_SSE4_2)
+#if defined(EIGEN_VECTORIZE_AVX512)
+  return "AVX512, FMA, AVX2, AVX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_AVX)
+  return "AVX SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
+#elif defined(EIGEN_VECTORIZE_SSE4_2)
   return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2";
 #elif defined(EIGEN_VECTORIZE_SSE4_1)
   return "SSE, SSE2, SSE3, SSSE3, SSE4.1";
@@ -193,8 +322,12 @@
   return "SSE, SSE2";
 #elif defined(EIGEN_VECTORIZE_ALTIVEC)
   return "AltiVec";
+#elif defined(EIGEN_VECTORIZE_VSX)
+  return "VSX";
 #elif defined(EIGEN_VECTORIZE_NEON)
   return "ARM NEON";
+#elif defined(EIGEN_VECTORIZE_ZVECTOR)
+  return "S390X ZVECTOR";
 #else
   return "None";
 #endif
@@ -202,42 +335,21 @@
 
 } // end namespace Eigen
 
-#define STAGE10_FULL_EIGEN2_API             10
-#define STAGE20_RESOLVE_API_CONFLICTS       20
-#define STAGE30_FULL_EIGEN3_API             30
-#define STAGE40_FULL_EIGEN3_STRICTNESS      40
-#define STAGE99_NO_EIGEN2_SUPPORT           99
-
-#if   defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE40_FULL_EIGEN3_STRICTNESS
-#elif defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API
-#elif defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE20_RESOLVE_API_CONFLICTS
-#elif defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API
-  #define EIGEN2_SUPPORT
-  #define EIGEN2_SUPPORT_STAGE STAGE10_FULL_EIGEN2_API
-#elif defined EIGEN2_SUPPORT
-  // default to stage 3, that's what it's always meant
-  #define EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API
-  #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API
-#else
-  #define EIGEN2_SUPPORT_STAGE STAGE99_NO_EIGEN2_SUPPORT
+#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS || defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API || defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS || defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API || defined EIGEN2_SUPPORT
+// This will generate an error message:
+#error Eigen2-support is only available up to version 3.2. Please go to "http://eigen.tuxfamily.org/index.php?title=Eigen2" for further information
 #endif
 
-#ifdef EIGEN2_SUPPORT
-#undef minor
-#endif
+namespace Eigen {
 
 // we use size_t frequently and we'll never remember to prepend it with std:: everytime just to
 // ensure QNX/QCC support
 using std::size_t;
-// gcc 4.6.0 wants std:: for ptrdiff_t 
+// gcc 4.6.0 wants std:: for ptrdiff_t
 using std::ptrdiff_t;
 
+}
+
 /** \defgroup Core_Module Core module
   * This is the main module of Eigen providing dense matrix and vector support
   * (both fixed and dynamic size) with all the features corresponding to a BLAS library
@@ -249,8 +361,8 @@
   */
 
 #include "src/Core/util/Constants.h"
-#include "src/Core/util/ForwardDeclarations.h"
 #include "src/Core/util/Meta.h"
+#include "src/Core/util/ForwardDeclarations.h"
 #include "src/Core/util/StaticAssert.h"
 #include "src/Core/util/XprHelper.h"
 #include "src/Core/util/Memory.h"
@@ -258,41 +370,94 @@
 #include "src/Core/NumTraits.h"
 #include "src/Core/MathFunctions.h"
 #include "src/Core/GenericPacketMath.h"
+#include "src/Core/MathFunctionsImpl.h"
+#include "src/Core/arch/Default/ConjHelper.h"
 
-#if defined EIGEN_VECTORIZE_SSE
+#if defined EIGEN_VECTORIZE_AVX512
+  #include "src/Core/arch/SSE/PacketMath.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
+  #include "src/Core/arch/AVX512/PacketMath.h"
+  #include "src/Core/arch/AVX512/MathFunctions.h"
+#elif defined EIGEN_VECTORIZE_AVX
+  // Use AVX for floats and doubles, SSE for integers
+  #include "src/Core/arch/SSE/PacketMath.h"
+  #include "src/Core/arch/SSE/Complex.h"
+  #include "src/Core/arch/SSE/MathFunctions.h"
+  #include "src/Core/arch/AVX/PacketMath.h"
+  #include "src/Core/arch/AVX/MathFunctions.h"
+  #include "src/Core/arch/AVX/Complex.h"
+  #include "src/Core/arch/AVX/TypeCasting.h"
+  #include "src/Core/arch/SSE/TypeCasting.h"
+#elif defined EIGEN_VECTORIZE_SSE
   #include "src/Core/arch/SSE/PacketMath.h"
   #include "src/Core/arch/SSE/MathFunctions.h"
   #include "src/Core/arch/SSE/Complex.h"
-#elif defined EIGEN_VECTORIZE_ALTIVEC
+  #include "src/Core/arch/SSE/TypeCasting.h"
+#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX)
   #include "src/Core/arch/AltiVec/PacketMath.h"
+  #include "src/Core/arch/AltiVec/MathFunctions.h"
   #include "src/Core/arch/AltiVec/Complex.h"
 #elif defined EIGEN_VECTORIZE_NEON
   #include "src/Core/arch/NEON/PacketMath.h"
+  #include "src/Core/arch/NEON/MathFunctions.h"
   #include "src/Core/arch/NEON/Complex.h"
+#elif defined EIGEN_VECTORIZE_ZVECTOR
+  #include "src/Core/arch/ZVector/PacketMath.h"
+  #include "src/Core/arch/ZVector/MathFunctions.h"
+  #include "src/Core/arch/ZVector/Complex.h"
+#endif
+
+// Half float support
+#include "src/Core/arch/CUDA/Half.h"
+#include "src/Core/arch/CUDA/PacketMathHalf.h"
+#include "src/Core/arch/CUDA/TypeCasting.h"
+
+#if defined EIGEN_VECTORIZE_CUDA
+  #include "src/Core/arch/CUDA/PacketMath.h"
+  #include "src/Core/arch/CUDA/MathFunctions.h"
 #endif
 
 #include "src/Core/arch/Default/Settings.h"
 
-#include "src/Core/Functors.h"
+#include "src/Core/functors/TernaryFunctors.h"
+#include "src/Core/functors/BinaryFunctors.h"
+#include "src/Core/functors/UnaryFunctors.h"
+#include "src/Core/functors/NullaryFunctors.h"
+#include "src/Core/functors/StlFunctors.h"
+#include "src/Core/functors/AssignmentFunctors.h"
+
+// Specialized functors to enable the processing of complex numbers
+// on CUDA devices
+#include "src/Core/arch/CUDA/Complex.h"
+
+#include "src/Core/IO.h"
 #include "src/Core/DenseCoeffsBase.h"
 #include "src/Core/DenseBase.h"
 #include "src/Core/MatrixBase.h"
 #include "src/Core/EigenBase.h"
 
+#include "src/Core/Product.h"
+#include "src/Core/CoreEvaluators.h"
+#include "src/Core/AssignEvaluator.h"
+
 #ifndef EIGEN_PARSED_BY_DOXYGEN // work around Doxygen bug triggered by Assign.h r814874
                                 // at least confirmed with Doxygen 1.5.5 and 1.5.6
   #include "src/Core/Assign.h"
 #endif
 
+#include "src/Core/ArrayBase.h"
 #include "src/Core/util/BlasUtil.h"
 #include "src/Core/DenseStorage.h"
 #include "src/Core/NestByValue.h"
-#include "src/Core/ForceAlignedAccess.h"
+
+// #include "src/Core/ForceAlignedAccess.h"
+
 #include "src/Core/ReturnByValue.h"
 #include "src/Core/NoAlias.h"
 #include "src/Core/PlainObjectBase.h"
 #include "src/Core/Matrix.h"
 #include "src/Core/Array.h"
+#include "src/Core/CwiseTernaryOp.h"
 #include "src/Core/CwiseBinaryOp.h"
 #include "src/Core/CwiseUnaryOp.h"
 #include "src/Core/CwiseNullaryOp.h"
@@ -300,32 +465,32 @@
 #include "src/Core/SelfCwiseBinaryOp.h"
 #include "src/Core/Dot.h"
 #include "src/Core/StableNorm.h"
-#include "src/Core/MapBase.h"
 #include "src/Core/Stride.h"
+#include "src/Core/MapBase.h"
 #include "src/Core/Map.h"
+#include "src/Core/Ref.h"
 #include "src/Core/Block.h"
 #include "src/Core/VectorBlock.h"
-#include "src/Core/Ref.h"
 #include "src/Core/Transpose.h"
 #include "src/Core/DiagonalMatrix.h"
 #include "src/Core/Diagonal.h"
 #include "src/Core/DiagonalProduct.h"
-#include "src/Core/PermutationMatrix.h"
-#include "src/Core/Transpositions.h"
 #include "src/Core/Redux.h"
 #include "src/Core/Visitor.h"
 #include "src/Core/Fuzzy.h"
-#include "src/Core/IO.h"
 #include "src/Core/Swap.h"
 #include "src/Core/CommaInitializer.h"
-#include "src/Core/Flagged.h"
-#include "src/Core/ProductBase.h"
 #include "src/Core/GeneralProduct.h"
+#include "src/Core/Solve.h"
+#include "src/Core/Inverse.h"
+#include "src/Core/SolverBase.h"
+#include "src/Core/PermutationMatrix.h"
+#include "src/Core/Transpositions.h"
 #include "src/Core/TriangularMatrix.h"
 #include "src/Core/SelfAdjointView.h"
 #include "src/Core/products/GeneralBlockPanelKernel.h"
 #include "src/Core/products/Parallelizer.h"
-#include "src/Core/products/CoeffBasedProduct.h"
+#include "src/Core/ProductEvaluators.h"
 #include "src/Core/products/GeneralMatrixVector.h"
 #include "src/Core/products/GeneralMatrixMatrix.h"
 #include "src/Core/SolveTriangular.h"
@@ -340,6 +505,7 @@
 #include "src/Core/products/TriangularSolverVector.h"
 #include "src/Core/BandMatrix.h"
 #include "src/Core/CoreIterators.h"
+#include "src/Core/ConditionEstimator.h"
 
 #include "src/Core/BooleanRedux.h"
 #include "src/Core/Select.h"
@@ -347,18 +513,17 @@
 #include "src/Core/Random.h"
 #include "src/Core/Replicate.h"
 #include "src/Core/Reverse.h"
-#include "src/Core/ArrayBase.h"
 #include "src/Core/ArrayWrapper.h"
 
 #ifdef EIGEN_USE_BLAS
-#include "src/Core/products/GeneralMatrixMatrix_MKL.h"
-#include "src/Core/products/GeneralMatrixVector_MKL.h"
-#include "src/Core/products/GeneralMatrixMatrixTriangular_MKL.h"
-#include "src/Core/products/SelfadjointMatrixMatrix_MKL.h"
-#include "src/Core/products/SelfadjointMatrixVector_MKL.h"
-#include "src/Core/products/TriangularMatrixMatrix_MKL.h"
-#include "src/Core/products/TriangularMatrixVector_MKL.h"
-#include "src/Core/products/TriangularSolverMatrix_MKL.h"
+#include "src/Core/products/GeneralMatrixMatrix_BLAS.h"
+#include "src/Core/products/GeneralMatrixVector_BLAS.h"
+#include "src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h"
+#include "src/Core/products/SelfadjointMatrixMatrix_BLAS.h"
+#include "src/Core/products/SelfadjointMatrixVector_BLAS.h"
+#include "src/Core/products/TriangularMatrixMatrix_BLAS.h"
+#include "src/Core/products/TriangularMatrixVector_BLAS.h"
+#include "src/Core/products/TriangularSolverMatrix_BLAS.h"
 #endif // EIGEN_USE_BLAS
 
 #ifdef EIGEN_USE_MKL_VML
@@ -369,8 +534,4 @@
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
-#ifdef EIGEN2_SUPPORT
-#include "Eigen2Support"
-#endif
-
 #endif // EIGEN_CORE_H
diff --git a/Eigen/Eigen b/Eigen/Eigen
index 19b40ea..654c8dc 100644
--- a/Eigen/Eigen
+++ b/Eigen/Eigen
@@ -1,2 +1,2 @@
 #include "Dense"
-//#include "Sparse"
+#include "Sparse"
diff --git a/Eigen/Eigen2Support b/Eigen/Eigen2Support
deleted file mode 100644
index 6aa009d..0000000
--- a/Eigen/Eigen2Support
+++ /dev/null
@@ -1,95 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2SUPPORT_H
-#define EIGEN2SUPPORT_H
-
-#if (!defined(EIGEN2_SUPPORT)) || (!defined(EIGEN_CORE_H))
-#error Eigen2 support must be enabled by defining EIGEN2_SUPPORT before including any Eigen header
-#endif
-
-#ifndef EIGEN_NO_EIGEN2_DEPRECATED_WARNING
-
-#if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__clang__)
-#warning "Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3. (Define EIGEN_NO_EIGEN2_DEPRECATED_WARNING to disable this warning)"
-#else
-#pragma message ("Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3. (Define EIGEN_NO_EIGEN2_DEPRECATED_WARNING to disable this warning)")
-#endif
-
-#endif // EIGEN_NO_EIGEN2_DEPRECATED_WARNING
-
-#include "src/Core/util/DisableStupidWarnings.h"
-
-/** \ingroup Support_modules
-  * \defgroup Eigen2Support_Module Eigen2 support module
-  *
-  * \warning Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3.
-  *
-  * This module provides a couple of deprecated functions improving the compatibility with Eigen2.
-  * 
-  * To use it, define EIGEN2_SUPPORT before including any Eigen header
-  * \code
-  * #define EIGEN2_SUPPORT
-  * \endcode
-  *
-  */
-
-#include "src/Eigen2Support/Macros.h"
-#include "src/Eigen2Support/Memory.h"
-#include "src/Eigen2Support/Meta.h"
-#include "src/Eigen2Support/Lazy.h"
-#include "src/Eigen2Support/Cwise.h"
-#include "src/Eigen2Support/CwiseOperators.h"
-#include "src/Eigen2Support/TriangularSolver.h"
-#include "src/Eigen2Support/Block.h"
-#include "src/Eigen2Support/VectorBlock.h"
-#include "src/Eigen2Support/Minor.h"
-#include "src/Eigen2Support/MathFunctions.h"
-
-
-#include "src/Core/util/ReenableStupidWarnings.h"
-
-// Eigen2 used to include iostream
-#include<iostream>
-
-#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \
-using Eigen::Matrix##SizeSuffix##TypeSuffix; \
-using Eigen::Vector##SizeSuffix##TypeSuffix; \
-using Eigen::RowVector##SizeSuffix##TypeSuffix;
-
-#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(TypeSuffix) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \
-
-#define EIGEN_USING_MATRIX_TYPEDEFS \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(i) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(f) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(d) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cf) \
-EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cd)
-
-#define USING_PART_OF_NAMESPACE_EIGEN \
-EIGEN_USING_MATRIX_TYPEDEFS \
-using Eigen::Matrix; \
-using Eigen::MatrixBase; \
-using Eigen::ei_random; \
-using Eigen::ei_real; \
-using Eigen::ei_imag; \
-using Eigen::ei_conj; \
-using Eigen::ei_abs; \
-using Eigen::ei_abs2; \
-using Eigen::ei_sqrt; \
-using Eigen::ei_exp; \
-using Eigen::ei_log; \
-using Eigen::ei_sin; \
-using Eigen::ei_cos;
-
-#endif // EIGEN2SUPPORT_H
diff --git a/Eigen/Eigenvalues b/Eigen/Eigenvalues
index 53c5a73..f3f661b 100644
--- a/Eigen/Eigenvalues
+++ b/Eigen/Eigenvalues
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_EIGENVALUES_MODULE_H
 #define EIGEN_EIGENVALUES_MODULE_H
 
@@ -25,6 +32,7 @@
   * \endcode
   */
 
+#include "src/misc/RealSvd2x2.h"
 #include "src/Eigenvalues/Tridiagonalization.h"
 #include "src/Eigenvalues/RealSchur.h"
 #include "src/Eigenvalues/EigenSolver.h"
@@ -37,9 +45,14 @@
 #include "src/Eigenvalues/GeneralizedEigenSolver.h"
 #include "src/Eigenvalues/MatrixBaseEigenvalues.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/Eigenvalues/RealSchur_MKL.h"
-#include "src/Eigenvalues/ComplexSchur_MKL.h"
-#include "src/Eigenvalues/SelfAdjointEigenSolver_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/Eigenvalues/RealSchur_LAPACKE.h"
+#include "src/Eigenvalues/ComplexSchur_LAPACKE.h"
+#include "src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/Eigen/Geometry b/Eigen/Geometry
index efd9d45..716d529 100644
--- a/Eigen/Geometry
+++ b/Eigen/Geometry
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_GEOMETRY_MODULE_H
 #define EIGEN_GEOMETRY_MODULE_H
 
@@ -9,21 +16,17 @@
 #include "LU"
 #include <limits>
 
-#ifndef M_PI
-#define M_PI 3.14159265358979323846
-#endif
-
 /** \defgroup Geometry_Module Geometry module
   *
-  *
-  *
   * This module provides support for:
   *  - fixed-size homogeneous transformations
   *  - translation, scaling, 2D and 3D rotations
-  *  - quaternions
-  *  - \ref MatrixBase::cross() "cross product"
-  *  - \ref MatrixBase::unitOrthogonal() "orthognal vector generation"
-  *  - some linear components: parametrized-lines and hyperplanes
+  *  - \link Quaternion quaternions \endlink
+  *  - cross products (\ref MatrixBase::cross, \ref MatrixBase::cross3)
+  *  - orthognal vector generation (\ref MatrixBase::unitOrthogonal)
+  *  - some linear components: \link ParametrizedLine parametrized-lines \endlink and \link Hyperplane hyperplanes \endlink
+  *  - \link AlignedBox axis aligned bounding boxes \endlink
+  *  - \link umeyama least-square transformation fitting \endlink
   *
   * \code
   * #include <Eigen/Geometry>
@@ -33,27 +36,23 @@
 #include "src/Geometry/OrthoMethods.h"
 #include "src/Geometry/EulerAngles.h"
 
-#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-  #include "src/Geometry/Homogeneous.h"
-  #include "src/Geometry/RotationBase.h"
-  #include "src/Geometry/Rotation2D.h"
-  #include "src/Geometry/Quaternion.h"
-  #include "src/Geometry/AngleAxis.h"
-  #include "src/Geometry/Transform.h"
-  #include "src/Geometry/Translation.h"
-  #include "src/Geometry/Scaling.h"
-  #include "src/Geometry/Hyperplane.h"
-  #include "src/Geometry/ParametrizedLine.h"
-  #include "src/Geometry/AlignedBox.h"
-  #include "src/Geometry/Umeyama.h"
+#include "src/Geometry/Homogeneous.h"
+#include "src/Geometry/RotationBase.h"
+#include "src/Geometry/Rotation2D.h"
+#include "src/Geometry/Quaternion.h"
+#include "src/Geometry/AngleAxis.h"
+#include "src/Geometry/Transform.h"
+#include "src/Geometry/Translation.h"
+#include "src/Geometry/Scaling.h"
+#include "src/Geometry/Hyperplane.h"
+#include "src/Geometry/ParametrizedLine.h"
+#include "src/Geometry/AlignedBox.h"
+#include "src/Geometry/Umeyama.h"
 
-  #if defined EIGEN_VECTORIZE_SSE
-    #include "src/Geometry/arch/Geometry_SSE.h"
-  #endif
-#endif
-
-#ifdef EIGEN2_SUPPORT
-#include "src/Eigen2Support/Geometry/All.h"
+// Use the SSE optimized version whenever possible. At the moment the
+// SSE version doesn't compile when AVX is enabled
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
+#include "src/Geometry/arch/Geometry_SSE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/Eigen/Householder b/Eigen/Householder
index 6e348db..89cd81b 100644
--- a/Eigen/Householder
+++ b/Eigen/Householder
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_HOUSEHOLDER_MODULE_H
 #define EIGEN_HOUSEHOLDER_MODULE_H
 
diff --git a/Eigen/IterativeLinearSolvers b/Eigen/IterativeLinearSolvers
index 0f4159d..957d575 100644
--- a/Eigen/IterativeLinearSolvers
+++ b/Eigen/IterativeLinearSolvers
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
 #define EIGEN_ITERATIVELINEARSOLVERS_MODULE_H
 
@@ -12,28 +19,29 @@
   * This module currently provides iterative methods to solve problems of the form \c A \c x = \c b, where \c A is a squared matrix, usually very large and sparse.
   * Those solvers are accessible via the following classes:
   *  - ConjugateGradient for selfadjoint (hermitian) matrices,
+  *  - LeastSquaresConjugateGradient for rectangular least-square problems,
   *  - BiCGSTAB for general square matrices.
   *
   * These iterative solvers are associated with some preconditioners:
   *  - IdentityPreconditioner - not really useful
-  *  - DiagonalPreconditioner - also called JAcobi preconditioner, work very well on diagonal dominant matrices.
-  *  - IncompleteILUT - incomplete LU factorization with dual thresholding
+  *  - DiagonalPreconditioner - also called Jacobi preconditioner, work very well on diagonal dominant matrices.
+  *  - IncompleteLUT - incomplete LU factorization with dual thresholding
   *
   * Such problems can also be solved using the direct sparse decomposition modules: SparseCholesky, CholmodSupport, UmfPackSupport, SuperLUSupport.
   *
-  * \code
-  * #include <Eigen/IterativeLinearSolvers>
-  * \endcode
+    \code
+    #include <Eigen/IterativeLinearSolvers>
+    \endcode
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
+#include "src/IterativeLinearSolvers/SolveWithGuess.h"
 #include "src/IterativeLinearSolvers/IterativeSolverBase.h"
 #include "src/IterativeLinearSolvers/BasicPreconditioners.h"
 #include "src/IterativeLinearSolvers/ConjugateGradient.h"
+#include "src/IterativeLinearSolvers/LeastSquareConjugateGradient.h"
 #include "src/IterativeLinearSolvers/BiCGSTAB.h"
 #include "src/IterativeLinearSolvers/IncompleteLUT.h"
+#include "src/IterativeLinearSolvers/IncompleteCholesky.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
diff --git a/Eigen/Jacobi b/Eigen/Jacobi
index ba8a4dc..17c1d78 100644
--- a/Eigen/Jacobi
+++ b/Eigen/Jacobi
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_JACOBI_MODULE_H
 #define EIGEN_JACOBI_MODULE_H
 
diff --git a/Eigen/LU b/Eigen/LU
index db57955..6418a86 100644
--- a/Eigen/LU
+++ b/Eigen/LU
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_LU_MODULE_H
 #define EIGEN_LU_MODULE_H
 
@@ -16,25 +23,27 @@
   * \endcode
   */
 
-#include "src/misc/Solve.h"
 #include "src/misc/Kernel.h"
 #include "src/misc/Image.h"
 #include "src/LU/FullPivLU.h"
 #include "src/LU/PartialPivLU.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/LU/PartialPivLU_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/LU/PartialPivLU_LAPACKE.h"
 #endif
 #include "src/LU/Determinant.h"
-#include "src/LU/Inverse.h"
+#include "src/LU/InverseImpl.h"
 
-#if defined EIGEN_VECTORIZE_SSE
+// Use the SSE optimized version whenever possible. At the moment the
+// SSE version doesn't compile when AVX is enabled
+#if defined EIGEN_VECTORIZE_SSE && !defined EIGEN_VECTORIZE_AVX
   #include "src/LU/arch/Inverse_SSE.h"
 #endif
 
-#ifdef EIGEN2_SUPPORT
-  #include "src/Eigen2Support/LU.h"
-#endif
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_LU_MODULE_H
diff --git a/Eigen/LeastSquares b/Eigen/LeastSquares
deleted file mode 100644
index 35137c2..0000000
--- a/Eigen/LeastSquares
+++ /dev/null
@@ -1,32 +0,0 @@
-#ifndef EIGEN_REGRESSION_MODULE_H
-#define EIGEN_REGRESSION_MODULE_H
-
-#ifndef EIGEN2_SUPPORT
-#error LeastSquares is only available in Eigen2 support mode (define EIGEN2_SUPPORT)
-#endif
-
-// exclude from normal eigen3-only documentation
-#ifdef EIGEN2_SUPPORT
-
-#include "Core"
-
-#include "src/Core/util/DisableStupidWarnings.h"
-
-#include "Eigenvalues"
-#include "Geometry"
-
-/** \defgroup LeastSquares_Module LeastSquares module
-  * This module provides linear regression and related features.
-  *
-  * \code
-  * #include <Eigen/LeastSquares>
-  * \endcode
-  */
-
-#include "src/Eigen2Support/LeastSquares.h"
-
-#include "src/Core/util/ReenableStupidWarnings.h"
-
-#endif // EIGEN2_SUPPORT
-
-#endif // EIGEN_REGRESSION_MODULE_H
diff --git a/Eigen/MetisSupport b/Eigen/MetisSupport
index 6a113f7..85c41bf 100644
--- a/Eigen/MetisSupport
+++ b/Eigen/MetisSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_METISSUPPORT_MODULE_H
 #define EIGEN_METISSUPPORT_MODULE_H
 
diff --git a/Eigen/OrderingMethods b/Eigen/OrderingMethods
index 7c0f1ff..d8ea361 100644
--- a/Eigen/OrderingMethods
+++ b/Eigen/OrderingMethods
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_ORDERINGMETHODS_MODULE_H
 #define EIGEN_ORDERINGMETHODS_MODULE_H
 
diff --git a/Eigen/PaStiXSupport b/Eigen/PaStiXSupport
index 7c616ee..de3a63b 100644
--- a/Eigen/PaStiXSupport
+++ b/Eigen/PaStiXSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_PASTIXSUPPORT_MODULE_H
 #define EIGEN_PASTIXSUPPORT_MODULE_H
 
@@ -5,7 +12,6 @@
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
-#include <complex.h>
 extern "C" {
 #include <pastix_nompi.h>
 #include <pastix.h>
@@ -35,12 +41,8 @@
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/PaStiXSupport/PaStiXSupport.h"
 
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_PASTIXSUPPORT_MODULE_H
diff --git a/Eigen/PardisoSupport b/Eigen/PardisoSupport
old mode 100644
new mode 100755
index 99330ce..340edf5
--- a/Eigen/PardisoSupport
+++ b/Eigen/PardisoSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_PARDISOSUPPORT_MODULE_H
 #define EIGEN_PARDISOSUPPORT_MODULE_H
 
@@ -7,8 +14,6 @@
 
 #include <mkl_pardiso.h>
 
-#include <unsupported/Eigen/SparseExtra>
-
 /** \ingroup Support_modules
   * \defgroup PardisoSupport_Module PardisoSupport module
   *
diff --git a/Eigen/QR b/Eigen/QR
index ac5b026..c7e9144 100644
--- a/Eigen/QR
+++ b/Eigen/QR
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_QR_MODULE_H
 #define EIGEN_QR_MODULE_H
 
@@ -15,31 +22,30 @@
   *
   * This module provides various QR decompositions
   * This module also provides some MatrixBase methods, including:
-  *  - MatrixBase::qr(),
+  *  - MatrixBase::householderQr()
+  *  - MatrixBase::colPivHouseholderQr()
+  *  - MatrixBase::fullPivHouseholderQr()
   *
   * \code
   * #include <Eigen/QR>
   * \endcode
   */
 
-#include "src/misc/Solve.h"
 #include "src/QR/HouseholderQR.h"
 #include "src/QR/FullPivHouseholderQR.h"
 #include "src/QR/ColPivHouseholderQR.h"
+#include "src/QR/CompleteOrthogonalDecomposition.h"
 #ifdef EIGEN_USE_LAPACKE
-#include "src/QR/HouseholderQR_MKL.h"
-#include "src/QR/ColPivHouseholderQR_MKL.h"
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
 #endif
-
-#ifdef EIGEN2_SUPPORT
-#include "src/Eigen2Support/QR.h"
+#include "src/QR/HouseholderQR_LAPACKE.h"
+#include "src/QR/ColPivHouseholderQR_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
-#ifdef EIGEN2_SUPPORT
-#include "Eigenvalues"
-#endif
-
 #endif // EIGEN_QR_MODULE_H
 /* vim: set filetype=cpp et sw=2 ts=2 ai: */
diff --git a/Eigen/QtAlignedMalloc b/Eigen/QtAlignedMalloc
index 46f7d83..4f07df0 100644
--- a/Eigen/QtAlignedMalloc
+++ b/Eigen/QtAlignedMalloc
@@ -1,3 +1,9 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
 #ifndef EIGEN_QTMALLOC_MODULE_H
 #define EIGEN_QTMALLOC_MODULE_H
@@ -8,7 +14,7 @@
 
 #include "src/Core/util/DisableStupidWarnings.h"
 
-void *qMalloc(size_t size)
+void *qMalloc(std::size_t size)
 {
   return Eigen::internal::aligned_malloc(size);
 }
@@ -18,10 +24,10 @@
   Eigen::internal::aligned_free(ptr);
 }
 
-void *qRealloc(void *ptr, size_t size)
+void *qRealloc(void *ptr, std::size_t size)
 {
   void* newPtr = Eigen::internal::aligned_malloc(size);
-  memcpy(newPtr, ptr, size);
+  std::memcpy(newPtr, ptr, size);
   Eigen::internal::aligned_free(ptr);
   return newPtr;
 }
diff --git a/Eigen/SPQRSupport b/Eigen/SPQRSupport
index 7701644..f70390c 100644
--- a/Eigen/SPQRSupport
+++ b/Eigen/SPQRSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPQRSUPPORT_MODULE_H
 #define EIGEN_SPQRSUPPORT_MODULE_H
 
@@ -10,7 +17,7 @@
 /** \ingroup Support_modules
   * \defgroup SPQRSupport_Module SuiteSparseQR module
   * 
-  * This module provides an interface to the SPQR library, which is part of the <a href="http://www.cise.ufl.edu/research/sparse/SuiteSparse/">suitesparse</a> package.
+  * This module provides an interface to the SPQR library, which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
   *
   * \code
   * #include <Eigen/SPQRSupport>
@@ -21,8 +28,6 @@
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
 #include "src/CholmodSupport/CholmodSupport.h"
 #include "src/SPQRSupport/SuiteSparseQRSupport.h"
 
diff --git a/Eigen/SVD b/Eigen/SVD
index fd31001..5d0e75f 100644
--- a/Eigen/SVD
+++ b/Eigen/SVD
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SVD_MODULE_H
 #define EIGEN_SVD_MODULE_H
 
@@ -12,23 +19,30 @@
   *
   *
   * This module provides SVD decomposition for matrices (both real and complex).
-  * This decomposition is accessible via the following MatrixBase method:
+  * Two decomposition algorithms are provided:
+  *  - JacobiSVD implementing two-sided Jacobi iterations is numerically very accurate, fast for small matrices, but very slow for larger ones.
+  *  - BDCSVD implementing a recursive divide & conquer strategy on top of an upper-bidiagonalization which remains fast for large problems.
+  * These decompositions are accessible via the respective classes and following MatrixBase methods:
   *  - MatrixBase::jacobiSvd()
+  *  - MatrixBase::bdcSvd()
   *
   * \code
   * #include <Eigen/SVD>
   * \endcode
   */
 
-#include "src/misc/Solve.h"
-#include "src/SVD/JacobiSVD.h"
-#if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT)
-#include "src/SVD/JacobiSVD_MKL.h"
-#endif
+#include "src/misc/RealSvd2x2.h"
 #include "src/SVD/UpperBidiagonalization.h"
-
-#ifdef EIGEN2_SUPPORT
-#include "src/Eigen2Support/SVD.h"
+#include "src/SVD/SVDBase.h"
+#include "src/SVD/JacobiSVD.h"
+#include "src/SVD/BDCSVD.h"
+#if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT)
+#ifdef EIGEN_USE_MKL
+#include "mkl_lapacke.h"
+#else
+#include "src/misc/lapacke.h"
+#endif
+#include "src/SVD/JacobiSVD_LAPACKE.h"
 #endif
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/Eigen/Sparse b/Eigen/Sparse
index 7cc9c09..136e681 100644
--- a/Eigen/Sparse
+++ b/Eigen/Sparse
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPARSE_MODULE_H
 #define EIGEN_SPARSE_MODULE_H
 
@@ -11,14 +18,16 @@
   * - \ref SparseQR_Module
   * - \ref IterativeLinearSolvers_Module
   *
-  * \code
-  * #include <Eigen/Sparse>
-  * \endcode
+    \code
+    #include <Eigen/Sparse>
+    \endcode
   */
 
 #include "SparseCore"
 #include "OrderingMethods"
+#ifndef EIGEN_MPL2_ONLY
 #include "SparseCholesky"
+#endif
 #include "SparseLU"
 #include "SparseQR"
 #include "IterativeLinearSolvers"
diff --git a/Eigen/SparseCholesky b/Eigen/SparseCholesky
index 9f5056a..b6a320c 100644
--- a/Eigen/SparseCholesky
+++ b/Eigen/SparseCholesky
@@ -34,8 +34,6 @@
 #error The SparseCholesky module has nothing to offer in MPL2 only mode
 #endif
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
 #include "src/SparseCholesky/SimplicialCholesky.h"
 
 #ifndef EIGEN_MPL2_ONLY
diff --git a/Eigen/SparseCore b/Eigen/SparseCore
index 9b5be5e..76966c4 100644
--- a/Eigen/SparseCore
+++ b/Eigen/SparseCore
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPARSECORE_MODULE_H
 #define EIGEN_SPARSECORE_MODULE_H
 
@@ -14,7 +21,7 @@
 /** 
   * \defgroup SparseCore_Module SparseCore module
   *
-  * This module provides a sparse matrix representation, and basic associatd matrix manipulations
+  * This module provides a sparse matrix representation, and basic associated matrix manipulations
   * and operations.
   *
   * See the \ref TutorialSparse "Sparse tutorial"
@@ -26,37 +33,35 @@
   * This module depends on: Core.
   */
 
-namespace Eigen {
-
-/** The type used to identify a general sparse storage. */
-struct Sparse {};
-
-}
-
 #include "src/SparseCore/SparseUtil.h"
 #include "src/SparseCore/SparseMatrixBase.h"
+#include "src/SparseCore/SparseAssign.h"
 #include "src/SparseCore/CompressedStorage.h"
 #include "src/SparseCore/AmbiVector.h"
+#include "src/SparseCore/SparseCompressedBase.h"
 #include "src/SparseCore/SparseMatrix.h"
+#include "src/SparseCore/SparseMap.h"
 #include "src/SparseCore/MappedSparseMatrix.h"
 #include "src/SparseCore/SparseVector.h"
-#include "src/SparseCore/SparseBlock.h"
-#include "src/SparseCore/SparseTranspose.h"
+#include "src/SparseCore/SparseRef.h"
 #include "src/SparseCore/SparseCwiseUnaryOp.h"
 #include "src/SparseCore/SparseCwiseBinaryOp.h"
+#include "src/SparseCore/SparseTranspose.h"
+#include "src/SparseCore/SparseBlock.h"
 #include "src/SparseCore/SparseDot.h"
-#include "src/SparseCore/SparsePermutation.h"
 #include "src/SparseCore/SparseRedux.h"
-#include "src/SparseCore/SparseFuzzy.h"
+#include "src/SparseCore/SparseView.h"
+#include "src/SparseCore/SparseDiagonalProduct.h"
 #include "src/SparseCore/ConservativeSparseSparseProduct.h"
 #include "src/SparseCore/SparseSparseProductWithPruning.h"
 #include "src/SparseCore/SparseProduct.h"
 #include "src/SparseCore/SparseDenseProduct.h"
-#include "src/SparseCore/SparseDiagonalProduct.h"
-#include "src/SparseCore/SparseTriangularView.h"
 #include "src/SparseCore/SparseSelfAdjointView.h"
+#include "src/SparseCore/SparseTriangularView.h"
 #include "src/SparseCore/TriangularSolver.h"
-#include "src/SparseCore/SparseView.h"
+#include "src/SparseCore/SparsePermutation.h"
+#include "src/SparseCore/SparseFuzzy.h"
+#include "src/SparseCore/SparseSolverBase.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
 
diff --git a/Eigen/SparseLU b/Eigen/SparseLU
index 8527a49..38b38b5 100644
--- a/Eigen/SparseLU
+++ b/Eigen/SparseLU
@@ -20,9 +20,6 @@
   * Please, see the documentation of the SparseLU class for more details.
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 // Ordering interface
 #include "OrderingMethods"
 
diff --git a/Eigen/SparseQR b/Eigen/SparseQR
index 4ee4206..a6f3b7f 100644
--- a/Eigen/SparseQR
+++ b/Eigen/SparseQR
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SPARSEQR_MODULE_H
 #define EIGEN_SPARSEQR_MODULE_H
 
@@ -21,9 +28,6 @@
   * 
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "OrderingMethods"
 #include "src/SparseCore/SparseColEtree.h"
 #include "src/SparseQR/SparseQR.h"
diff --git a/Eigen/StdDeque b/Eigen/StdDeque
index f272347..bc68397 100644
--- a/Eigen/StdDeque
+++ b/Eigen/StdDeque
@@ -14,7 +14,7 @@
 #include "Core"
 #include <deque>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...)
 
diff --git a/Eigen/StdList b/Eigen/StdList
index 225c1e1..4c6262c 100644
--- a/Eigen/StdList
+++ b/Eigen/StdList
@@ -13,7 +13,7 @@
 #include "Core"
 #include <list>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */    
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...)
 
diff --git a/Eigen/StdVector b/Eigen/StdVector
index 6b22627..0c4697a 100644
--- a/Eigen/StdVector
+++ b/Eigen/StdVector
@@ -14,7 +14,7 @@
 #include "Core"
 #include <vector>
 
-#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */
+#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */
 
 #define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...)
 
diff --git a/Eigen/SuperLUSupport b/Eigen/SuperLUSupport
index 575e14f..59312a8 100644
--- a/Eigen/SuperLUSupport
+++ b/Eigen/SuperLUSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_SUPERLUSUPPORT_MODULE_H
 #define EIGEN_SUPERLUSUPPORT_MODULE_H
 
@@ -36,6 +43,8 @@
   * - class SuperLU: a supernodal sequential LU factorization.
   * - class SuperILU: a supernodal sequential incomplete LU factorization (to be used as a preconditioner for iterative methods).
   *
+  * \warning This wrapper requires at least versions 4.0 of SuperLU. The 3.x versions are not supported.
+  *
   * \warning When including this module, you have to use SUPERLU_EMPTY instead of EMPTY which is no longer defined because it is too polluting.
   *
   * \code
@@ -48,12 +57,8 @@
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/SuperLUSupport/SuperLUSupport.h"
 
-
 #include "src/Core/util/ReenableStupidWarnings.h"
 
 #endif // EIGEN_SUPERLUSUPPORT_MODULE_H
diff --git a/Eigen/UmfPackSupport b/Eigen/UmfPackSupport
index 984f64a..00eec80 100644
--- a/Eigen/UmfPackSupport
+++ b/Eigen/UmfPackSupport
@@ -1,3 +1,10 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
 #ifndef EIGEN_UMFPACKSUPPORT_MODULE_H
 #define EIGEN_UMFPACKSUPPORT_MODULE_H
 
@@ -12,7 +19,7 @@
 /** \ingroup Support_modules
   * \defgroup UmfPackSupport_Module UmfPackSupport module
   *
-  * This module provides an interface to the UmfPack library which is part of the <a href="http://www.cise.ufl.edu/research/sparse/SuiteSparse/">suitesparse</a> package.
+  * This module provides an interface to the UmfPack library which is part of the <a href="http://www.suitesparse.com">suitesparse</a> package.
   * It provides the following factorization class:
   * - class UmfPackLU: a multifrontal sequential LU factorization.
   *
@@ -26,9 +33,6 @@
   *
   */
 
-#include "src/misc/Solve.h"
-#include "src/misc/SparseSolve.h"
-
 #include "src/UmfPackSupport/UmfPackSupport.h"
 
 #include "src/Core/util/ReenableStupidWarnings.h"
diff --git a/Eigen/src/CMakeLists.txt b/Eigen/src/CMakeLists.txt
deleted file mode 100644
index c326f37..0000000
--- a/Eigen/src/CMakeLists.txt
+++ /dev/null
@@ -1,7 +0,0 @@
-file(GLOB Eigen_src_subdirectories "*")
-escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}")
-foreach(f ${Eigen_src_subdirectories})
-  if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" )
-    add_subdirectory(${f})
-  endif()
-endforeach()
diff --git a/Eigen/src/Cholesky/CMakeLists.txt b/Eigen/src/Cholesky/CMakeLists.txt
deleted file mode 100644
index d01488b..0000000
--- a/Eigen/src/Cholesky/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Cholesky_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Cholesky_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Cholesky COMPONENT Devel
-  )
diff --git a/Eigen/src/Cholesky/LDLT.h b/Eigen/src/Cholesky/LDLT.h
index e01ae82..15ccf24 100644
--- a/Eigen/src/Cholesky/LDLT.h
+++ b/Eigen/src/Cholesky/LDLT.h
@@ -13,7 +13,7 @@
 #ifndef EIGEN_LDLT_H
 #define EIGEN_LDLT_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
   template<typename MatrixType, int UpLo> struct LDLT_Traits;
@@ -28,8 +28,8 @@
   *
   * \brief Robust Cholesky decomposition of a matrix with pivoting
   *
-  * \param MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
-  * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  * \tparam _MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition
+  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
   *             The other triangular part won't be read.
   *
   * Perform a robust Cholesky decomposition of a positive semidefinite or negative semidefinite
@@ -43,7 +43,9 @@
   * Remember that Cholesky decompositions are not rank-revealing. Also, do not use a Cholesky
   * decomposition to determine whether a system of equations has a solution.
   *
-  * \sa MatrixBase::ldlt(), class LLT
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
+  * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT
   */
 template<typename _MatrixType, int _UpLo> class LDLT
 {
@@ -52,15 +54,15 @@
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options & ~RowMajorBit, // these are the options for the TmpMatrixType, we need a ColMajor matrix here!
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
       UpLo = _UpLo
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef Matrix<Scalar, RowsAtCompileTime, 1, Options, MaxRowsAtCompileTime, 1> TmpMatrixType;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef Matrix<Scalar, RowsAtCompileTime, 1, 0, MaxRowsAtCompileTime, 1> TmpMatrixType;
 
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
@@ -72,11 +74,11 @@
       * The default constructor is useful in cases in which the user intends to
       * perform decompositions via LDLT::compute(const MatrixType&).
       */
-    LDLT() 
-      : m_matrix(), 
-        m_transpositions(), 
+    LDLT()
+      : m_matrix(),
+        m_transpositions(),
         m_sign(internal::ZeroSign),
-        m_isInitialized(false) 
+        m_isInitialized(false)
     {}
 
     /** \brief Default Constructor with memory preallocation
@@ -85,7 +87,7 @@
       * according to the specified problem \a size.
       * \sa LDLT()
       */
-    LDLT(Index size)
+    explicit LDLT(Index size)
       : m_matrix(size, size),
         m_transpositions(size),
         m_temporary(size),
@@ -96,16 +98,35 @@
     /** \brief Constructor with decomposition
       *
       * This calculates the decomposition for the input \a matrix.
+      *
       * \sa LDLT(Index size)
       */
-    LDLT(const MatrixType& matrix)
+    template<typename InputType>
+    explicit LDLT(const EigenBase<InputType>& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_transpositions(matrix.rows()),
         m_temporary(matrix.rows()),
         m_sign(internal::ZeroSign),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a LDLT factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa LDLT(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit LDLT(EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
+        m_transpositions(matrix.rows()),
+        m_temporary(matrix.rows()),
+        m_sign(internal::ZeroSign),
+        m_isInitialized(false)
+    {
+      compute(matrix.derived());
     }
 
     /** Clear any existing decomposition
@@ -151,13 +172,6 @@
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
       return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign;
     }
-    
-    #ifdef EIGEN2_SUPPORT
-    inline bool isPositiveDefinite() const
-    {
-      return isPositive();
-    }
-    #endif
 
     /** \returns true if the matrix is negative (semidefinite) */
     inline bool isNegative(void) const
@@ -173,37 +187,38 @@
       * \note_about_checking_solutions
       *
       * More precisely, this method solves \f$ A x = b \f$ using the decomposition \f$ A = P^T L D L^* P \f$
-      * by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$, 
+      * by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$,
       * \f$ L^* y_4 = y_3 \f$ and \f$ P x = y_4 \f$ in succession. If the matrix \f$ A \f$ is singular, then
       * \f$ D \f$ will also be singular (all the other matrices are invertible). In that case, the
       * least-square solution of \f$ D y_3 = y_2 \f$ is computed. This does not mean that this function
       * computes the least-square solution of \f$ A x = b \f$ is \f$ A \f$ is singular.
       *
-      * \sa MatrixBase::ldlt()
+      * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<LDLT, Rhs>
+    inline const Solve<LDLT, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
       eigen_assert(m_matrix.rows()==b.rows()
                 && "LDLT::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<LDLT, Rhs>(*this, b.derived());
+      return Solve<LDLT, Rhs>(*this, b.derived());
     }
 
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived, typename ResultType>
-    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
-    {
-      *result = this->solve(b);
-      return true;
-    }
-    #endif
-
     template<typename Derived>
     bool solveInPlace(MatrixBase<Derived> &bAndX) const;
 
-    LDLT& compute(const MatrixType& matrix);
+    template<typename InputType>
+    LDLT& compute(const EigenBase<InputType>& matrix);
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of
+     *  which \c *this is the LDLT decomposition.
+     */
+    RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "LDLT is not initialized.");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
+    }
 
     template <typename Derived>
     LDLT& rankUpdate(const MatrixBase<Derived>& w, const RealScalar& alpha=1);
@@ -220,22 +235,35 @@
 
     MatrixType reconstructedMatrix() const;
 
+    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
+      *
+      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
+      * \code x = decomposition.adjoint().solve(b) \endcode
+      */
+    const LDLT& adjoint() const { return *this; };
+
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
 
     /** \brief Reports whether previous computation was successful.
       *
       * \returns \c Success if computation was succesful,
-      *          \c NumericalIssue if the matrix.appears to be negative.
+      *          \c NumericalIssue if the factorization failed because of a zero pivot.
       */
     ComputationInfo info() const
     {
       eigen_assert(m_isInitialized && "LDLT is not initialized.");
-      return Success;
+      return m_info;
     }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
+
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
@@ -248,10 +276,12 @@
       * is not stored), and the diagonal entries correspond to D.
       */
     MatrixType m_matrix;
+    RealScalar m_l1_norm;
     TranspositionType m_transpositions;
     TmpMatrixType m_temporary;
     internal::SignMatrix m_sign;
     bool m_isInitialized;
+    ComputationInfo m_info;
 };
 
 namespace internal {
@@ -266,15 +296,18 @@
     using std::abs;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename TranspositionType::StorageIndex IndexType;
     eigen_assert(mat.rows()==mat.cols());
     const Index size = mat.rows();
+    bool found_zero_pivot = false;
+    bool ret = true;
 
     if (size <= 1)
     {
       transpositions.setIdentity();
-      if (numext::real(mat.coeff(0,0)) > 0) sign = PositiveSemiDef;
-      else if (numext::real(mat.coeff(0,0)) < 0) sign = NegativeSemiDef;
+      if(size==0) sign = ZeroSign;
+      else if (numext::real(mat.coeff(0,0)) > static_cast<RealScalar>(0) ) sign = PositiveSemiDef;
+      else if (numext::real(mat.coeff(0,0)) < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
       else sign = ZeroSign;
       return true;
     }
@@ -286,7 +319,7 @@
       mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner);
       index_of_biggest_in_corner += k;
 
-      transpositions.coeffRef(k) = index_of_biggest_in_corner;
+      transpositions.coeffRef(k) = IndexType(index_of_biggest_in_corner);
       if(k != index_of_biggest_in_corner)
       {
         // apply the transposition while taking care to consider only
@@ -295,7 +328,7 @@
         mat.row(k).head(k).swap(mat.row(index_of_biggest_in_corner).head(k));
         mat.col(k).tail(s).swap(mat.col(index_of_biggest_in_corner).tail(s));
         std::swap(mat.coeffRef(k,k),mat.coeffRef(index_of_biggest_in_corner,index_of_biggest_in_corner));
-        for(int i=k+1;i<index_of_biggest_in_corner;++i)
+        for(Index i=k+1;i<index_of_biggest_in_corner;++i)
         {
           Scalar tmp = mat.coeffRef(i,k);
           mat.coeffRef(i,k) = numext::conj(mat.coeffRef(index_of_biggest_in_corner,i));
@@ -321,26 +354,46 @@
         if(rs>0)
           A21.noalias() -= A20 * temp.head(k);
       }
-      
+
       // In some previous versions of Eigen (e.g., 3.2.1), the scaling was omitted if the pivot
-      // was smaller than the cutoff value. However, soince LDLT is not rank-revealing
-      // we should only make sure we do not introduce INF or NaN values.
-      // LAPACK also uses 0 as the cutoff value.
+      // was smaller than the cutoff value. However, since LDLT is not rank-revealing
+      // we should only make sure that we do not introduce INF or NaN values.
+      // Remark that LAPACK also uses 0 as the cutoff value.
       RealScalar realAkk = numext::real(mat.coeffRef(k,k));
-      if((rs>0) && (abs(realAkk) > RealScalar(0)))
+      bool pivot_is_valid = (abs(realAkk) > RealScalar(0));
+
+      if(k==0 && !pivot_is_valid)
+      {
+        // The entire diagonal is zero, there is nothing more to do
+        // except filling the transpositions, and checking whether the matrix is zero.
+        sign = ZeroSign;
+        for(Index j = 0; j<size; ++j)
+        {
+          transpositions.coeffRef(j) = IndexType(j);
+          ret = ret && (mat.col(j).tail(size-j-1).array()==Scalar(0)).all();
+        }
+        return ret;
+      }
+
+      if((rs>0) && pivot_is_valid)
         A21 /= realAkk;
+      else if(rs>0)
+        ret = ret && (A21.array()==Scalar(0)).all();
+
+      if(found_zero_pivot && pivot_is_valid) ret = false; // factorization failed
+      else if(!pivot_is_valid) found_zero_pivot = true;
 
       if (sign == PositiveSemiDef) {
-        if (realAkk < 0) sign = Indefinite;
+        if (realAkk < static_cast<RealScalar>(0)) sign = Indefinite;
       } else if (sign == NegativeSemiDef) {
-        if (realAkk > 0) sign = Indefinite;
+        if (realAkk > static_cast<RealScalar>(0)) sign = Indefinite;
       } else if (sign == ZeroSign) {
-        if (realAkk > 0) sign = PositiveSemiDef;
-        else if (realAkk < 0) sign = NegativeSemiDef;
+        if (realAkk > static_cast<RealScalar>(0)) sign = PositiveSemiDef;
+        else if (realAkk < static_cast<RealScalar>(0)) sign = NegativeSemiDef;
       }
     }
 
-    return true;
+    return ret;
   }
 
   // Reference for the algorithm: Davis and Hager, "Multiple Rank
@@ -356,7 +409,6 @@
     using numext::isfinite;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
 
     const Index size = mat.rows();
     eigen_assert(mat.cols() == size && w.size()==size);
@@ -420,16 +472,16 @@
 {
   typedef const TriangularView<const MatrixType, UnitLower> MatrixL;
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
 };
 
 template<typename MatrixType> struct LDLT_Traits<MatrixType,Upper>
 {
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, UnitLower> MatrixL;
   typedef const TriangularView<const MatrixType, UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); }
-  static inline MatrixU getU(const MatrixType& m) { return m; }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
 };
 
 } // end namespace internal
@@ -437,21 +489,35 @@
 /** Compute / recompute the LDLT decomposition A = L D L^* = U^* D U of \a matrix
   */
 template<typename MatrixType, int _UpLo>
-LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const MatrixType& a)
+template<typename InputType>
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
 {
   check_template_parameters();
-  
+
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
 
-  m_matrix = a;
+  m_matrix = a.derived();
+
+  // Compute matrix L1 norm = max abs column sum.
+  m_l1_norm = RealScalar(0);
+  // TODO move this code to SelfAdjointView
+  for (Index col = 0; col < size; ++col) {
+    RealScalar abs_col_sum;
+    if (_UpLo == Lower)
+      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
+    else
+      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
+    if (abs_col_sum > m_l1_norm)
+      m_l1_norm = abs_col_sum;
+  }
 
   m_transpositions.resize(size);
   m_isInitialized = false;
   m_temporary.resize(size);
   m_sign = internal::ZeroSign;
 
-  internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign);
+  m_info = internal::ldlt_inplace<UpLo>::unblocked(m_matrix, m_transpositions, m_temporary, m_sign) ? Success : NumericalIssue;
 
   m_isInitialized = true;
   return *this;
@@ -464,20 +530,21 @@
   */
 template<typename MatrixType, int _UpLo>
 template<typename Derived>
-LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Derived>& w, const typename NumTraits<typename MatrixType::Scalar>::Real& sigma)
+LDLT<MatrixType,_UpLo>& LDLT<MatrixType,_UpLo>::rankUpdate(const MatrixBase<Derived>& w, const typename LDLT<MatrixType,_UpLo>::RealScalar& sigma)
 {
+  typedef typename TranspositionType::StorageIndex IndexType;
   const Index size = w.rows();
   if (m_isInitialized)
   {
     eigen_assert(m_matrix.rows()==size);
   }
   else
-  {    
+  {
     m_matrix.resize(size,size);
     m_matrix.setZero();
     m_transpositions.resize(size);
     for (Index i = 0; i < size; i++)
-      m_transpositions.coeffRef(i) = i;
+      m_transpositions.coeffRef(i) = IndexType(i);
     m_temporary.resize(size);
     m_sign = sigma>=0 ? internal::PositiveSemiDef : internal::NegativeSemiDef;
     m_isInitialized = true;
@@ -488,53 +555,46 @@
   return *this;
 }
 
-namespace internal {
-template<typename _MatrixType, int _UpLo, typename Rhs>
-struct solve_retval<LDLT<_MatrixType,_UpLo>, Rhs>
-  : solve_retval_base<LDLT<_MatrixType,_UpLo>, Rhs>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType, int _UpLo>
+template<typename RhsType, typename DstType>
+void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  typedef LDLT<_MatrixType,_UpLo> LDLTType;
-  EIGEN_MAKE_SOLVE_HELPERS(LDLTType,Rhs)
+  eigen_assert(rhs.rows() == rows());
+  // dst = P b
+  dst = m_transpositions * rhs;
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  // dst = L^-1 (P b)
+  matrixL().solveInPlace(dst);
+
+  // dst = D^-1 (L^-1 P b)
+  // more precisely, use pseudo-inverse of D (see bug 241)
+  using std::abs;
+  const typename Diagonal<const MatrixType>::RealReturnType vecD(vectorD());
+  // In some previous versions, tolerance was set to the max of 1/highest (or rather numeric_limits::min())
+  // and the maximal diagonal entry * epsilon as motivated by LAPACK's xGELSS:
+  // RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
+  // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
+  // diagonal element is not well justified and leads to numerical issues in some cases.
+  // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
+  // Using numeric_limits::min() gives us more robustness to denormals.
+  RealScalar tolerance = (std::numeric_limits<RealScalar>::min)();
+
+  for (Index i = 0; i < vecD.size(); ++i)
   {
-    eigen_assert(rhs().rows() == dec().matrixLDLT().rows());
-    // dst = P b
-    dst = dec().transpositionsP() * rhs();
-
-    // dst = L^-1 (P b)
-    dec().matrixL().solveInPlace(dst);
-
-    // dst = D^-1 (L^-1 P b)
-    // more precisely, use pseudo-inverse of D (see bug 241)
-    using std::abs;
-    using std::max;
-    typedef typename LDLTType::MatrixType MatrixType;
-    typedef typename LDLTType::RealScalar RealScalar;
-    const typename Diagonal<const MatrixType>::RealReturnType vectorD(dec().vectorD());
-    // In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon
-    // as motivated by LAPACK's xGELSS:
-    // RealScalar tolerance = (max)(vectorD.array().abs().maxCoeff() *NumTraits<RealScalar>::epsilon(),RealScalar(1) / NumTraits<RealScalar>::highest());
-    // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest
-    // diagonal element is not well justified and to numerical issues in some cases.
-    // Moreover, Lapack's xSYTRS routines use 0 for the tolerance.
-    RealScalar tolerance = RealScalar(1) / NumTraits<RealScalar>::highest();
-    
-    for (Index i = 0; i < vectorD.size(); ++i) {
-      if(abs(vectorD(i)) > tolerance)
-        dst.row(i) /= vectorD(i);
-      else
-        dst.row(i).setZero();
-    }
-
-    // dst = L^-T (D^-1 L^-1 P b)
-    dec().matrixU().solveInPlace(dst);
-
-    // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
-    dst = dec().transpositionsP().transpose() * dst;
+    if(abs(vecD(i)) > tolerance)
+      dst.row(i) /= vecD(i);
+    else
+      dst.row(i).setZero();
   }
-};
+
+  // dst = L^-T (D^-1 L^-1 P b)
+  matrixU().solveInPlace(dst);
+
+  // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b
+  dst = m_transpositions.transpose() * dst;
 }
+#endif
 
 /** \internal use x = ldlt_object.solve(x);
   *
@@ -588,6 +648,7 @@
 
 /** \cholesky_module
   * \returns the Cholesky decomposition with full pivoting without square root of \c *this
+  * \sa MatrixBase::ldlt()
   */
 template<typename MatrixType, unsigned int UpLo>
 inline const LDLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
@@ -598,6 +659,7 @@
 
 /** \cholesky_module
   * \returns the Cholesky decomposition with full pivoting without square root of \c *this
+  * \sa SelfAdjointView::ldlt()
   */
 template<typename Derived>
 inline const LDLT<typename MatrixBase<Derived>::PlainObject>
diff --git a/Eigen/src/Cholesky/LLT.h b/Eigen/src/Cholesky/LLT.h
index 59723a6..e1624d2 100644
--- a/Eigen/src/Cholesky/LLT.h
+++ b/Eigen/src/Cholesky/LLT.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_LLT_H
 #define EIGEN_LLT_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal{
 template<typename MatrixType, int UpLo> struct LLT_Traits;
@@ -22,9 +22,9 @@
   *
   * \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features
   *
-  * \param MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
-  * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
-  *             The other triangular part won't be read.
+  * \tparam _MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition
+  * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper.
+  *               The other triangular part won't be read.
   *
   * This class performs a LL^T Cholesky decomposition of a symmetric, positive definite
   * matrix A such that A = LL^* = U^*U, where L is lower triangular.
@@ -40,12 +40,18 @@
   *
   * Example: \include LLT_example.cpp
   * Output: \verbinclude LLT_example.out
-  *    
-  * \sa MatrixBase::llt(), class LDLT
-  */
- /* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH)
-  * Note that during the decomposition, only the upper triangular part of A is considered. Therefore,
-  * the strict lower part does not have to store correct values.
+  *
+  * \b Performance: for best performance, it is recommended to use a column-major storage format
+  * with the Lower triangular part (the default), or, equivalently, a row-major storage format
+  * with the Upper triangular part. Otherwise, you might get a 20% slowdown for the full factorization
+  * step, and rank-updates can be up to 3 times slower.
+  *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  *
+  * Note that during the decomposition, only the lower (or upper, as defined by _UpLo) triangular part of A is considered.
+  * Therefore, the strict lower part does not have to store correct values.
+  *
+  * \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT
   */
 template<typename _MatrixType, int _UpLo> class LLT
 {
@@ -54,12 +60,12 @@
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename MatrixType::StorageIndex StorageIndex;
 
     enum {
       PacketSize = internal::packet_traits<Scalar>::size,
@@ -83,14 +89,30 @@
       * according to the specified problem \a size.
       * \sa LLT()
       */
-    LLT(Index size) : m_matrix(size, size),
+    explicit LLT(Index size) : m_matrix(size, size),
                     m_isInitialized(false) {}
 
-    LLT(const MatrixType& matrix)
+    template<typename InputType>
+    explicit LLT(const EigenBase<InputType>& matrix)
       : m_matrix(matrix.rows(), matrix.cols()),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a LDLT factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
+      * \c MatrixType is a Eigen::Ref.
+      *
+      * \sa LLT(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit LLT(EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
+        m_isInitialized(false)
+    {
+      compute(matrix.derived());
     }
 
     /** \returns a view of the upper triangular matrix U */
@@ -115,33 +137,33 @@
       * Example: \include LLT_solve.cpp
       * Output: \verbinclude LLT_solve.out
       *
-      * \sa solveInPlace(), MatrixBase::llt()
+      * \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<LLT, Rhs>
+    inline const Solve<LLT, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "LLT is not initialized.");
       eigen_assert(m_matrix.rows()==b.rows()
                 && "LLT::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<LLT, Rhs>(*this, b.derived());
+      return Solve<LLT, Rhs>(*this, b.derived());
     }
 
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived, typename ResultType>
-    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
-    {
-      *result = this->solve(b);
-      return true;
-    }
-    
-    bool isPositiveDefinite() const { return true; }
-    #endif
-
     template<typename Derived>
-    void solveInPlace(MatrixBase<Derived> &bAndX) const;
+    void solveInPlace(const MatrixBase<Derived> &bAndX) const;
 
-    LLT& compute(const MatrixType& matrix);
+    template<typename InputType>
+    LLT& compute(const EigenBase<InputType>& matrix);
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of
+      *  which \c *this is the Cholesky decomposition.
+      */
+    RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "LLT is not initialized.");
+      eigen_assert(m_info == Success && "LLT failed because matrix appears to be negative");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
+    }
 
     /** \returns the LLT decomposition matrix
       *
@@ -159,7 +181,7 @@
     /** \brief Reports whether previous computation was successful.
       *
       * \returns \c Success if computation was succesful,
-      *          \c NumericalIssue if the matrix.appears to be negative.
+      *          \c NumericalIssue if the matrix.appears not to be positive definite.
       */
     ComputationInfo info() const
     {
@@ -167,24 +189,38 @@
       return m_info;
     }
 
+    /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint.
+      *
+      * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as:
+      * \code x = decomposition.adjoint().solve(b) \endcode
+      */
+    const LLT& adjoint() const { return *this; };
+
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
 
     template<typename VectorType>
     LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1);
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
+
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
     /** \internal
       * Used to compute and store L
       * The strict upper part is not used and even not initialized.
       */
     MatrixType m_matrix;
+    RealScalar m_l1_norm;
     bool m_isInitialized;
     ComputationInfo m_info;
 };
@@ -194,12 +230,11 @@
 template<typename Scalar, int UpLo> struct llt_inplace;
 
 template<typename MatrixType, typename VectorType>
-static typename MatrixType::Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
+static Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma)
 {
   using std::sqrt;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename MatrixType::RealScalar RealScalar;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::ColXpr ColXpr;
   typedef typename internal::remove_all<ColXpr>::type ColXprCleaned;
   typedef typename ColXprCleaned::SegmentReturnType ColXprSegment;
@@ -268,11 +303,10 @@
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
   template<typename MatrixType>
-  static typename MatrixType::Index unblocked(MatrixType& mat)
+  static Index unblocked(MatrixType& mat)
   {
     using std::sqrt;
-    typedef typename MatrixType::Index Index;
-    
+
     eigen_assert(mat.rows()==mat.cols());
     const Index size = mat.rows();
     for(Index k = 0; k < size; ++k)
@@ -289,15 +323,14 @@
         return k;
       mat.coeffRef(k,k) = x = sqrt(x);
       if (k>0 && rs>0) A21.noalias() -= A20 * A10.adjoint();
-      if (rs>0) A21 *= RealScalar(1)/x;
+      if (rs>0) A21 /= x;
     }
     return -1;
   }
 
   template<typename MatrixType>
-  static typename MatrixType::Index blocked(MatrixType& m)
+  static Index blocked(MatrixType& m)
   {
-    typedef typename MatrixType::Index Index;
     eigen_assert(m.rows()==m.cols());
     Index size = m.rows();
     if(size<32)
@@ -322,36 +355,36 @@
       Index ret;
       if((ret=unblocked(A11))>=0) return k+ret;
       if(rs>0) A11.adjoint().template triangularView<Upper>().template solveInPlace<OnTheRight>(A21);
-      if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,-1); // bottleneck
+      if(rs>0) A22.template selfadjointView<Lower>().rankUpdate(A21,typename NumTraits<RealScalar>::Literal(-1)); // bottleneck
     }
     return -1;
   }
 
   template<typename MatrixType, typename VectorType>
-  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
   {
     return Eigen::internal::llt_rank_update_lower(mat, vec, sigma);
   }
 };
-  
+
 template<typename Scalar> struct llt_inplace<Scalar, Upper>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
 
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE typename MatrixType::Index unblocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE Index unblocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::unblocked(matt);
   }
   template<typename MatrixType>
-  static EIGEN_STRONG_INLINE typename MatrixType::Index blocked(MatrixType& mat)
+  static EIGEN_STRONG_INLINE Index blocked(MatrixType& mat)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::blocked(matt);
   }
   template<typename MatrixType, typename VectorType>
-  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma)
   {
     Transpose<MatrixType> matt(mat);
     return llt_inplace<Scalar, Lower>::rankUpdate(matt, vec.conjugate(), sigma);
@@ -362,8 +395,8 @@
 {
   typedef const TriangularView<const MatrixType, Lower> MatrixL;
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, Upper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
   static bool inplace_decomposition(MatrixType& m)
   { return llt_inplace<typename MatrixType::Scalar, Lower>::blocked(m)==-1; }
 };
@@ -372,8 +405,8 @@
 {
   typedef const TriangularView<const typename MatrixType::AdjointReturnType, Lower> MatrixL;
   typedef const TriangularView<const MatrixType, Upper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); }
-  static inline MatrixU getU(const MatrixType& m) { return m; }
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); }
   static bool inplace_decomposition(MatrixType& m)
   { return llt_inplace<typename MatrixType::Scalar, Upper>::blocked(m)==-1; }
 };
@@ -388,14 +421,29 @@
   * Output: \verbinclude TutorialLinAlgComputeTwice.out
   */
 template<typename MatrixType, int _UpLo>
-LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const MatrixType& a)
+template<typename InputType>
+LLT<MatrixType,_UpLo>& LLT<MatrixType,_UpLo>::compute(const EigenBase<InputType>& a)
 {
   check_template_parameters();
-  
+
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
   m_matrix.resize(size, size);
-  m_matrix = a;
+  if (!internal::is_same_dense(m_matrix, a.derived()))
+    m_matrix = a.derived();
+
+  // Compute matrix L1 norm = max abs column sum.
+  m_l1_norm = RealScalar(0);
+  // TODO move this code to SelfAdjointView
+  for (Index col = 0; col < size; ++col) {
+    RealScalar abs_col_sum;
+    if (_UpLo == Lower)
+      abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>();
+    else
+      abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>();
+    if (abs_col_sum > m_l1_norm)
+      m_l1_norm = abs_col_sum;
+  }
 
   m_isInitialized = true;
   bool ok = Traits::inplace_decomposition(m_matrix);
@@ -423,39 +471,33 @@
 
   return *this;
 }
-    
-namespace internal {
-template<typename _MatrixType, int UpLo, typename Rhs>
-struct solve_retval<LLT<_MatrixType, UpLo>, Rhs>
-  : solve_retval_base<LLT<_MatrixType, UpLo>, Rhs>
-{
-  typedef LLT<_MatrixType,UpLo> LLTType;
-  EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs)
 
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dst = rhs();
-    dec().solveInPlace(dst);
-  }
-};
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType,int _UpLo>
+template<typename RhsType, typename DstType>
+void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  dst = rhs;
+  solveInPlace(dst);
 }
+#endif
 
 /** \internal use x = llt_object.solve(x);
-  * 
+  *
   * This is the \em in-place version of solve().
   *
   * \param bAndX represents both the right-hand side matrix b and result x.
   *
-  * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD.
+  * This version avoids a copy when the right hand side matrix b is not needed anymore.
   *
-  * This version avoids a copy when the right hand side matrix b is not
-  * needed anymore.
+  * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
+  * This function will const_cast it, so constness isn't honored here.
   *
   * \sa LLT::solve(), MatrixBase::llt()
   */
 template<typename MatrixType, int _UpLo>
 template<typename Derived>
-void LLT<MatrixType,_UpLo>::solveInPlace(MatrixBase<Derived> &bAndX) const
+void LLT<MatrixType,_UpLo>::solveInPlace(const MatrixBase<Derived> &bAndX) const
 {
   eigen_assert(m_isInitialized && "LLT is not initialized.");
   eigen_assert(m_matrix.rows()==bAndX.rows());
@@ -475,6 +517,7 @@
 
 /** \cholesky_module
   * \returns the LLT decomposition of \c *this
+  * \sa SelfAdjointView::llt()
   */
 template<typename Derived>
 inline const LLT<typename MatrixBase<Derived>::PlainObject>
@@ -485,6 +528,7 @@
 
 /** \cholesky_module
   * \returns the LLT decomposition of \c *this
+  * \sa SelfAdjointView::llt()
   */
 template<typename MatrixType, unsigned int UpLo>
 inline const LLT<typename SelfAdjointView<MatrixType, UpLo>::PlainObject, UpLo>
diff --git a/Eigen/src/Cholesky/LLT_MKL.h b/Eigen/src/Cholesky/LLT_LAPACKE.h
similarity index 70%
rename from Eigen/src/Cholesky/LLT_MKL.h
rename to Eigen/src/Cholesky/LLT_LAPACKE.h
index 66675d7..bc6489e 100644
--- a/Eigen/src/Cholesky/LLT_MKL.h
+++ b/Eigen/src/Cholesky/LLT_LAPACKE.h
@@ -25,41 +25,38 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to LAPACKe
  *     LLt decomposition based on LAPACKE_?potrf function.
  ********************************************************************************
 */
 
-#ifndef EIGEN_LLT_MKL_H
-#define EIGEN_LLT_MKL_H
-
-#include "Eigen/src/Core/util/MKL_support.h"
-#include <iostream>
+#ifndef EIGEN_LLT_LAPACKE_H
+#define EIGEN_LLT_LAPACKE_H
 
 namespace Eigen { 
 
 namespace internal {
 
-template<typename Scalar> struct mkl_llt;
+template<typename Scalar> struct lapacke_llt;
 
-#define EIGEN_MKL_LLT(EIGTYPE, MKLTYPE, MKLPREFIX) \
-template<> struct mkl_llt<EIGTYPE> \
+#define EIGEN_LAPACKE_LLT(EIGTYPE, BLASTYPE, LAPACKE_PREFIX) \
+template<> struct lapacke_llt<EIGTYPE> \
 { \
   template<typename MatrixType> \
-  static inline typename MatrixType::Index potrf(MatrixType& m, char uplo) \
+  static inline Index potrf(MatrixType& m, char uplo) \
   { \
     lapack_int matrix_order; \
     lapack_int size, lda, info, StorageOrder; \
     EIGTYPE* a; \
     eigen_assert(m.rows()==m.cols()); \
     /* Set up parameters for ?potrf */ \
-    size = m.rows(); \
+    size = convert_index<lapack_int>(m.rows()); \
     StorageOrder = MatrixType::Flags&RowMajorBit?RowMajor:ColMajor; \
     matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
     a = &(m.coeffRef(0,0)); \
-    lda = m.outerStride(); \
+    lda = convert_index<lapack_int>(m.outerStride()); \
 \
-    info = LAPACKE_##MKLPREFIX##potrf( matrix_order, uplo, size, (MKLTYPE*)a, lda ); \
+    info = LAPACKE_##LAPACKE_PREFIX##potrf( matrix_order, uplo, size, (BLASTYPE*)a, lda ); \
     info = (info==0) ? -1 : info>0 ? info-1 : size; \
     return info; \
   } \
@@ -67,36 +64,36 @@
 template<> struct llt_inplace<EIGTYPE, Lower> \
 { \
   template<typename MatrixType> \
-  static typename MatrixType::Index blocked(MatrixType& m) \
+  static Index blocked(MatrixType& m) \
   { \
-    return mkl_llt<EIGTYPE>::potrf(m, 'L'); \
+    return lapacke_llt<EIGTYPE>::potrf(m, 'L'); \
   } \
   template<typename MatrixType, typename VectorType> \
-  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
   { return Eigen::internal::llt_rank_update_lower(mat, vec, sigma); } \
 }; \
 template<> struct llt_inplace<EIGTYPE, Upper> \
 { \
   template<typename MatrixType> \
-  static typename MatrixType::Index blocked(MatrixType& m) \
+  static Index blocked(MatrixType& m) \
   { \
-    return mkl_llt<EIGTYPE>::potrf(m, 'U'); \
+    return lapacke_llt<EIGTYPE>::potrf(m, 'U'); \
   } \
   template<typename MatrixType, typename VectorType> \
-  static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
+  static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \
   { \
     Transpose<MatrixType> matt(mat); \
     return llt_inplace<EIGTYPE, Lower>::rankUpdate(matt, vec.conjugate(), sigma); \
   } \
 };
 
-EIGEN_MKL_LLT(double, double, d)
-EIGEN_MKL_LLT(float, float, s)
-EIGEN_MKL_LLT(dcomplex, MKL_Complex16, z)
-EIGEN_MKL_LLT(scomplex, MKL_Complex8, c)
+EIGEN_LAPACKE_LLT(double, double, d)
+EIGEN_LAPACKE_LLT(float, float, s)
+EIGEN_LAPACKE_LLT(dcomplex, lapack_complex_double, z)
+EIGEN_LAPACKE_LLT(scomplex, lapack_complex_float, c)
 
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_LLT_MKL_H
+#endif // EIGEN_LLT_LAPACKE_H
diff --git a/Eigen/src/CholmodSupport/CMakeLists.txt b/Eigen/src/CholmodSupport/CMakeLists.txt
deleted file mode 100644
index 814dfa6..0000000
--- a/Eigen/src/CholmodSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_CholmodSupport_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_CholmodSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/CholmodSupport COMPONENT Devel
-  )
diff --git a/Eigen/src/CholmodSupport/CholmodSupport.h b/Eigen/src/CholmodSupport/CholmodSupport.h
index c449960..5719720 100644
--- a/Eigen/src/CholmodSupport/CholmodSupport.h
+++ b/Eigen/src/CholmodSupport/CholmodSupport.h
@@ -14,46 +14,52 @@
 
 namespace internal {
 
-template<typename Scalar, typename CholmodType>
-void cholmod_configure_matrix(CholmodType& mat)
-{
-  if (internal::is_same<Scalar,float>::value)
-  {
-    mat.xtype = CHOLMOD_REAL;
-    mat.dtype = CHOLMOD_SINGLE;
-  }
-  else if (internal::is_same<Scalar,double>::value)
-  {
+template<typename Scalar> struct cholmod_configure_matrix;
+
+template<> struct cholmod_configure_matrix<double> {
+  template<typename CholmodType>
+  static void run(CholmodType& mat) {
     mat.xtype = CHOLMOD_REAL;
     mat.dtype = CHOLMOD_DOUBLE;
   }
-  else if (internal::is_same<Scalar,std::complex<float> >::value)
-  {
-    mat.xtype = CHOLMOD_COMPLEX;
-    mat.dtype = CHOLMOD_SINGLE;
-  }
-  else if (internal::is_same<Scalar,std::complex<double> >::value)
-  {
+};
+
+template<> struct cholmod_configure_matrix<std::complex<double> > {
+  template<typename CholmodType>
+  static void run(CholmodType& mat) {
     mat.xtype = CHOLMOD_COMPLEX;
     mat.dtype = CHOLMOD_DOUBLE;
   }
-  else
-  {
-    eigen_assert(false && "Scalar type not supported by CHOLMOD");
-  }
-}
+};
+
+// Other scalar types are not yet suppotred by Cholmod
+// template<> struct cholmod_configure_matrix<float> {
+//   template<typename CholmodType>
+//   static void run(CholmodType& mat) {
+//     mat.xtype = CHOLMOD_REAL;
+//     mat.dtype = CHOLMOD_SINGLE;
+//   }
+// };
+//
+// template<> struct cholmod_configure_matrix<std::complex<float> > {
+//   template<typename CholmodType>
+//   static void run(CholmodType& mat) {
+//     mat.xtype = CHOLMOD_COMPLEX;
+//     mat.dtype = CHOLMOD_SINGLE;
+//   }
+// };
 
 } // namespace internal
 
 /** Wraps the Eigen sparse matrix \a mat into a Cholmod sparse matrix object.
   * Note that the data are shared.
   */
-template<typename _Scalar, int _Options, typename _Index>
-cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+cholmod_sparse viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_StorageIndex> > mat)
 {
   cholmod_sparse res;
   res.nzmax   = mat.nonZeros();
-  res.nrow    = mat.rows();;
+  res.nrow    = mat.rows();
   res.ncol    = mat.cols();
   res.p       = mat.outerIndexPtr();
   res.i       = mat.innerIndexPtr();
@@ -74,11 +80,11 @@
   res.dtype   = 0;
   res.stype   = -1;
   
-  if (internal::is_same<_Index,int>::value)
+  if (internal::is_same<_StorageIndex,int>::value)
   {
     res.itype = CHOLMOD_INT;
   }
-  else if (internal::is_same<_Index,UF_long>::value)
+  else if (internal::is_same<_StorageIndex,long>::value)
   {
     res.itype = CHOLMOD_LONG;
   }
@@ -88,7 +94,7 @@
   }
 
   // setup res.xtype
-  internal::cholmod_configure_matrix<_Scalar>(res);
+  internal::cholmod_configure_matrix<_Scalar>::run(res);
   
   res.stype = 0;
   
@@ -98,16 +104,23 @@
 template<typename _Scalar, int _Options, typename _Index>
 const cholmod_sparse viewAsCholmod(const SparseMatrix<_Scalar,_Options,_Index>& mat)
 {
-  cholmod_sparse res = viewAsCholmod(mat.const_cast_derived());
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
+  return res;
+}
+
+template<typename _Scalar, int _Options, typename _Index>
+const cholmod_sparse viewAsCholmod(const SparseVector<_Scalar,_Options,_Index>& mat)
+{
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.const_cast_derived()));
   return res;
 }
 
 /** Returns a view of the Eigen sparse matrix \a mat as Cholmod sparse matrix.
   * The data are not copied but shared. */
 template<typename _Scalar, int _Options, typename _Index, unsigned int UpLo>
-cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
+cholmod_sparse viewAsCholmod(const SparseSelfAdjointView<const SparseMatrix<_Scalar,_Options,_Index>, UpLo>& mat)
 {
-  cholmod_sparse res = viewAsCholmod(mat.matrix().const_cast_derived());
+  cholmod_sparse res = viewAsCholmod(Ref<SparseMatrix<_Scalar,_Options,_Index> >(mat.matrix().const_cast_derived()));
   
   if(UpLo==Upper) res.stype =  1;
   if(UpLo==Lower) res.stype = -1;
@@ -131,19 +144,19 @@
   res.x      = (void*)(mat.derived().data());
   res.z      = 0;
 
-  internal::cholmod_configure_matrix<Scalar>(res);
+  internal::cholmod_configure_matrix<Scalar>::run(res);
 
   return res;
 }
 
 /** Returns a view of the Cholmod sparse matrix \a cm as an Eigen sparse matrix.
   * The data are not copied but shared. */
-template<typename Scalar, int Flags, typename Index>
-MappedSparseMatrix<Scalar,Flags,Index> viewAsEigen(cholmod_sparse& cm)
+template<typename Scalar, int Flags, typename StorageIndex>
+MappedSparseMatrix<Scalar,Flags,StorageIndex> viewAsEigen(cholmod_sparse& cm)
 {
-  return MappedSparseMatrix<Scalar,Flags,Index>
-         (cm.nrow, cm.ncol, static_cast<Index*>(cm.p)[cm.ncol],
-          static_cast<Index*>(cm.p), static_cast<Index*>(cm.i),static_cast<Scalar*>(cm.x) );
+  return MappedSparseMatrix<Scalar,Flags,StorageIndex>
+         (cm.nrow, cm.ncol, static_cast<StorageIndex*>(cm.p)[cm.ncol],
+          static_cast<StorageIndex*>(cm.p), static_cast<StorageIndex*>(cm.i),static_cast<Scalar*>(cm.x) );
 }
 
 enum CholmodMode {
@@ -157,29 +170,39 @@
   * \sa class CholmodSupernodalLLT, class CholmodSimplicialLDLT, class CholmodSimplicialLLT
   */
 template<typename _MatrixType, int _UpLo, typename Derived>
-class CholmodBase : internal::noncopyable
+class CholmodBase : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
   public:
     typedef _MatrixType MatrixType;
     enum { UpLo = _UpLo };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
     typedef MatrixType CholMatrixType;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
 
     CholmodBase()
-      : m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
+      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
     {
-      m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
+      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
+      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
       cholmod_start(&m_cholmod);
     }
 
-    CholmodBase(const MatrixType& matrix)
-      : m_cholmodFactor(0), m_info(Success), m_isInitialized(false)
+    explicit CholmodBase(const MatrixType& matrix)
+      : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false)
     {
-      m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0);
+      EIGEN_STATIC_ASSERT((internal::is_same<double,RealScalar>::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY);
+      m_shiftOffset[0] = m_shiftOffset[1] = 0.0;
       cholmod_start(&m_cholmod);
       compute(matrix);
     }
@@ -191,11 +214,8 @@
       cholmod_finish(&m_cholmod);
     }
     
-    inline Index cols() const { return m_cholmodFactor->n; }
-    inline Index rows() const { return m_cholmodFactor->n; }
-    
-    Derived& derived() { return *static_cast<Derived*>(this); }
-    const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    inline StorageIndex cols() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
+    inline StorageIndex rows() const { return internal::convert_index<StorageIndex, Index>(m_cholmodFactor->n); }
     
     /** \brief Reports whether previous computation was successful.
       *
@@ -216,34 +236,6 @@
       return derived();
     }
     
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<CholmodBase, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "LLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<CholmodBase, Rhs>(*this, b.derived());
-    }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<CholmodBase, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "LLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<CholmodBase, Rhs>(*this, b.derived());
-    }
-    
     /** Performs a symbolic decomposition on the sparsity pattern of \a matrix.
       *
       * This function is particularly useful when solving for several problems having the same structure.
@@ -277,7 +269,7 @@
       eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
       cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView<UpLo>());
       cholmod_factorize_p(&A, m_shiftOffset, 0, 0, m_cholmodFactor, &m_cholmod);
-      
+
       // If the factorization failed, minor is the column at which it did. On success minor == n.
       this->m_info = (m_cholmodFactor->minor == m_cholmodFactor->n ? Success : NumericalIssue);
       m_factorizationIsOk = true;
@@ -290,20 +282,22 @@
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       const Index size = m_cholmodFactor->n;
       EIGEN_UNUSED_VARIABLE(size);
       eigen_assert(size==b.rows());
+      
+      // Cholmod needs column-major stoarge without inner-stride, which corresponds to the default behavior of Ref.
+      Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b.derived());
 
-      // note: cd stands for Cholmod Dense
-      Rhs& b_ref(b.const_cast_derived());
       cholmod_dense b_cd = viewAsCholmod(b_ref);
       cholmod_dense* x_cd = cholmod_solve(CHOLMOD_A, m_cholmodFactor, &b_cd, &m_cholmod);
       if(!x_cd)
       {
         this->m_info = NumericalIssue;
+        return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
       dest = Matrix<Scalar,Dest::RowsAtCompileTime,Dest::ColsAtCompileTime>::Map(reinterpret_cast<Scalar*>(x_cd->x),b.rows(),b.cols());
@@ -311,8 +305,8 @@
     }
     
     /** \internal */
-    template<typename RhsScalar, int RhsOptions, typename RhsIndex, typename DestScalar, int DestOptions, typename DestIndex>
-    void _solve(const SparseMatrix<RhsScalar,RhsOptions,RhsIndex> &b, SparseMatrix<DestScalar,DestOptions,DestIndex> &dest) const
+    template<typename RhsDerived, typename DestDerived>
+    void _solve_impl(const SparseMatrixBase<RhsDerived> &b, SparseMatrixBase<DestDerived> &dest) const
     {
       eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       const Index size = m_cholmodFactor->n;
@@ -320,14 +314,16 @@
       eigen_assert(size==b.rows());
 
       // note: cs stands for Cholmod Sparse
-      cholmod_sparse b_cs = viewAsCholmod(b);
+      Ref<SparseMatrix<typename RhsDerived::Scalar,ColMajor,typename RhsDerived::StorageIndex> > b_ref(b.const_cast_derived());
+      cholmod_sparse b_cs = viewAsCholmod(b_ref);
       cholmod_sparse* x_cs = cholmod_spsolve(CHOLMOD_A, m_cholmodFactor, &b_cs, &m_cholmod);
       if(!x_cs)
       {
         this->m_info = NumericalIssue;
+        return;
       }
       // TODO optimize this copy by swapping when possible (be careful with alignment, etc.)
-      dest = viewAsEigen<DestScalar,DestOptions,DestIndex>(*x_cs);
+      dest.derived() = viewAsEigen<typename DestDerived::Scalar,ColMajor,typename DestDerived::StorageIndex>(*x_cs);
       cholmod_free_sparse(&x_cs, &m_cholmod);
     }
     #endif // EIGEN_PARSED_BY_DOXYGEN
@@ -344,10 +340,61 @@
       */
     Derived& setShift(const RealScalar& offset)
     {
-      m_shiftOffset[0] = offset;
+      m_shiftOffset[0] = double(offset);
       return derived();
     }
     
+    /** \returns the determinant of the underlying matrix from the current factorization */
+    Scalar determinant() const
+    {
+      using std::exp;
+      return exp(logDeterminant());
+    }
+
+    /** \returns the log determinant of the underlying matrix from the current factorization */
+    Scalar logDeterminant() const
+    {
+      using std::log;
+      using numext::real;
+      eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
+
+      RealScalar logDet = 0;
+      Scalar *x = static_cast<Scalar*>(m_cholmodFactor->x);
+      if (m_cholmodFactor->is_super)
+      {
+        // Supernodal factorization stored as a packed list of dense column-major blocs,
+        // as described by the following structure:
+
+        // super[k] == index of the first column of the j-th super node
+        StorageIndex *super = static_cast<StorageIndex*>(m_cholmodFactor->super);
+        // pi[k] == offset to the description of row indices
+        StorageIndex *pi = static_cast<StorageIndex*>(m_cholmodFactor->pi);
+        // px[k] == offset to the respective dense block
+        StorageIndex *px = static_cast<StorageIndex*>(m_cholmodFactor->px);
+
+        Index nb_super_nodes = m_cholmodFactor->nsuper;
+        for (Index k=0; k < nb_super_nodes; ++k)
+        {
+          StorageIndex ncols = super[k + 1] - super[k];
+          StorageIndex nrows = pi[k + 1] - pi[k];
+
+          Map<const Array<Scalar,1,Dynamic>, 0, InnerStride<> > sk(x + px[k], ncols, InnerStride<>(nrows+1));
+          logDet += sk.real().log().sum();
+        }
+      }
+      else
+      {
+        // Simplicial factorization stored as standard CSC matrix.
+        StorageIndex *p = static_cast<StorageIndex*>(m_cholmodFactor->p);
+        Index size = m_cholmodFactor->n;
+        for (Index k=0; k<size; ++k)
+          logDet += log(real( x[p[k]] ));
+      }
+      if (m_cholmodFactor->is_ll)
+        logDet *= 2.0;
+      return logDet;
+    };
+
     template<typename Stream>
     void dumpMemory(Stream& /*s*/)
     {}
@@ -355,9 +402,8 @@
   protected:
     mutable cholmod_common m_cholmod;
     cholmod_factor* m_cholmodFactor;
-    RealScalar m_shiftOffset[2];
+    double m_shiftOffset[2];
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
 };
@@ -376,9 +422,13 @@
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLLT
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLLT
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT<_MatrixType, _UpLo> >
@@ -395,7 +445,7 @@
     CholmodSimplicialLLT(const MatrixType& matrix) : Base()
     {
       init();
-      compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodSimplicialLLT() {}
@@ -423,9 +473,13 @@
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLDLT
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLDLT
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT<_MatrixType, _UpLo> >
@@ -442,7 +496,7 @@
     CholmodSimplicialLDLT(const MatrixType& matrix) : Base()
     {
       init();
-      compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodSimplicialLDLT() {}
@@ -468,9 +522,13 @@
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT<_MatrixType, _UpLo> >
@@ -487,7 +545,7 @@
     CholmodSupernodalLLT(const MatrixType& matrix) : Base()
     {
       init();
-      compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodSupernodalLLT() {}
@@ -515,9 +573,13 @@
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
   *               or Upper. Default is Lower.
   *
+  * \implsparsesolverconcept
+  *
   * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \warning Only double precision real and complex scalar types are supported by Cholmod.
+  *
+  * \sa \ref TutorialSparseSolverConcept
   */
 template<typename _MatrixType, int _UpLo = Lower>
 class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecomposition<_MatrixType, _UpLo> >
@@ -534,7 +596,7 @@
     CholmodDecomposition(const MatrixType& matrix) : Base()
     {
       init();
-      compute(matrix);
+      this->compute(matrix);
     }
 
     ~CholmodDecomposition() {}
@@ -572,36 +634,6 @@
     }
 };
 
-namespace internal {
-  
-template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
-struct solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-  : solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-{
-  typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, int _UpLo, typename Derived, typename Rhs>
-struct sparse_solve_retval<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-  : sparse_solve_retval_base<CholmodBase<_MatrixType,_UpLo,Derived>, Rhs>
-{
-  typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_CHOLMODSUPPORT_H
diff --git a/Eigen/src/Core/Array.h b/Eigen/src/Core/Array.h
index 0ab03ef..16770fc 100644
--- a/Eigen/src/Core/Array.h
+++ b/Eigen/src/Core/Array.h
@@ -12,7 +12,16 @@
 
 namespace Eigen {
 
-/** \class Array 
+namespace internal {
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+  typedef ArrayXpr XprKind;
+  typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
+};
+}
+
+/** \class Array
   * \ingroup Core_Module
   *
   * \brief General-purpose arrays with easy API for coefficient-wise operations
@@ -24,20 +33,14 @@
   * API for the %Matrix class provides easy access to linear-algebra
   * operations.
   *
-  * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
+  * See documentation of class Matrix for detailed information on the template parameters
+  * storage layout.
   *
-  * \sa \ref TutorialArrayClass, \ref TopicClassHierarchy
+  * This class can be extended with the help of the plugin mechanism described on the page
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN.
+  *
+  * \sa \blank \ref TutorialArrayClass, \ref TopicClassHierarchy
   */
-namespace internal {
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct traits<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > : traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-{
-  typedef ArrayXpr XprKind;
-  typedef ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > XprBase;
-};
-}
-
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Array
   : public PlainObjectBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
@@ -69,11 +72,27 @@
       * the usage of 'using'. This should be done only for operator=.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array& operator=(const EigenBase<OtherDerived> &other)
     {
       return Base::operator=(other);
     }
 
+    /** Set all the entries to \a value.
+      * \sa DenseBase::setConstant(), DenseBase::fill()
+      */
+    /* This overload is needed because the usage of
+      *   using Base::operator=;
+      * fails on MSVC. Since the code below is working with GCC and MSVC, we skipped
+      * the usage of 'using'. This should be done only for operator=.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const Scalar &value)
+    {
+      Base::setConstant(value);
+      return *this;
+    }
+
     /** Copies the value of the expression \a other into \c *this with automatic resizing.
       *
       * *this might be resized to match the dimensions of \a other. If *this was a null matrix (not already initialized),
@@ -84,7 +103,8 @@
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array& operator=(const ArrayBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array& operator=(const DenseBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -92,11 +112,12 @@
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array& operator=(const Array& other)
     {
       return Base::_set(other);
     }
-
+    
     /** Default constructor.
       *
       * For fixed-size matrices, does nothing.
@@ -107,6 +128,7 @@
       *
       * \sa resize(Index,Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array() : Base()
     {
       Base::_check_template_params();
@@ -116,6 +138,7 @@
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     // FIXME is it still needed ??
     /** \internal */
+    EIGEN_DEVICE_FUNC
     Array(internal::constructor_without_unaligned_array_assert)
       : Base(internal::constructor_without_unaligned_array_assert())
     {
@@ -124,41 +147,62 @@
     }
 #endif
 
-    /** Constructs a vector or row-vector with given dimension. \only_for_vectors
-      *
-      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
-      * it is redundant to pass the dimension here, so it makes more sense to use the default
-      * constructor Matrix() instead.
-      */
-    EIGEN_STRONG_INLINE explicit Array(Index dim)
-      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
+      : Base(std::move(other))
     {
       Base::_check_template_params();
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Array)
-      eigen_assert(dim >= 0);
-      eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
-      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
+    EIGEN_DEVICE_FUNC
+    Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
+    {
+      other.swap(*this);
+      return *this;
+    }
+#endif
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Array(const T& x)
+    {
+      Base::_check_template_params();
+      Base::template _init1<T>(x);
+    }
+
     template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const T0& val0, const T1& val1)
     {
       Base::_check_template_params();
       this->template _init2<T0,T1>(val0, val1);
     }
     #else
-    /** constructs an uninitialized matrix with \a rows rows and \a cols columns.
+    /** \brief Constructs a fixed-sized array initialized with coefficients starting at \a data */
+    EIGEN_DEVICE_FUNC explicit Array(const Scalar *data);
+    /** Constructs a vector or row-vector with given dimension. \only_for_vectors
       *
-      * This is useful for dynamic-size matrices. For fixed-size matrices,
+      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
+      * it is redundant to pass the dimension here, so it makes more sense to use the default
+      * constructor Array() instead.
+      */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Array(Index dim);
+    /** constructs an initialized 1x1 Array with the given coefficient */
+    Array(const Scalar& value);
+    /** constructs an uninitialized array with \a rows rows and \a cols columns.
+      *
+      * This is useful for dynamic-size arrays. For fixed-size arrays,
       * it is redundant to pass these parameters, so one should use the default constructor
-      * Matrix() instead. */
+      * Array() instead. */
     Array(Index rows, Index cols);
     /** constructs an initialized 2D vector with given coefficients */
     Array(const Scalar& val0, const Scalar& val1);
     #endif
 
     /** constructs an initialized 3D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2)
     {
       Base::_check_template_params();
@@ -168,6 +212,7 @@
       m_storage.data()[2] = val2;
     }
     /** constructs an initialized 4D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3)
     {
       Base::_check_template_params();
@@ -178,51 +223,27 @@
       m_storage.data()[3] = val3;
     }
 
-    explicit Array(const Scalar *data);
-
-    /** Constructor copying the value of the expression \a other */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const ArrayBase<OtherDerived>& other)
-             : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
     /** Copy constructor */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Array(const Array& other)
-            : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
-    /** Copy constructor with in-place evaluation */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const ReturnByValue<OtherDerived>& other)
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      other.evalTo(*this);
-    }
+            : Base(other)
+    { }
+
+  private:
+    struct PrivateType {};
+  public:
 
     /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other)
-      : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      Base::_check_template_params();
-      Base::_resize_to_match(other);
-      *this = other;
-    }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Array(const EigenBase<OtherDerived> &other,
+                              typename internal::enable_if<internal::is_convertible<typename OtherDerived::Scalar,Scalar>::value,
+                                                           PrivateType>::type = PrivateType())
+      : Base(other.derived())
+    { }
 
-    /** Override MatrixBase::swap() since for dynamic-sized matrices of same type it is enough to swap the
-      * data pointers.
-      */
-    template<typename OtherDerived>
-    void swap(ArrayBase<OtherDerived> const & other)
-    { this->_swap(other.derived()); }
-
-    inline Index innerStride() const { return 1; }
-    inline Index outerStride() const { return this->innerSize(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
 
     #ifdef EIGEN_ARRAY_PLUGIN
     #include EIGEN_ARRAY_PLUGIN
diff --git a/Eigen/src/Core/ArrayBase.h b/Eigen/src/Core/ArrayBase.h
index 3885260..3dbc708 100644
--- a/Eigen/src/Core/ArrayBase.h
+++ b/Eigen/src/Core/ArrayBase.h
@@ -32,7 +32,7 @@
   * \tparam Derived is the derived type, e.g., an array or an expression type.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN.
   *
   * \sa class MatrixBase, \ref TopicClassHierarchy
   */
@@ -46,11 +46,7 @@
 
     typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl;
 
-    using internal::special_scalar_op_base<Derived,typename internal::traits<Derived>::Scalar,
-                typename NumTraits<typename internal::traits<Derived>::Scalar>::Real>::operator*;
-
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -64,8 +60,7 @@
     using Base::MaxSizeAtCompileTime;
     using Base::IsVectorAtCompileTime;
     using Base::Flags;
-    using Base::CoeffReadCost;
-
+    
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -85,25 +80,14 @@
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-    /** \internal the plain matrix type corresponding to this expression. Note that is not necessarily
-      * exactly the return type of eval(): in the case of plain matrices, the return type of eval() is a const
-      * reference to a matrix, not a matrix! It is however guaranteed that the return type of eval() is either
-      * PlainObject or const PlainObject&.
-      */
-    typedef Array<typename internal::traits<Derived>::Scalar,
-                internal::traits<Derived>::RowsAtCompileTime,
-                internal::traits<Derived>::ColsAtCompileTime,
-                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
-                internal::traits<Derived>::MaxRowsAtCompileTime,
-                internal::traits<Derived>::MaxColsAtCompileTime
-          > PlainObject;
-
+    typedef typename Base::PlainObject PlainObject;
 
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase
+#define EIGEN_DOC_UNARY_ADDONS(X,Y)
 #   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
 #   include "../plugins/ArrayCwiseUnaryOps.h"
@@ -114,44 +98,62 @@
 #     include EIGEN_ARRAYBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_UNARY_ADDONS
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const ArrayBase& other)
     {
-      return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
+      internal::call_assignment(derived(), other.derived());
+      return derived();
     }
+    
+    /** Set all the entries to \a value.
+      * \sa DenseBase::setConstant(), DenseBase::fill() */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator=(const Scalar &value)
+    { Base::setConstant(value); return derived(); }
 
-    Derived& operator+=(const Scalar& scalar)
-    { return *this = derived() + scalar; }
-    Derived& operator-=(const Scalar& scalar)
-    { return *this = derived() - scalar; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator+=(const Scalar& scalar);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator-=(const Scalar& scalar);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const ArrayBase<OtherDerived>& other);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator*=(const ArrayBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator/=(const ArrayBase<OtherDerived>& other);
 
   public:
+    EIGEN_DEVICE_FUNC
     ArrayBase<Derived>& array() { return *this; }
+    EIGEN_DEVICE_FUNC
     const ArrayBase<Derived>& array() const { return *this; }
 
     /** \returns an \link Eigen::MatrixBase Matrix \endlink expression of this array
       * \sa MatrixBase::array() */
-    MatrixWrapper<Derived> matrix() { return derived(); }
-    const MatrixWrapper<const Derived> matrix() const { return derived(); }
+    EIGEN_DEVICE_FUNC
+    MatrixWrapper<Derived> matrix() { return MatrixWrapper<Derived>(derived()); }
+    EIGEN_DEVICE_FUNC
+    const MatrixWrapper<const Derived> matrix() const { return MatrixWrapper<const Derived>(derived()); }
 
 //     template<typename Dest>
 //     inline void evalTo(Dest& dst) const { dst = matrix(); }
 
   protected:
+    EIGEN_DEVICE_FUNC
     ArrayBase() : Base() {}
 
   private:
@@ -173,11 +175,10 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator-=(const ArrayBase<OtherDerived> &other)
 {
-  SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -187,11 +188,10 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator+=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -201,11 +201,10 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator*=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::mul_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -215,11 +214,10 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE Derived &
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived &
 ArrayBase<Derived>::operator/=(const ArrayBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_quotient_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::div_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
diff --git a/Eigen/src/Core/ArrayWrapper.h b/Eigen/src/Core/ArrayWrapper.h
index b4641e2..688aadd 100644
--- a/Eigen/src/Core/ArrayWrapper.h
+++ b/Eigen/src/Core/ArrayWrapper.h
@@ -32,7 +32,8 @@
   // Let's remove NestByRefBit
   enum {
     Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    Flags = Flags0 & ~NestByRefBit
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
   };
 };
 }
@@ -44,6 +45,7 @@
     typedef ArrayBase<ArrayWrapper> Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(ArrayWrapper)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ArrayWrapper)
+    typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
 
     typedef typename internal::conditional<
                        internal::is_lvalue<ExpressionType>::value,
@@ -51,76 +53,45 @@
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
 
-    inline ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+    using Base::coeffRef;
 
+    EIGEN_DEVICE_FUNC
+    explicit EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
+    EIGEN_DEVICE_FUNC
     inline const Scalar* data() const { return m_expression.data(); }
 
-    inline CoeffReturnType coeff(Index rowId, Index colId) const
-    {
-      return m_expression.coeff(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index rowId, Index colId)
-    {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
-    }
-
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
+      return m_expression.coeffRef(rowId, colId);
     }
 
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return m_expression.coeff(index);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index rowId, Index colId) const
-    {
-      return m_expression.template packet<LoadMode>(rowId, colId);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return m_expression.template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
+      return m_expression.coeffRef(index);
     }
 
     template<typename Dest>
+    EIGEN_DEVICE_FUNC
     inline void evalTo(Dest& dst) const { dst = m_expression; }
 
     const typename internal::remove_all<NestedExpressionType>::type& 
+    EIGEN_DEVICE_FUNC
     nestedExpression() const 
     {
       return m_expression;
@@ -128,10 +99,12 @@
 
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index)  */
-    void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize) { m_expression.resize(newSize); }
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index,Index)*/
-    void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
 
   protected:
     NestedExpressionType m_expression;
@@ -157,7 +130,8 @@
   // Let's remove NestByRefBit
   enum {
     Flags0 = traits<typename remove_all<typename ExpressionType::Nested>::type >::Flags,
-    Flags = Flags0 & ~NestByRefBit
+    LvalueBitFlag = is_lvalue<ExpressionType>::value ? LvalueBit : 0,
+    Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag
   };
 };
 }
@@ -169,6 +143,7 @@
     typedef MatrixBase<MatrixWrapper<ExpressionType> > Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(MatrixWrapper)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(MatrixWrapper)
+    typedef typename internal::remove_all<ExpressionType>::type NestedExpression;
 
     typedef typename internal::conditional<
                        internal::is_lvalue<ExpressionType>::value,
@@ -176,72 +151,40 @@
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    typedef typename internal::nested<ExpressionType>::type NestedExpressionType;
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type NestedExpressionType;
 
-    inline MatrixWrapper(ExpressionType& a_matrix) : m_expression(a_matrix) {}
+    using Base::coeffRef;
 
+    EIGEN_DEVICE_FUNC
+    explicit inline MatrixWrapper(ExpressionType& matrix) : m_expression(matrix) {}
+
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); }
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
+    EIGEN_DEVICE_FUNC
     inline const Scalar* data() const { return m_expression.data(); }
 
-    inline CoeffReturnType coeff(Index rowId, Index colId) const
-    {
-      return m_expression.coeff(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index rowId, Index colId)
-    {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
-    }
-
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return m_expression.derived().coeffRef(rowId, colId);
     }
 
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return m_expression.coeff(index);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
-      return m_expression.const_cast_derived().coeffRef(index);
+      return m_expression.coeffRef(index);
     }
 
-    template<int LoadMode>
-    inline const PacketScalar packet(Index rowId, Index colId) const
-    {
-      return m_expression.template packet<LoadMode>(rowId, colId);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(rowId, colId, val);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return m_expression.template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& val)
-    {
-      m_expression.const_cast_derived().template writePacket<LoadMode>(index, val);
-    }
-
+    EIGEN_DEVICE_FUNC
     const typename internal::remove_all<NestedExpressionType>::type& 
     nestedExpression() const 
     {
@@ -250,10 +193,12 @@
 
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index)  */
-    void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index newSize) { m_expression.resize(newSize); }
     /** Forwards the resizing request to the nested expression
       * \sa DenseBase::resize(Index,Index)*/
-    void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); }
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols) { m_expression.resize(rows,cols); }
 
   protected:
     NestedExpressionType m_expression;
diff --git a/Eigen/src/Core/Assign.h b/Eigen/src/Core/Assign.h
index bcfc261..53806ba 100644
--- a/Eigen/src/Core/Assign.h
+++ b/Eigen/src/Core/Assign.h
@@ -14,478 +14,6 @@
 
 namespace Eigen {
 
-namespace internal {
-
-/***************************************************************************
-* Part 1 : the logic deciding a strategy for traversal and unrolling       *
-***************************************************************************/
-
-template <typename Derived, typename OtherDerived>
-struct assign_traits
-{
-public:
-  enum {
-    DstIsAligned = Derived::Flags & AlignedBit,
-    DstHasDirectAccess = Derived::Flags & DirectAccessBit,
-    SrcIsAligned = OtherDerived::Flags & AlignedBit,
-    JointAlignment = bool(DstIsAligned) && bool(SrcIsAligned) ? Aligned : Unaligned
-  };
-
-private:
-  enum {
-    InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime)
-              : int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime)
-              : int(Derived::RowsAtCompileTime),
-    InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime)
-              : int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime)
-              : int(Derived::MaxRowsAtCompileTime),
-    MaxSizeAtCompileTime = Derived::SizeAtCompileTime,
-    PacketSize = packet_traits<typename Derived::Scalar>::size
-  };
-
-  enum {
-    StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)),
-    MightVectorize = StorageOrdersAgree
-                  && (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit),
-    MayInnerVectorize  = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0
-                       && int(DstIsAligned) && int(SrcIsAligned),
-    MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit),
-    MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess
-                       && (DstIsAligned || MaxSizeAtCompileTime == Dynamic),
-      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
-         so it's only good for large enough sizes. */
-    MaySliceVectorize  = MightVectorize && DstHasDirectAccess
-                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize)
-      /* slice vectorization can be slow, so we only want it if the slices are big, which is
-         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
-         in a fixed-size matrix */
-  };
-
-public:
-  enum {
-    Traversal = int(MayInnerVectorize)  ? int(InnerVectorizedTraversal)
-              : int(MayLinearVectorize) ? int(LinearVectorizedTraversal)
-              : int(MaySliceVectorize)  ? int(SliceVectorizedTraversal)
-              : int(MayLinearize)       ? int(LinearTraversal)
-                                        : int(DefaultTraversal),
-    Vectorized = int(Traversal) == InnerVectorizedTraversal
-              || int(Traversal) == LinearVectorizedTraversal
-              || int(Traversal) == SliceVectorizedTraversal
-  };
-
-private:
-  enum {
-    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1),
-    MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic
-                       && int(OtherDerived::CoeffReadCost) != Dynamic
-                       && int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit),
-    MayUnrollInner      = int(InnerSize) != Dynamic
-                       && int(OtherDerived::CoeffReadCost) != Dynamic
-                       && int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit)
-  };
-
-public:
-  enum {
-    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
-                ? (
-                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
-                  : int(MayUnrollInner)      ? int(InnerUnrolling)
-                                             : int(NoUnrolling)
-                  )
-              : int(Traversal) == int(LinearVectorizedTraversal)
-                ? ( bool(MayUnrollCompletely) && bool(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) )
-              : int(Traversal) == int(LinearTraversal)
-                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) : int(NoUnrolling) )
-              : int(NoUnrolling)
-  };
-
-#ifdef EIGEN_DEBUG_ASSIGN
-  static void debug()
-  {
-    EIGEN_DEBUG_VAR(DstIsAligned)
-    EIGEN_DEBUG_VAR(SrcIsAligned)
-    EIGEN_DEBUG_VAR(JointAlignment)
-    EIGEN_DEBUG_VAR(InnerSize)
-    EIGEN_DEBUG_VAR(InnerMaxSize)
-    EIGEN_DEBUG_VAR(PacketSize)
-    EIGEN_DEBUG_VAR(StorageOrdersAgree)
-    EIGEN_DEBUG_VAR(MightVectorize)
-    EIGEN_DEBUG_VAR(MayLinearize)
-    EIGEN_DEBUG_VAR(MayInnerVectorize)
-    EIGEN_DEBUG_VAR(MayLinearVectorize)
-    EIGEN_DEBUG_VAR(MaySliceVectorize)
-    EIGEN_DEBUG_VAR(Traversal)
-    EIGEN_DEBUG_VAR(UnrollingLimit)
-    EIGEN_DEBUG_VAR(MayUnrollCompletely)
-    EIGEN_DEBUG_VAR(MayUnrollInner)
-    EIGEN_DEBUG_VAR(Unrolling)
-  }
-#endif
-};
-
-/***************************************************************************
-* Part 2 : meta-unrollers
-***************************************************************************/
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_DefaultTraversal_CompleteUnrolling
-{
-  enum {
-    outer = Index / Derived1::InnerSizeAtCompileTime,
-    inner = Index % Derived1::InnerSizeAtCompileTime
-  };
-
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.copyCoeffByOuterInner(outer, inner, src);
-    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_DefaultTraversal_InnerUnrolling
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
-  {
-    dst.copyCoeffByOuterInner(outer, Index, src);
-    assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src, outer);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_LinearTraversal_CompleteUnrolling
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.copyCoeff(Index, src);
-    assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Index+1, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_innervec_CompleteUnrolling
-{
-  enum {
-    outer = Index / Derived1::InnerSizeAtCompileTime,
-    inner = Index % Derived1::InnerSizeAtCompileTime,
-    JointAlignment = assign_traits<Derived1,Derived2>::JointAlignment
-  };
-
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    dst.template copyPacketByOuterInner<Derived2, Aligned, JointAlignment>(outer, inner, src);
-    assign_innervec_CompleteUnrolling<Derived1, Derived2,
-      Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_innervec_CompleteUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, int Index, int Stop>
-struct assign_innervec_InnerUnrolling
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer)
-  {
-    dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, Index, src);
-    assign_innervec_InnerUnrolling<Derived1, Derived2,
-      Index+packet_traits<typename Derived1::Scalar>::size, Stop>::run(dst, src, outer);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Stop>
-struct assign_innervec_InnerUnrolling<Derived1, Derived2, Stop, Stop>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {}
-};
-
-/***************************************************************************
-* Part 3 : implementation of all cases
-***************************************************************************/
-
-template<typename Derived1, typename Derived2,
-         int Traversal = assign_traits<Derived1, Derived2>::Traversal,
-         int Unrolling = assign_traits<Derived1, Derived2>::Unrolling,
-         int Version = Specialized>
-struct assign_impl;
-
-/************************
-*** Default traversal ***
-************************/
-
-template<typename Derived1, typename Derived2, int Unrolling, int Version>
-struct assign_impl<Derived1, Derived2, InvalidTraversal, Unrolling, Version>
-{
-  static inline void run(Derived1 &, const Derived2 &) { }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, DefaultTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      for(Index inner = 0; inner < innerSize; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, DefaultTraversal, CompleteUnrolling, Version>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, DefaultTraversal, InnerUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      assign_DefaultTraversal_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
-        ::run(dst, src, outer);
-  }
-};
-
-/***********************
-*** Linear traversal ***
-***********************/
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index size = dst.size();
-    for(Index i = 0; i < size; ++i)
-      dst.copyCoeff(i, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearTraversal, CompleteUnrolling, Version>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    assign_LinearTraversal_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-/**************************
-*** Inner vectorization ***
-**************************/
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    const Index packetSize = packet_traits<typename Derived1::Scalar>::size;
-    for(Index outer = 0; outer < outerSize; ++outer)
-      for(Index inner = 0; inner < innerSize; inner+=packetSize)
-        dst.template copyPacketByOuterInner<Derived2, Aligned, Aligned>(outer, inner, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, CompleteUnrolling, Version>
-{
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, Derived1::SizeAtCompileTime>
-      ::run(dst, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, InnerVectorizedTraversal, InnerUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer)
-      assign_innervec_InnerUnrolling<Derived1, Derived2, 0, Derived1::InnerSizeAtCompileTime>
-        ::run(dst, src, outer);
-  }
-};
-
-/***************************
-*** Linear vectorization ***
-***************************/
-
-template <bool IsAligned = false>
-struct unaligned_assign_impl
-{
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_STRONG_INLINE void run(const Derived&, OtherDerived&, typename Derived::Index, typename Derived::Index) {}
-};
-
-template <>
-struct unaligned_assign_impl<false>
-{
-  // MSVC must not inline this functions. If it does, it fails to optimize the
-  // packet access path.
-#ifdef _MSC_VER
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_DONT_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
-#else
-  template <typename Derived, typename OtherDerived>
-  static EIGEN_STRONG_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end)
-#endif
-  {
-    for (typename Derived::Index index = start; index < end; ++index)
-      dst.copyCoeff(index, src);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    const Index size = dst.size();
-    typedef packet_traits<typename Derived1::Scalar> PacketTraits;
-    enum {
-      packetSize = PacketTraits::size,
-      dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : int(assign_traits<Derived1,Derived2>::DstIsAligned) ,
-      srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
-    };
-    const Index alignedStart = assign_traits<Derived1,Derived2>::DstIsAligned ? 0
-                             : internal::first_aligned(&dst.coeffRef(0), size);
-    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
-
-    unaligned_assign_impl<assign_traits<Derived1,Derived2>::DstIsAligned!=0>::run(src,dst,0,alignedStart);
-
-    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
-    {
-      dst.template copyPacket<Derived2, dstAlignment, srcAlignment>(index, src);
-    }
-
-    unaligned_assign_impl<>::run(src,dst,alignedEnd,size);
-  }
-};
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, LinearVectorizedTraversal, CompleteUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src)
-  {
-    enum { size = Derived1::SizeAtCompileTime,
-           packetSize = packet_traits<typename Derived1::Scalar>::size,
-           alignedSize = (size/packetSize)*packetSize };
-
-    assign_innervec_CompleteUnrolling<Derived1, Derived2, 0, alignedSize>::run(dst, src);
-    assign_DefaultTraversal_CompleteUnrolling<Derived1, Derived2, alignedSize, size>::run(dst, src);
-  }
-};
-
-/**************************
-*** Slice vectorization ***
-***************************/
-
-template<typename Derived1, typename Derived2, int Version>
-struct assign_impl<Derived1, Derived2, SliceVectorizedTraversal, NoUnrolling, Version>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    typedef typename Derived1::Scalar Scalar;
-    typedef packet_traits<Scalar> PacketTraits;
-    enum {
-      packetSize = PacketTraits::size,
-      alignable = PacketTraits::AlignedOnScalar,
-      dstIsAligned = assign_traits<Derived1,Derived2>::DstIsAligned,
-      dstAlignment = alignable ? Aligned : int(dstIsAligned),
-      srcAlignment = assign_traits<Derived1,Derived2>::JointAlignment
-    };
-    const Scalar *dst_ptr = &dst.coeffRef(0,0);
-    if((!bool(dstIsAligned)) && (Index(dst_ptr) % sizeof(Scalar))>0)
-    {
-      // the pointer is not aligend-on scalar, so alignment is not possible
-      return assign_impl<Derived1,Derived2,DefaultTraversal,NoUnrolling>::run(dst, src);
-    }
-    const Index packetAlignedMask = packetSize - 1;
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    const Index alignedStep = alignable ? (packetSize - dst.outerStride() % packetSize) & packetAlignedMask : 0;
-    Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned(dst_ptr, innerSize);
-
-    for(Index outer = 0; outer < outerSize; ++outer)
-    {
-      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
-      // do the non-vectorizable part of the assignment
-      for(Index inner = 0; inner<alignedStart ; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-
-      // do the vectorizable part of the assignment
-      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
-        dst.template copyPacketByOuterInner<Derived2, dstAlignment, Unaligned>(outer, inner, src);
-
-      // do the non-vectorizable part of the assignment
-      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
-        dst.copyCoeffByOuterInner(outer, inner, src);
-
-      alignedStart = std::min<Index>((alignedStart+alignedStep)%packetSize, innerSize);
-    }
-  }
-};
-
-} // end namespace internal
-
-/***************************************************************************
-* Part 4 : implementation of DenseBase methods
-***************************************************************************/
-
 template<typename Derived>
 template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived& DenseBase<Derived>
@@ -499,90 +27,62 @@
   EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived)
   EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
 
-#ifdef EIGEN_DEBUG_ASSIGN
-  internal::assign_traits<Derived, OtherDerived>::debug();
-#endif
   eigen_assert(rows() == other.rows() && cols() == other.cols());
-  internal::assign_impl<Derived, OtherDerived, int(SameType) ? int(internal::assign_traits<Derived, OtherDerived>::Traversal)
-                                                       : int(InvalidTraversal)>::run(derived(),other.derived());
-#ifndef EIGEN_NO_DEBUG
-  checkTransposeAliasing(other.derived());
-#endif
+  internal::call_assignment_no_alias(derived(),other.derived());
+  
   return derived();
 }
 
-namespace internal {
-
-template<typename Derived, typename OtherDerived,
-         bool EvalBeforeAssigning = (int(internal::traits<OtherDerived>::Flags) & EvalBeforeAssigningBit) != 0,
-         bool NeedToTranspose = ((int(Derived::RowsAtCompileTime) == 1 && int(OtherDerived::ColsAtCompileTime) == 1)
-                              |   // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&".
-                                  // revert to || as soon as not needed anymore.
-                                  (int(Derived::ColsAtCompileTime) == 1 && int(OtherDerived::RowsAtCompileTime) == 1))
-                              && int(Derived::SizeAtCompileTime) != 1>
-struct assign_selector;
-
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,false,false> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); }
-  template<typename ActualDerived, typename ActualOtherDerived>
-  static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { other.evalTo(dst); return dst; }
-};
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,true,false> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.eval()); }
-};
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,false,true> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); }
-  template<typename ActualDerived, typename ActualOtherDerived>
-  static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { Transpose<ActualDerived> dstTrans(dst); other.evalTo(dstTrans); return dst; }
-};
-template<typename Derived, typename OtherDerived>
-struct assign_selector<Derived,OtherDerived,true,true> {
-  static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose().eval()); }
-};
-
-} // end namespace internal
-
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator=(const DenseBase& other)
 {
-  return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const MatrixBase& other)
 {
-  return internal::assign_selector<Derived,Derived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
 template <typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const DenseBase<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
 template <typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const EigenBase<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
 {
-  return internal::assign_selector<Derived,OtherDerived,false>::evalTo(derived(), other.derived());
+  other.derived().evalTo(derived());
+  return derived();
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/AssignEvaluator.h b/Eigen/src/Core/AssignEvaluator.h
new file mode 100644
index 0000000..dbe435d
--- /dev/null
+++ b/Eigen/src/Core/AssignEvaluator.h
@@ -0,0 +1,935 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ASSIGN_EVALUATOR_H
+#define EIGEN_ASSIGN_EVALUATOR_H
+
+namespace Eigen {
+
+// This implementation is based on Assign.h
+
+namespace internal {
+  
+/***************************************************************************
+* Part 1 : the logic deciding a strategy for traversal and unrolling       *
+***************************************************************************/
+
+// copy_using_evaluator_traits is based on assign_traits
+
+template <typename DstEvaluator, typename SrcEvaluator, typename AssignFunc>
+struct copy_using_evaluator_traits
+{
+  typedef typename DstEvaluator::XprType Dst;
+  typedef typename Dst::Scalar DstScalar;
+  
+  enum {
+    DstFlags = DstEvaluator::Flags,
+    SrcFlags = SrcEvaluator::Flags
+  };
+  
+public:
+  enum {
+    DstAlignment = DstEvaluator::Alignment,
+    SrcAlignment = SrcEvaluator::Alignment,
+    DstHasDirectAccess = (DstFlags & DirectAccessBit) == DirectAccessBit,
+    JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment)
+  };
+
+private:
+  enum {
+    InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
+              : int(DstFlags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
+              : int(Dst::RowsAtCompileTime),
+    InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime)
+              : int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime)
+              : int(Dst::MaxRowsAtCompileTime),
+    OuterStride = int(outer_stride_at_compile_time<Dst>::ret),
+    MaxSizeAtCompileTime = Dst::SizeAtCompileTime
+  };
+
+  // TODO distinguish between linear traversal and inner-traversals
+  typedef typename find_best_packet<DstScalar,Dst::SizeAtCompileTime>::type LinearPacketType;
+  typedef typename find_best_packet<DstScalar,InnerSize>::type InnerPacketType;
+
+  enum {
+    LinearPacketSize = unpacket_traits<LinearPacketType>::size,
+    InnerPacketSize = unpacket_traits<InnerPacketType>::size
+  };
+
+public:
+  enum {
+    LinearRequiredAlignment = unpacket_traits<LinearPacketType>::alignment,
+    InnerRequiredAlignment = unpacket_traits<InnerPacketType>::alignment
+  };
+
+private:
+  enum {
+    DstIsRowMajor = DstFlags&RowMajorBit,
+    SrcIsRowMajor = SrcFlags&RowMajorBit,
+    StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)),
+    MightVectorize = bool(StorageOrdersAgree)
+                  && (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit)
+                  && bool(functor_traits<AssignFunc>::PacketAccess),
+    MayInnerVectorize  = MightVectorize
+                       && int(InnerSize)!=Dynamic && int(InnerSize)%int(InnerPacketSize)==0
+                       && int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0
+                       && (EIGEN_UNALIGNED_VECTORIZE  || int(JointAlignment)>=int(InnerRequiredAlignment)),
+    MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit),
+    MayLinearVectorize = bool(MightVectorize) && bool(MayLinearize) && bool(DstHasDirectAccess)
+                       && (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || MaxSizeAtCompileTime == Dynamic),
+      /* If the destination isn't aligned, we have to do runtime checks and we don't unroll,
+         so it's only good for large enough sizes. */
+    MaySliceVectorize  = bool(MightVectorize) && bool(DstHasDirectAccess)
+                       && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=(EIGEN_UNALIGNED_VECTORIZE?InnerPacketSize:(3*InnerPacketSize)))
+      /* slice vectorization can be slow, so we only want it if the slices are big, which is
+         indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block
+         in a fixed-size matrix
+         However, with EIGEN_UNALIGNED_VECTORIZE and unrolling, slice vectorization is still worth it */
+  };
+
+public:
+  enum {
+    Traversal = int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize) ? int(LinearVectorizedTraversal)
+              : int(MayInnerVectorize)   ? int(InnerVectorizedTraversal)
+              : int(MayLinearVectorize)  ? int(LinearVectorizedTraversal)
+              : int(MaySliceVectorize)   ? int(SliceVectorizedTraversal)
+              : int(MayLinearize)        ? int(LinearTraversal)
+                                         : int(DefaultTraversal),
+    Vectorized = int(Traversal) == InnerVectorizedTraversal
+              || int(Traversal) == LinearVectorizedTraversal
+              || int(Traversal) == SliceVectorizedTraversal
+  };
+
+  typedef typename conditional<int(Traversal)==LinearVectorizedTraversal, LinearPacketType, InnerPacketType>::type PacketType;
+
+private:
+  enum {
+    ActualPacketSize    = int(Traversal)==LinearVectorizedTraversal ? LinearPacketSize
+                        : Vectorized ? InnerPacketSize
+                        : 1,
+    UnrollingLimit      = EIGEN_UNROLLING_LIMIT * ActualPacketSize,
+    MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic
+                       && int(Dst::SizeAtCompileTime) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit),
+    MayUnrollInner      = int(InnerSize) != Dynamic
+                       && int(InnerSize) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit)
+  };
+
+public:
+  enum {
+    Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal))
+                ? (
+                    int(MayUnrollCompletely) ? int(CompleteUnrolling)
+                  : int(MayUnrollInner)      ? int(InnerUnrolling)
+                                             : int(NoUnrolling)
+                  )
+              : int(Traversal) == int(LinearVectorizedTraversal)
+                ? ( bool(MayUnrollCompletely) && ( EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)))
+                          ? int(CompleteUnrolling)
+                          : int(NoUnrolling) )
+              : int(Traversal) == int(LinearTraversal)
+                ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) 
+                                              : int(NoUnrolling) )
+#if EIGEN_UNALIGNED_VECTORIZE
+              : int(Traversal) == int(SliceVectorizedTraversal)
+                ? ( bool(MayUnrollInner) ? int(InnerUnrolling)
+                                         : int(NoUnrolling) )
+#endif
+              : int(NoUnrolling)
+  };
+
+#ifdef EIGEN_DEBUG_ASSIGN
+  static void debug()
+  {
+    std::cerr << "DstXpr: " << typeid(typename DstEvaluator::XprType).name() << std::endl;
+    std::cerr << "SrcXpr: " << typeid(typename SrcEvaluator::XprType).name() << std::endl;
+    std::cerr.setf(std::ios::hex, std::ios::basefield);
+    std::cerr << "DstFlags" << " = " << DstFlags << " (" << demangle_flags(DstFlags) << " )" << std::endl;
+    std::cerr << "SrcFlags" << " = " << SrcFlags << " (" << demangle_flags(SrcFlags) << " )" << std::endl;
+    std::cerr.unsetf(std::ios::hex);
+    EIGEN_DEBUG_VAR(DstAlignment)
+    EIGEN_DEBUG_VAR(SrcAlignment)
+    EIGEN_DEBUG_VAR(LinearRequiredAlignment)
+    EIGEN_DEBUG_VAR(InnerRequiredAlignment)
+    EIGEN_DEBUG_VAR(JointAlignment)
+    EIGEN_DEBUG_VAR(InnerSize)
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(LinearPacketSize)
+    EIGEN_DEBUG_VAR(InnerPacketSize)
+    EIGEN_DEBUG_VAR(ActualPacketSize)
+    EIGEN_DEBUG_VAR(StorageOrdersAgree)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearize)
+    EIGEN_DEBUG_VAR(MayInnerVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl;
+    EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost)
+    EIGEN_DEBUG_VAR(UnrollingLimit)
+    EIGEN_DEBUG_VAR(MayUnrollCompletely)
+    EIGEN_DEBUG_VAR(MayUnrollInner)
+    std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl;
+    std::cerr << std::endl;
+  }
+#endif
+};
+
+/***************************************************************************
+* Part 2 : meta-unrollers
+***************************************************************************/
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  
+  enum {
+    outer = Index / DstXprType::InnerSizeAtCompileTime,
+    inner = Index % DstXprType::InnerSizeAtCompileTime
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    kernel.assignCoeffByOuterInner(outer, inner);
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+template<typename Kernel, int Index_, int Stop>
+struct copy_using_evaluator_DefaultTraversal_InnerUnrolling
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
+  {
+    kernel.assignCoeffByOuterInner(outer, Index_);
+    copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Index_+1, Stop>::run(kernel, outer);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index) { }
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_LinearTraversal_CompleteUnrolling
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel& kernel)
+  {
+    kernel.assignCoeff(Index);
+    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Index+1, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Kernel, int Index, int Stop>
+struct copy_using_evaluator_innervec_CompleteUnrolling
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  typedef typename Kernel::PacketType PacketType;
+  
+  enum {
+    outer = Index / DstXprType::InnerSizeAtCompileTime,
+    inner = Index % DstXprType::InnerSizeAtCompileTime,
+    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
+    DstAlignment = Kernel::AssignmentTraits::DstAlignment
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
+    enum { NextIndex = Index + unpacket_traits<PacketType>::size };
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, NextIndex, Stop>::run(kernel);
+  }
+};
+
+template<typename Kernel, int Stop>
+struct copy_using_evaluator_innervec_CompleteUnrolling<Kernel, Stop, Stop>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { }
+};
+
+template<typename Kernel, int Index_, int Stop, int SrcAlignment, int DstAlignment>
+struct copy_using_evaluator_innervec_InnerUnrolling
+{
+  typedef typename Kernel::PacketType PacketType;
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer)
+  {
+    kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, Index_);
+    enum { NextIndex = Index_ + unpacket_traits<PacketType>::size };
+    copy_using_evaluator_innervec_InnerUnrolling<Kernel, NextIndex, Stop, SrcAlignment, DstAlignment>::run(kernel, outer);
+  }
+};
+
+template<typename Kernel, int Stop, int SrcAlignment, int DstAlignment>
+struct copy_using_evaluator_innervec_InnerUnrolling<Kernel, Stop, Stop, SrcAlignment, DstAlignment>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &, Index) { }
+};
+
+/***************************************************************************
+* Part 3 : implementation of all cases
+***************************************************************************/
+
+// dense_assignment_loop is based on assign_impl
+
+template<typename Kernel,
+         int Traversal = Kernel::AssignmentTraits::Traversal,
+         int Unrolling = Kernel::AssignmentTraits::Unrolling>
+struct dense_assignment_loop;
+
+/************************
+*** Default traversal ***
+************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel &kernel)
+  {
+    for(Index outer = 0; outer < kernel.outerSize(); ++outer) {
+      for(Index inner = 0; inner < kernel.innerSize(); ++inner) {
+        kernel.assignCoeffByOuterInner(outer, inner);
+      }
+    }
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, DefaultTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+
+    const Index outerSize = kernel.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime>::run(kernel, outer);
+  }
+};
+
+/***************************
+*** Linear vectorization ***
+***************************/
+
+
+// The goal of unaligned_dense_assignment_loop is simply to factorize the handling
+// of the non vectorizable beginning and ending parts
+
+template <bool IsAligned = false>
+struct unaligned_dense_assignment_loop
+{
+  // if IsAligned = true, then do nothing
+  template <typename Kernel>
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index, Index) {}
+};
+
+template <>
+struct unaligned_dense_assignment_loop<false>
+{
+  // MSVC must not inline this functions. If it does, it fails to optimize the
+  // packet access path.
+  // FIXME check which version exhibits this issue
+#if EIGEN_COMP_MSVC
+  template <typename Kernel>
+  static EIGEN_DONT_INLINE void run(Kernel &kernel,
+                                    Index start,
+                                    Index end)
+#else
+  template <typename Kernel>
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel,
+                                      Index start,
+                                      Index end)
+#endif
+  {
+    for (Index index = start; index < end; ++index)
+      kernel.assignCoeff(index);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index size = kernel.size();
+    typedef typename Kernel::Scalar Scalar;
+    typedef typename Kernel::PacketType PacketType;
+    enum {
+      requestedAlignment = Kernel::AssignmentTraits::LinearRequiredAlignment,
+      packetSize = unpacket_traits<PacketType>::size,
+      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
+      dstAlignment = packet_traits<Scalar>::AlignedOnScalar ? int(requestedAlignment)
+                                                            : int(Kernel::AssignmentTraits::DstAlignment),
+      srcAlignment = Kernel::AssignmentTraits::JointAlignment
+    };
+    const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned<requestedAlignment>(kernel.dstDataPtr(), size);
+    const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize;
+
+    unaligned_dense_assignment_loop<dstIsAligned!=0>::run(kernel, 0, alignedStart);
+
+    for(Index index = alignedStart; index < alignedEnd; index += packetSize)
+      kernel.template assignPacket<dstAlignment, srcAlignment, PacketType>(index);
+
+    unaligned_dense_assignment_loop<>::run(kernel, alignedEnd, size);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearVectorizedTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::PacketType PacketType;
+    
+    enum { size = DstXprType::SizeAtCompileTime,
+           packetSize =unpacket_traits<PacketType>::size,
+           alignedSize = (size/packetSize)*packetSize };
+
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, alignedSize>::run(kernel);
+    copy_using_evaluator_DefaultTraversal_CompleteUnrolling<Kernel, alignedSize, size>::run(kernel);
+  }
+};
+
+/**************************
+*** Inner vectorization ***
+**************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, NoUnrolling>
+{
+  typedef typename Kernel::PacketType PacketType;
+  enum {
+    SrcAlignment = Kernel::AssignmentTraits::SrcAlignment,
+    DstAlignment = Kernel::AssignmentTraits::DstAlignment
+  };
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index innerSize = kernel.innerSize();
+    const Index outerSize = kernel.outerSize();
+    const Index packetSize = unpacket_traits<PacketType>::size;
+    for(Index outer = 0; outer < outerSize; ++outer)
+      for(Index inner = 0; inner < innerSize; inner+=packetSize)
+        kernel.template assignPacketByOuterInner<DstAlignment, SrcAlignment, PacketType>(outer, inner);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_innervec_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, InnerVectorizedTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::AssignmentTraits Traits;
+    const Index outerSize = kernel.outerSize();
+    for(Index outer = 0; outer < outerSize; ++outer)
+      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, DstXprType::InnerSizeAtCompileTime,
+                                                   Traits::SrcAlignment, Traits::DstAlignment>::run(kernel, outer);
+  }
+};
+
+/***********************
+*** Linear traversal ***
+***********************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    const Index size = kernel.size();
+    for(Index i = 0; i < size; ++i)
+      kernel.assignCoeff(i);
+  }
+};
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, LinearTraversal, CompleteUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    copy_using_evaluator_LinearTraversal_CompleteUnrolling<Kernel, 0, DstXprType::SizeAtCompileTime>::run(kernel);
+  }
+};
+
+/**************************
+*** Slice vectorization ***
+***************************/
+
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, NoUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::Scalar Scalar;
+    typedef typename Kernel::PacketType PacketType;
+    enum {
+      packetSize = unpacket_traits<PacketType>::size,
+      requestedAlignment = int(Kernel::AssignmentTraits::InnerRequiredAlignment),
+      alignable = packet_traits<Scalar>::AlignedOnScalar || int(Kernel::AssignmentTraits::DstAlignment)>=sizeof(Scalar),
+      dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment),
+      dstAlignment = alignable ? int(requestedAlignment)
+                               : int(Kernel::AssignmentTraits::DstAlignment)
+    };
+    const Scalar *dst_ptr = kernel.dstDataPtr();
+    if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0)
+    {
+      // the pointer is not aligend-on scalar, so alignment is not possible
+      return dense_assignment_loop<Kernel,DefaultTraversal,NoUnrolling>::run(kernel);
+    }
+    const Index packetAlignedMask = packetSize - 1;
+    const Index innerSize = kernel.innerSize();
+    const Index outerSize = kernel.outerSize();
+    const Index alignedStep = alignable ? (packetSize - kernel.outerStride() % packetSize) & packetAlignedMask : 0;
+    Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned<requestedAlignment>(dst_ptr, innerSize);
+
+    for(Index outer = 0; outer < outerSize; ++outer)
+    {
+      const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask);
+      // do the non-vectorizable part of the assignment
+      for(Index inner = 0; inner<alignedStart ; ++inner)
+        kernel.assignCoeffByOuterInner(outer, inner);
+
+      // do the vectorizable part of the assignment
+      for(Index inner = alignedStart; inner<alignedEnd; inner+=packetSize)
+        kernel.template assignPacketByOuterInner<dstAlignment, Unaligned, PacketType>(outer, inner);
+
+      // do the non-vectorizable part of the assignment
+      for(Index inner = alignedEnd; inner<innerSize ; ++inner)
+        kernel.assignCoeffByOuterInner(outer, inner);
+
+      alignedStart = numext::mini((alignedStart+alignedStep)%packetSize, innerSize);
+    }
+  }
+};
+
+#if EIGEN_UNALIGNED_VECTORIZE
+template<typename Kernel>
+struct dense_assignment_loop<Kernel, SliceVectorizedTraversal, InnerUnrolling>
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel)
+  {
+    typedef typename Kernel::DstEvaluatorType::XprType DstXprType;
+    typedef typename Kernel::PacketType PacketType;
+
+    enum { size = DstXprType::InnerSizeAtCompileTime,
+           packetSize =unpacket_traits<PacketType>::size,
+           vectorizableSize = (size/packetSize)*packetSize };
+
+    for(Index outer = 0; outer < kernel.outerSize(); ++outer)
+    {
+      copy_using_evaluator_innervec_InnerUnrolling<Kernel, 0, vectorizableSize, 0, 0>::run(kernel, outer);
+      copy_using_evaluator_DefaultTraversal_InnerUnrolling<Kernel, vectorizableSize, size>::run(kernel, outer);
+    }
+  }
+};
+#endif
+
+
+/***************************************************************************
+* Part 4 : Generic dense assignment kernel
+***************************************************************************/
+
+// This class generalize the assignment of a coefficient (or packet) from one dense evaluator
+// to another dense writable evaluator.
+// It is parametrized by the two evaluators, and the actual assignment functor.
+// This abstraction level permits to keep the evaluation loops as simple and as generic as possible.
+// One can customize the assignment using this generic dense_assignment_kernel with different
+// functors, or by completely overloading it, by-passing a functor.
+template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
+class generic_dense_assignment_kernel
+{
+protected:
+  typedef typename DstEvaluatorTypeT::XprType DstXprType;
+  typedef typename SrcEvaluatorTypeT::XprType SrcXprType;
+public:
+  
+  typedef DstEvaluatorTypeT DstEvaluatorType;
+  typedef SrcEvaluatorTypeT SrcEvaluatorType;
+  typedef typename DstEvaluatorType::Scalar Scalar;
+  typedef copy_using_evaluator_traits<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor> AssignmentTraits;
+  typedef typename AssignmentTraits::PacketType PacketType;
+  
+  
+  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
+    : m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr)
+  {
+    #ifdef EIGEN_DEBUG_ASSIGN
+    AssignmentTraits::debug();
+    #endif
+  }
+  
+  EIGEN_DEVICE_FUNC Index size() const        { return m_dstExpr.size(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const   { return m_dstExpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const   { return m_dstExpr.outerSize(); }
+  EIGEN_DEVICE_FUNC Index rows() const        { return m_dstExpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const        { return m_dstExpr.cols(); }
+  EIGEN_DEVICE_FUNC Index outerStride() const { return m_dstExpr.outerStride(); }
+  
+  EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() { return m_dst; }
+  EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const { return m_src; }
+  
+  /// Assign src(row,col) to dst(row,col) through the assignment functor.
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col)
+  {
+    m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col));
+  }
+  
+  /// \sa assignCoeff(Index,Index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index index)
+  {
+    m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index));
+  }
+  
+  /// \sa assignCoeff(Index,Index)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeffByOuterInner(Index outer, Index inner)
+  {
+    Index row = rowIndexByOuterInner(outer, inner); 
+    Index col = colIndexByOuterInner(outer, inner); 
+    assignCoeff(row, col);
+  }
+  
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index row, Index col)
+  {
+    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(row,col), m_src.template packet<LoadMode,PacketType>(row,col));
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index index)
+  {
+    m_functor.template assignPacket<StoreMode>(&m_dst.coeffRef(index), m_src.template packet<LoadMode,PacketType>(index));
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner)
+  {
+    Index row = rowIndexByOuterInner(outer, inner); 
+    Index col = colIndexByOuterInner(outer, inner);
+    assignPacket<StoreMode,LoadMode,PacketType>(row, col);
+  }
+  
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner)
+  {
+    typedef typename DstEvaluatorType::ExpressionTraits Traits;
+    return int(Traits::RowsAtCompileTime) == 1 ? 0
+      : int(Traits::ColsAtCompileTime) == 1 ? inner
+      : int(DstEvaluatorType::Flags)&RowMajorBit ? outer
+      : inner;
+  }
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner)
+  {
+    typedef typename DstEvaluatorType::ExpressionTraits Traits;
+    return int(Traits::ColsAtCompileTime) == 1 ? 0
+      : int(Traits::RowsAtCompileTime) == 1 ? inner
+      : int(DstEvaluatorType::Flags)&RowMajorBit ? inner
+      : outer;
+  }
+
+  EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const
+  {
+    return m_dstExpr.data();
+  }
+  
+protected:
+  DstEvaluatorType& m_dst;
+  const SrcEvaluatorType& m_src;
+  const Functor &m_functor;
+  // TODO find a way to avoid the needs of the original expression
+  DstXprType& m_dstExpr;
+};
+
+/***************************************************************************
+* Part 5 : Entry point for dense rectangular assignment
+***************************************************************************/
+
+template<typename DstXprType,typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const Functor &/*func*/)
+{
+  EIGEN_ONLY_USED_FOR_DEBUG(dst);
+  EIGEN_ONLY_USED_FOR_DEBUG(src);
+  eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+}
+
+template<typename DstXprType,typename SrcXprType, typename T1, typename T2>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const internal::assign_op<T1,T2> &/*func*/)
+{
+  Index dstRows = src.rows();
+  Index dstCols = src.cols();
+  if(((dst.rows()!=dstRows) || (dst.cols()!=dstCols)))
+    dst.resize(dstRows, dstCols);
+  eigen_assert(dst.rows() == dstRows && dst.cols() == dstCols);
+}
+
+template<typename DstXprType, typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
+{
+  typedef evaluator<DstXprType> DstEvaluatorType;
+  typedef evaluator<SrcXprType> SrcEvaluatorType;
+
+  SrcEvaluatorType srcEvaluator(src);
+
+  // NOTE To properly handle A = (A*A.transpose())/s with A rectangular,
+  // we need to resize the destination after the source evaluator has been created.
+  resize_if_allowed(dst, src, func);
+
+  DstEvaluatorType dstEvaluator(dst);
+    
+  typedef generic_dense_assignment_kernel<DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
+  Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
+
+  dense_assignment_loop<Kernel>::run(kernel);
+}
+
+template<typename DstXprType, typename SrcXprType>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src)
+{
+  call_dense_assignment_loop(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
+}
+
+/***************************************************************************
+* Part 6 : Generic assignment
+***************************************************************************/
+
+// Based on the respective shapes of the destination and source,
+// the class AssignmentKind determine the kind of assignment mechanism.
+// AssignmentKind must define a Kind typedef.
+template<typename DstShape, typename SrcShape> struct AssignmentKind;
+
+// Assignement kind defined in this file:
+struct Dense2Dense {};
+struct EigenBase2EigenBase {};
+
+template<typename,typename> struct AssignmentKind { typedef EigenBase2EigenBase Kind; };
+template<> struct AssignmentKind<DenseShape,DenseShape> { typedef Dense2Dense Kind; };
+    
+// This is the main assignment class
+template< typename DstXprType, typename SrcXprType, typename Functor,
+          typename Kind = typename AssignmentKind< typename evaluator_traits<DstXprType>::Shape , typename evaluator_traits<SrcXprType>::Shape >::Kind,
+          typename EnableIf = void>
+struct Assignment;
+
+
+// The only purpose of this call_assignment() function is to deal with noalias() / "assume-aliasing" and automatic transposition.
+// Indeed, I (Gael) think that this concept of "assume-aliasing" was a mistake, and it makes thing quite complicated.
+// So this intermediate function removes everything related to "assume-aliasing" such that Assignment
+// does not has to bother about these annoying details.
+
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(const Dst& dst, const Src& src)
+{
+  call_assignment(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+                     
+// Deal with "assume-aliasing"
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+{
+  typename plain_matrix_type<Src>::type tmp(src);
+  call_assignment_no_alias(dst, tmp, func);
+}
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if<!evaluator_assume_aliasing<Src>::value, void*>::type = 0)
+{
+  call_assignment_no_alias(dst, src, func);
+}
+
+// by-pass "assume-aliasing"
+// When there is no aliasing, we require that 'dst' has been properly resized
+template<typename Dst, template <typename> class StorageBase, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment(NoAlias<Dst,StorageBase>& dst, const Src& src, const Func& func)
+{
+  call_assignment_no_alias(dst.expression(), src, func);
+}
+
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func)
+{
+  enum {
+    NeedToTranspose = (    (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1)
+                        || (int(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1)
+                      ) && int(Dst::SizeAtCompileTime) != 1
+  };
+
+  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst>::type ActualDstTypeCleaned;
+  typedef typename internal::conditional<NeedToTranspose, Transpose<Dst>, Dst&>::type ActualDstType;
+  ActualDstType actualDst(dst);
+  
+  // TODO check whether this is the right place to perform these checks:
+  EIGEN_STATIC_ASSERT_LVALUE(Dst)
+  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(ActualDstTypeCleaned,Src)
+  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar);
+  
+  Assignment<ActualDstTypeCleaned,Src,Func>::run(actualDst, src, func);
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias(Dst& dst, const Src& src)
+{
+  call_assignment_no_alias(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+
+template<typename Dst, typename Src, typename Func>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func)
+{
+  // TODO check whether this is the right place to perform these checks:
+  EIGEN_STATIC_ASSERT_LVALUE(Dst)
+  EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Dst,Src)
+  EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar);
+
+  Assignment<Dst,Src,Func>::run(dst, src, func);
+}
+template<typename Dst, typename Src>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src)
+{
+  call_assignment_no_alias_no_transpose(dst, src, internal::assign_op<typename Dst::Scalar,typename Src::Scalar>());
+}
+
+// forward declaration
+template<typename Dst, typename Src> void check_for_aliasing(const Dst &dst, const Src &src);
+
+// Generic Dense to Dense assignment
+// Note that the last template argument "Weak" is needed to make it possible to perform
+// both partial specialization+SFINAE without ambiguous specialization
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
+struct Assignment<DstXprType, SrcXprType, Functor, Dense2Dense, Weak>
+{
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+#ifndef EIGEN_NO_DEBUG
+    internal::check_for_aliasing(dst, src);
+#endif
+    
+    call_dense_assignment_loop(dst, src, func);
+  }
+};
+
+// Generic assignment through evalTo.
+// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism.
+// Note that the last template argument "Weak" is needed to make it possible to perform
+// both partial specialization+SFINAE without ambiguous specialization
+template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak>
+struct Assignment<DstXprType, SrcXprType, Functor, EigenBase2EigenBase, Weak>
+{
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.evalTo(dst);
+  }
+
+  // NOTE The following two functions are templated to avoid their instanciation if not needed
+  //      This is needed because some expressions supports evalTo only and/or have 'void' as scalar type.
+  template<typename SrcScalarType>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.addTo(dst);
+  }
+
+  template<typename SrcScalarType>
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,SrcScalarType> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    src.subTo(dst);
+  }
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_ASSIGN_EVALUATOR_H
diff --git a/Eigen/src/Core/Assign_MKL.h b/Eigen/src/Core/Assign_MKL.h
old mode 100644
new mode 100755
index 7772951..6866095
--- a/Eigen/src/Core/Assign_MKL.h
+++ b/Eigen/src/Core/Assign_MKL.h
@@ -1,6 +1,7 @@
 /*
  Copyright (c) 2011, Intel Corporation. All rights reserved.
-
+ Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+ 
  Redistribution and use in source and binary forms, with or without modification,
  are permitted provided that the following conditions are met:
 
@@ -37,17 +38,13 @@
 
 namespace internal {
 
-template<typename Op> struct vml_call
-{ enum { IsSupported = 0 }; };
-
-template<typename Dst, typename Src, typename UnaryOp>
+template<typename Dst, typename Src>
 class vml_assign_traits
 {
   private:
     enum {
       DstHasDirectAccess = Dst::Flags & DirectAccessBit,
       SrcHasDirectAccess = Src::Flags & DirectAccessBit,
-
       StorageOrdersAgree = (int(Dst::IsRowMajor) == int(Src::IsRowMajor)),
       InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime)
                 : int(Dst::Flags)&RowMajorBit ? int(Dst::ColsAtCompileTime)
@@ -57,165 +54,122 @@
                     : int(Dst::MaxRowsAtCompileTime),
       MaxSizeAtCompileTime = Dst::SizeAtCompileTime,
 
-      MightEnableVml =  vml_call<UnaryOp>::IsSupported && StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess
-                     && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
+      MightEnableVml = StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1,
       MightLinearize = MightEnableVml && (int(Dst::Flags) & int(Src::Flags) & LinearAccessBit),
       VmlSize = MightLinearize ? MaxSizeAtCompileTime : InnerMaxSize,
-      LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD,
-      MayEnableVml = MightEnableVml && LargeEnough,
-      MayLinearize = MayEnableVml && MightLinearize
+      LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD
     };
   public:
     enum {
-      Traversal = MayLinearize ? LinearVectorizedTraversal
-                : MayEnableVml ? InnerVectorizedTraversal
-                : DefaultTraversal
+      EnableVml = MightEnableVml && LargeEnough,
+      Traversal = MightLinearize ? LinearTraversal : DefaultTraversal
     };
 };
 
-template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling,
-         int VmlTraversal = vml_assign_traits<Derived1, Derived2, UnaryOp>::Traversal >
-struct vml_assign_impl
-  : assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>
-{
-};
-
-template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
-struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, InnerVectorizedTraversal>
-{
-  typedef typename Derived1::Scalar Scalar;
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
-  {
-    // in case we want to (or have to) skip VML at runtime we can call:
-    // assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
-    const Index innerSize = dst.innerSize();
-    const Index outerSize = dst.outerSize();
-    for(Index outer = 0; outer < outerSize; ++outer) {
-      const Scalar *src_ptr = src.IsRowMajor ?  &(src.nestedExpression().coeffRef(outer,0)) :
-                                                &(src.nestedExpression().coeffRef(0, outer));
-      Scalar *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));
-      vml_call<UnaryOp>::run(src.functor(), innerSize, src_ptr, dst_ptr );
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, typename UnaryOp, int Traversal, int Unrolling>
-struct vml_assign_impl<Derived1, Derived2, UnaryOp, Traversal, Unrolling, LinearVectorizedTraversal>
-{
-  static inline void run(Derived1& dst, const CwiseUnaryOp<UnaryOp, Derived2>& src)
-  {
-    // in case we want to (or have to) skip VML at runtime we can call:
-    // assign_impl<Derived1,Eigen::CwiseUnaryOp<UnaryOp, Derived2>,Traversal,Unrolling,BuiltIn>::run(dst,src);
-    vml_call<UnaryOp>::run(src.functor(), dst.size(), src.nestedExpression().data(), dst.data() );
-  }
-};
-
-// Macroses
-
-#define EIGEN_MKL_VML_SPECIALIZE_ASSIGN(TRAVERSAL,UNROLLING) \
-  template<typename Derived1, typename Derived2, typename UnaryOp> \
-  struct assign_impl<Derived1, Eigen::CwiseUnaryOp<UnaryOp, Derived2>, TRAVERSAL, UNROLLING, Specialized>  {  \
-    static inline void run(Derived1 &dst, const Eigen::CwiseUnaryOp<UnaryOp, Derived2> &src) { \
-      vml_assign_impl<Derived1,Derived2,UnaryOp,TRAVERSAL,UNROLLING>::run(dst, src); \
-    } \
-  };
-
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,InnerUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,InnerUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,CompleteUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,NoUnrolling)
-EIGEN_MKL_VML_SPECIALIZE_ASSIGN(SliceVectorizedTraversal,NoUnrolling)
-
-
+#define EIGEN_PP_EXPAND(ARG) ARG
 #if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1)
-#define  EIGEN_MKL_VML_MODE VML_HA
+#define EIGEN_VMLMODE_EXPAND_LA , VML_HA
 #else
-#define  EIGEN_MKL_VML_MODE VML_LA
+#define EIGEN_VMLMODE_EXPAND_LA , VML_LA
 #endif
 
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)     \
-  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
-    enum { IsSupported = 1 };                                                    \
-    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/,        \
-                            int size, const EIGENTYPE* src, EIGENTYPE* dst) {    \
-      VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst);                           \
-    }                                                                            \
+#define EIGEN_VMLMODE_EXPAND__ 
+
+#define EIGEN_VMLMODE_PREFIX_LA vm
+#define EIGEN_VMLMODE_PREFIX__  v
+#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_,VMLMODE)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
+  template< typename DstXprType, typename SrcXprNested>                                                                         \
+  struct Assignment<DstXprType, CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested>, assign_op<EIGENTYPE,EIGENTYPE>,   \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {              \
+    typedef CwiseUnaryOp<scalar_##EIGENOP##_op<EIGENTYPE>, SrcXprNested> SrcXprType;                                            \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                       \
+      resize_if_allowed(dst, src, func);                                                                                        \
+      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                       \
+      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal) {                                              \
+        VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(),                                                        \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                           \
+      } else {                                                                                                                  \
+        const Index outerSize = dst.outerSize();                                                                                \
+        for(Index outer = 0; outer < outerSize; ++outer) {                                                                      \
+          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) :                             \
+                                                      &(src.nestedExpression().coeffRef(0, outer));                             \
+          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                           \
+          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr,                                                                      \
+                (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                             \
+        }                                                                                                                       \
+      }                                                                                                                         \
+    }                                                                                                                           \
+  };                                                                                                                            \
+
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),s##VMLOP), float, float, VMLMODE)           \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),d##VMLOP), double, double, VMLMODE)
+
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)                                                         \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),c##VMLOP), scomplex, MKL_Complex8, VMLMODE) \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),z##VMLOP), dcomplex, MKL_Complex16, VMLMODE)
+  
+#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP, VMLMODE)                                                              \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE)                                                               \
+  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE)
+
+  
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sin,   Sin,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(asin,  Asin,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sinh,  Sinh,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cos,   Cos,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(acos,  Acos,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cosh,  Cosh,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tan,   Tan,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(atan,  Atan,  LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tanh,  Tanh,  LA)
+// EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,   Abs,    _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(exp,   Exp,   LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log,   Ln,    LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log10, Log10, LA)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sqrt,  Sqrt,  _)
+
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr,   _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(arg, Arg,      _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(round, Round,  _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor,  _)
+EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil,  Ceil,   _)
+
+#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE)                                           \
+  template< typename DstXprType, typename SrcXprNested, typename Plain>                                                       \
+  struct Assignment<DstXprType, CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                       \
+                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> >, assign_op<EIGENTYPE,EIGENTYPE>,    \
+                   Dense2Dense, typename enable_if<vml_assign_traits<DstXprType,SrcXprNested>::EnableVml>::type> {            \
+    typedef CwiseBinaryOp<scalar_##EIGENOP##_op<EIGENTYPE,EIGENTYPE>, SrcXprNested,                                           \
+                    const CwiseNullaryOp<internal::scalar_constant_op<EIGENTYPE>,Plain> > SrcXprType;                         \
+    static void run(DstXprType &dst, const SrcXprType &src, const assign_op<EIGENTYPE,EIGENTYPE> &func) {                     \
+      resize_if_allowed(dst, src, func);                                                                                      \
+      eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());                                                     \
+      VMLTYPE exponent = reinterpret_cast<const VMLTYPE&>(src.rhs().functor().m_other);                                       \
+      if(vml_assign_traits<DstXprType,SrcXprNested>::Traversal==LinearTraversal)                                              \
+      {                                                                                                                       \
+        VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent,                                                        \
+              (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE) );                                         \
+      } else {                                                                                                                \
+        const Index outerSize = dst.outerSize();                                                                              \
+        for(Index outer = 0; outer < outerSize; ++outer) {                                                                    \
+          const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) :                                        \
+                                                      &(src.lhs().coeffRef(0, outer));                                        \
+          EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer));                         \
+          VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent,                                                          \
+                 (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_##VMLMODE));                                          \
+        }                                                                                                                     \
+      }                                                                                                                       \
+    }                                                                                                                         \
   };
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)  \
-  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
-    enum { IsSupported = 1 };                                                    \
-    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& /*func*/,        \
-                            int size, const EIGENTYPE* src, EIGENTYPE* dst) {    \
-      MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE;                                    \
-      VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst, vmlMode);                  \
-    }                                                                            \
-  };
-
-#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE)       \
-  template<> struct vml_call< scalar_##EIGENOP##_op<EIGENTYPE> > {               \
-    enum { IsSupported = 1 };                                                    \
-    static inline void run( const scalar_##EIGENOP##_op<EIGENTYPE>& func,        \
-                          int size, const EIGENTYPE* src, EIGENTYPE* dst) {      \
-      EIGENTYPE exponent = func.m_exponent;                                      \
-      MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE;                                    \
-      VMLOP(&size, (const VMLTYPE*)src, (const VMLTYPE*)&exponent,               \
-                        (VMLTYPE*)dst, &vmlMode);                                \
-    }                                                                            \
-  };
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP)                   \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vs##VMLOP, float, float)             \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vd##VMLOP, double, double)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP)                \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vc##VMLOP, scomplex, MKL_Complex8)   \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vz##VMLOP, dcomplex, MKL_Complex16)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP)                        \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP)                         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP)
-
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP)                \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vms##VMLOP, float, float)         \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmd##VMLOP, double, double)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP)             \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmc##VMLOP, scomplex, MKL_Complex8)  \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmz##VMLOP, dcomplex, MKL_Complex16)
-
-#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(EIGENOP, VMLOP)                     \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP)                      \
-  EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP)
-
-
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sin,  Sin)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(asin, Asin)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(cos,  Cos)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(acos, Acos)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(tan,  Tan)
-//EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs,  Abs)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(exp,  Exp)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(log,  Ln)
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sqrt, Sqrt)
-
-EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr)
-
-// The vm*powx functions are not avaibale in the windows version of MKL.
-#ifndef _WIN32
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmspowx_, float, float)
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdpowx_, double, double)
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcpowx_, scomplex, MKL_Complex8)
-EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzpowx_, dcomplex, MKL_Complex16)
-#endif
+  
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmsPowx, float,    float,         LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdPowx, double,   double,        LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcPowx, scomplex, MKL_Complex8,  LA)
+EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzPowx, dcomplex, MKL_Complex16, LA)
 
 } // end namespace internal
 
diff --git a/Eigen/src/Core/BandMatrix.h b/Eigen/src/Core/BandMatrix.h
index ffd7fe8..4978c91 100644
--- a/Eigen/src/Core/BandMatrix.h
+++ b/Eigen/src/Core/BandMatrix.h
@@ -32,7 +32,7 @@
     };
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
-    typedef typename DenseMatrixType::Index Index;
+    typedef typename DenseMatrixType::StorageIndex StorageIndex;
     typedef typename internal::traits<Derived>::CoefficientsType CoefficientsType;
     typedef EigenBase<Derived> Base;
 
@@ -161,15 +161,15 @@
   *
   * \brief Represents a rectangular matrix with a banded storage
   *
-  * \param _Scalar Numeric type, i.e. float, double, int
-  * \param Rows Number of rows, or \b Dynamic
-  * \param Cols Number of columns, or \b Dynamic
-  * \param Supers Number of super diagonal
-  * \param Subs Number of sub diagonal
-  * \param _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint
-  *                 The former controls \ref TopicStorageOrders "storage order", and defaults to
-  *                 column-major. The latter controls whether the matrix represents a selfadjoint 
-  *                 matrix in which case either Supers of Subs have to be null.
+  * \tparam _Scalar Numeric type, i.e. float, double, int
+  * \tparam _Rows Number of rows, or \b Dynamic
+  * \tparam _Cols Number of columns, or \b Dynamic
+  * \tparam _Supers Number of super diagonal
+  * \tparam _Subs Number of sub diagonal
+  * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint
+  *                  The former controls \ref TopicStorageOrders "storage order", and defaults to
+  *                  column-major. The latter controls whether the matrix represents a selfadjoint
+  *                  matrix in which case either Supers of Subs have to be null.
   *
   * \sa class TridiagonalMatrix
   */
@@ -179,7 +179,7 @@
 {
   typedef _Scalar Scalar;
   typedef Dense StorageKind;
-  typedef DenseIndex Index;
+  typedef Eigen::Index StorageIndex;
   enum {
     CoeffReadCost = NumTraits<Scalar>::ReadCost,
     RowsAtCompileTime = _Rows,
@@ -201,10 +201,10 @@
   public:
 
     typedef typename internal::traits<BandMatrix>::Scalar Scalar;
-    typedef typename internal::traits<BandMatrix>::Index Index;
+    typedef typename internal::traits<BandMatrix>::StorageIndex StorageIndex;
     typedef typename internal::traits<BandMatrix>::CoefficientsType CoefficientsType;
 
-    inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
+    explicit inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs)
       : m_coeffs(1+supers+subs,cols),
         m_rows(rows), m_supers(supers), m_subs(subs)
     {
@@ -241,7 +241,7 @@
 {
   typedef typename _CoefficientsType::Scalar Scalar;
   typedef typename _CoefficientsType::StorageKind StorageKind;
-  typedef typename _CoefficientsType::Index Index;
+  typedef typename _CoefficientsType::StorageIndex StorageIndex;
   enum {
     CoeffReadCost = internal::traits<_CoefficientsType>::CoeffReadCost,
     RowsAtCompileTime = _Rows,
@@ -264,9 +264,9 @@
 
     typedef typename internal::traits<BandMatrixWrapper>::Scalar Scalar;
     typedef typename internal::traits<BandMatrixWrapper>::CoefficientsType CoefficientsType;
-    typedef typename internal::traits<BandMatrixWrapper>::Index Index;
+    typedef typename internal::traits<BandMatrixWrapper>::StorageIndex StorageIndex;
 
-    inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
+    explicit inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs)
       : m_coeffs(coeffs),
         m_rows(rows), m_supers(supers), m_subs(subs)
     {
@@ -302,9 +302,9 @@
   *
   * \brief Represents a tridiagonal matrix with a compact banded storage
   *
-  * \param _Scalar Numeric type, i.e. float, double, int
-  * \param Size Number of rows and cols, or \b Dynamic
-  * \param _Options Can be 0 or \b SelfAdjoint
+  * \tparam Scalar Numeric type, i.e. float, double, int
+  * \tparam Size Number of rows and cols, or \b Dynamic
+  * \tparam Options Can be 0 or \b SelfAdjoint
   *
   * \sa class BandMatrix
   */
@@ -312,9 +312,9 @@
 class TridiagonalMatrix : public BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor>
 {
     typedef BandMatrix<Scalar,Size,Size,Options&SelfAdjoint?0:1,1,Options|RowMajor> Base;
-    typedef typename Base::Index Index;
+    typedef typename Base::StorageIndex StorageIndex;
   public:
-    TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
+    explicit TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {}
 
     inline typename Base::template DiagonalIntReturnType<1>::Type super()
     { return Base::template diagonal<1>(); }
@@ -327,6 +327,25 @@
   protected:
 };
 
+
+struct BandShape {};
+
+template<typename _Scalar, int _Rows, int _Cols, int _Supers, int _Subs, int _Options>
+struct evaluator_traits<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
+  : public evaluator_traits_base<BandMatrix<_Scalar,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef BandShape Shape;
+};
+
+template<typename _CoefficientsType,int _Rows, int _Cols, int _Supers, int _Subs,int _Options>
+struct evaluator_traits<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+  : public evaluator_traits_base<BandMatrixWrapper<_CoefficientsType,_Rows,_Cols,_Supers,_Subs,_Options> >
+{
+  typedef BandShape Shape;
+};
+
+template<> struct AssignmentKind<DenseShape,BandShape> { typedef EigenBase2EigenBase Kind; };
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/Block.h b/Eigen/src/Core/Block.h
index 8278944..11de45c 100644
--- a/Eigen/src/Core/Block.h
+++ b/Eigen/src/Core/Block.h
@@ -13,14 +13,70 @@
 
 namespace Eigen { 
 
+namespace internal {
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
+struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
+{
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename traits<XprType>::StorageKind StorageKind;
+  typedef typename traits<XprType>::XprKind XprKind;
+  typedef typename ref_selector<XprType>::type XprTypeNested;
+  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
+  enum{
+    MatrixRows = traits<XprType>::RowsAtCompileTime,
+    MatrixCols = traits<XprType>::ColsAtCompileTime,
+    RowsAtCompileTime = MatrixRows == 0 ? 0 : BlockRows,
+    ColsAtCompileTime = MatrixCols == 0 ? 0 : BlockCols,
+    MaxRowsAtCompileTime = BlockRows==0 ? 0
+                         : RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime)
+                         : int(traits<XprType>::MaxRowsAtCompileTime),
+    MaxColsAtCompileTime = BlockCols==0 ? 0
+                         : ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
+                         : int(traits<XprType>::MaxColsAtCompileTime),
+
+    XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
+    IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
+               : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
+               : XprTypeIsRowMajor,
+    HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
+    InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
+    InnerStrideAtCompileTime = HasSameStorageOrderAsXprType
+                             ? int(inner_stride_at_compile_time<XprType>::ret)
+                             : int(outer_stride_at_compile_time<XprType>::ret),
+    OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
+                             ? int(outer_stride_at_compile_time<XprType>::ret)
+                             : int(inner_stride_at_compile_time<XprType>::ret),
+
+    // FIXME, this traits is rather specialized for dense object and it needs to be cleaned further
+    FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
+    FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
+    Flags = (traits<XprType>::Flags & (DirectAccessBit | (InnerPanel?CompressedAccessBit:0))) | FlagsLvalueBit | FlagsRowMajorBit,
+    // FIXME DirectAccessBit should not be handled by expressions
+    // 
+    // Alignment is needed by MapBase's assertions
+    // We can sefely set it to false here. Internal alignment errors will be detected by an eigen_internal_assert in the respective evaluator
+    Alignment = 0
+  };
+};
+
+template<typename XprType, int BlockRows=Dynamic, int BlockCols=Dynamic, bool InnerPanel = false,
+         bool HasDirectAccess = internal::has_direct_access<XprType>::ret> class BlockImpl_dense;
+         
+} // end namespace internal
+
+template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, typename StorageKind> class BlockImpl;
+
 /** \class Block
   * \ingroup Core_Module
   *
   * \brief Expression of a fixed-size or dynamic-size block
   *
-  * \param XprType the type of the expression in which we are taking a block
-  * \param BlockRows the number of rows of the block we are taking at compile time (optional)
-  * \param BlockCols the number of columns of the block we are taking at compile time (optional)
+  * \tparam XprType the type of the expression in which we are taking a block
+  * \tparam BlockRows the number of rows of the block we are taking at compile time (optional)
+  * \tparam BlockCols the number of columns of the block we are taking at compile time (optional)
+  * \tparam InnerPanel is true, if the block maps to a set of rows of a row major matrix or
+  *         to set of columns of a column major matrix (optional). The parameter allows to determine
+  *         at compile time whether aligned access is possible on the block expression.
   *
   * This class represents an expression of either a fixed-size or dynamic-size block. It is the return
   * type of DenseBase::block(Index,Index,Index,Index) and DenseBase::block<int,int>(Index,Index) and
@@ -44,62 +100,6 @@
   *
   * \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock
   */
-
-namespace internal {
-template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
-struct traits<Block<XprType, BlockRows, BlockCols, InnerPanel> > : traits<XprType>
-{
-  typedef typename traits<XprType>::Scalar Scalar;
-  typedef typename traits<XprType>::StorageKind StorageKind;
-  typedef typename traits<XprType>::XprKind XprKind;
-  typedef typename nested<XprType>::type XprTypeNested;
-  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
-  enum{
-    MatrixRows = traits<XprType>::RowsAtCompileTime,
-    MatrixCols = traits<XprType>::ColsAtCompileTime,
-    RowsAtCompileTime = MatrixRows == 0 ? 0 : BlockRows,
-    ColsAtCompileTime = MatrixCols == 0 ? 0 : BlockCols,
-    MaxRowsAtCompileTime = BlockRows==0 ? 0
-                         : RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime)
-                         : int(traits<XprType>::MaxRowsAtCompileTime),
-    MaxColsAtCompileTime = BlockCols==0 ? 0
-                         : ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime)
-                         : int(traits<XprType>::MaxColsAtCompileTime),
-    XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0,
-    IsDense = is_same<StorageKind,Dense>::value,
-    IsRowMajor = (IsDense&&MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
-               : (IsDense&&MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
-               : XprTypeIsRowMajor,
-    HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor),
-    InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
-    InnerStrideAtCompileTime = HasSameStorageOrderAsXprType
-                             ? int(inner_stride_at_compile_time<XprType>::ret)
-                             : int(outer_stride_at_compile_time<XprType>::ret),
-    OuterStrideAtCompileTime = HasSameStorageOrderAsXprType
-                             ? int(outer_stride_at_compile_time<XprType>::ret)
-                             : int(inner_stride_at_compile_time<XprType>::ret),
-    MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits<Scalar>::size) == 0)
-                       && (InnerStrideAtCompileTime == 1)
-                        ? PacketAccessBit : 0,
-    MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0,
-    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (traits<XprType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,
-    FlagsLvalueBit = is_lvalue<XprType>::value ? LvalueBit : 0,
-    FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0,
-    Flags0 = traits<XprType>::Flags & ( (HereditaryBits & ~RowMajorBit) |
-                                        DirectAccessBit |
-                                        MaskPacketAccessBit |
-                                        MaskAlignedBit),
-    Flags = Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit
-  };
-};
-
-template<typename XprType, int BlockRows=Dynamic, int BlockCols=Dynamic, bool InnerPanel = false,
-         bool HasDirectAccess = internal::has_direct_access<XprType>::ret> class BlockImpl_dense;
-         
-} // end namespace internal
-
-template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel, typename StorageKind> class BlockImpl;
-
 template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel> class Block
   : public BlockImpl<XprType, BlockRows, BlockCols, InnerPanel, typename internal::traits<XprType>::StorageKind>
 {
@@ -109,9 +109,12 @@
     typedef Impl Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(Block)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block)
+    
+    typedef typename internal::remove_all<XprType>::type NestedExpression;
   
     /** Column or Row constructor
       */
+    EIGEN_DEVICE_FUNC
     inline Block(XprType& xpr, Index i) : Impl(xpr,i)
     {
       eigen_assert( (i>=0) && (
@@ -121,25 +124,27 @@
 
     /** Fixed-size constructor
       */
-    inline Block(XprType& xpr, Index a_startRow, Index a_startCol)
-      : Impl(xpr, a_startRow, a_startCol)
+    EIGEN_DEVICE_FUNC
+    inline Block(XprType& xpr, Index startRow, Index startCol)
+      : Impl(xpr, startRow, startCol)
     {
       EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE)
-      eigen_assert(a_startRow >= 0 && BlockRows >= 1 && a_startRow + BlockRows <= xpr.rows()
-             && a_startCol >= 0 && BlockCols >= 1 && a_startCol + BlockCols <= xpr.cols());
+      eigen_assert(startRow >= 0 && BlockRows >= 0 && startRow + BlockRows <= xpr.rows()
+             && startCol >= 0 && BlockCols >= 0 && startCol + BlockCols <= xpr.cols());
     }
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline Block(XprType& xpr,
-          Index a_startRow, Index a_startCol,
+          Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : Impl(xpr, a_startRow, a_startCol, blockRows, blockCols)
+      : Impl(xpr, startRow, startCol, blockRows, blockCols)
     {
       eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows)
           && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols));
-      eigen_assert(a_startRow >= 0 && blockRows >= 0 && a_startRow  <= xpr.rows() - blockRows
-          && a_startCol >= 0 && blockCols >= 0 && a_startCol <= xpr.cols() - blockCols);
+      eigen_assert(startRow >= 0 && blockRows >= 0 && startRow  <= xpr.rows() - blockRows
+          && startCol >= 0 && blockCols >= 0 && startCol <= xpr.cols() - blockCols);
     }
 };
          
@@ -150,14 +155,15 @@
   : public internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel>
 {
     typedef internal::BlockImpl_dense<XprType, BlockRows, BlockCols, InnerPanel> Impl;
-    typedef typename XprType::Index Index;
+    typedef typename XprType::StorageIndex StorageIndex;
   public:
     typedef Impl Base;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
-    inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
-    inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol) : Impl(xpr, a_startRow, a_startCol) {}
-    inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol, Index blockRows, Index blockCols)
-      : Impl(xpr, a_startRow, a_startCol, blockRows, blockCols) {}
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {}
+    EIGEN_DEVICE_FUNC inline BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {}
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : Impl(xpr, startRow, startCol, blockRows, blockCols) {}
 };
 
 namespace internal {
@@ -167,16 +173,18 @@
   : public internal::dense_xpr_base<Block<XprType, BlockRows, BlockCols, InnerPanel> >::type
 {
     typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
   public:
 
     typedef typename internal::dense_xpr_base<BlockType>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(BlockType)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense)
 
-    class InnerIterator;
+    // class InnerIterator; // FIXME apparently never used
 
     /** Column or Row constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, Index i)
       : m_xpr(xpr),
         // It is a row if and only if BlockRows==1 and BlockCols==XprType::ColsAtCompileTime,
@@ -191,75 +199,76 @@
 
     /** Fixed-size constructor
       */
-    inline BlockImpl_dense(XprType& xpr, Index a_startRow, Index a_startCol)
-      : m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
+    EIGEN_DEVICE_FUNC
+    inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
+      : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
                     m_blockRows(BlockRows), m_blockCols(BlockCols)
     {}
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr,
-          Index a_startRow, Index a_startCol,
+          Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : m_xpr(xpr), m_startRow(a_startRow), m_startCol(a_startCol),
+      : m_xpr(xpr), m_startRow(startRow), m_startCol(startCol),
                     m_blockRows(blockRows), m_blockCols(blockCols)
     {}
 
-    inline Index rows() const { return m_blockRows.value(); }
-    inline Index cols() const { return m_blockCols.value(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_blockRows.value(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_blockCols.value(); }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index rowId, Index colId)
     {
       EIGEN_STATIC_ASSERT_LVALUE(XprType)
-      return m_xpr.const_cast_derived()
-               .coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
-      return m_xpr.derived()
-               .coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.derived().coeffRef(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index rowId, Index colId) const
     {
       return m_xpr.coeff(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index index)
     {
       EIGEN_STATIC_ASSERT_LVALUE(XprType)
-      return m_xpr.const_cast_derived()
-             .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                       m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_xpr.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
-      return m_xpr.const_cast_derived()
-             .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                       m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_xpr.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
+    EIGEN_DEVICE_FUNC
     inline const CoeffReturnType coeff(Index index) const
     {
-      return m_xpr
-             .coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                    m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_xpr.coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                         m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
     template<int LoadMode>
     inline PacketScalar packet(Index rowId, Index colId) const
     {
-      return m_xpr.template packet<Unaligned>
-              (rowId + m_startRow.value(), colId + m_startCol.value());
+      return m_xpr.template packet<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value());
     }
 
     template<int LoadMode>
     inline void writePacket(Index rowId, Index colId, const PacketScalar& val)
     {
-      m_xpr.const_cast_derived().template writePacket<Unaligned>
-              (rowId + m_startRow.value(), colId + m_startCol.value(), val);
+      m_xpr.template writePacket<Unaligned>(rowId + m_startRow.value(), colId + m_startCol.value(), val);
     }
 
     template<int LoadMode>
@@ -273,40 +282,46 @@
     template<int LoadMode>
     inline void writePacket(Index index, const PacketScalar& val)
     {
-      m_xpr.const_cast_derived().template writePacket<Unaligned>
+      m_xpr.template writePacket<Unaligned>
          (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
           m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0), val);
     }
 
     #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \sa MapBase::data() */
-    inline const Scalar* data() const;
-    inline Index innerStride() const;
-    inline Index outerStride() const;
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const;
+    EIGEN_DEVICE_FUNC inline Index innerStride() const;
+    EIGEN_DEVICE_FUNC inline Index outerStride() const;
     #endif
 
-    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
     { 
       return m_xpr; 
     }
+
+    EIGEN_DEVICE_FUNC
+    XprType& nestedExpression() { return m_xpr; }
       
-    Index startRow() const 
+    EIGEN_DEVICE_FUNC
+    StorageIndex startRow() const
     { 
       return m_startRow.value(); 
     }
       
-    Index startCol() const 
+    EIGEN_DEVICE_FUNC
+    StorageIndex startCol() const
     { 
       return m_startCol.value(); 
     }
 
   protected:
 
-    const typename XprType::Nested m_xpr;
-    const internal::variable_if_dynamic<Index, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
-    const internal::variable_if_dynamic<Index, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
-    const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_blockRows;
-    const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_blockCols;
+    XprTypeNested m_xpr;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
+    const internal::variable_if_dynamic<StorageIndex, RowsAtCompileTime> m_blockRows;
+    const internal::variable_if_dynamic<StorageIndex, ColsAtCompileTime> m_blockCols;
 };
 
 /** \internal Internal implementation of dense Blocks in the direct access case.*/
@@ -315,6 +330,10 @@
   : public MapBase<Block<XprType, BlockRows, BlockCols, InnerPanel> >
 {
     typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef typename internal::ref_selector<XprType>::non_const_type XprTypeNested;
+    enum {
+      XprTypeIsRowMajor = (int(traits<XprType>::Flags)&RowMajorBit) != 0
+    };
   public:
 
     typedef MapBase<BlockType> Base;
@@ -323,42 +342,52 @@
 
     /** Column or Row constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, Index i)
-      : Base(internal::const_cast_ptr(&xpr.coeffRef(
-              (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0,
-              (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)),
+      : Base(xpr.data() + i * (    ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor)) 
+                                || ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride()),
              BlockRows==1 ? 1 : xpr.rows(),
              BlockCols==1 ? 1 : xpr.cols()),
-        m_xpr(xpr)
+        m_xpr(xpr),
+        m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0),
+        m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)
     {
       init();
     }
 
     /** Fixed-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol)
-      : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol))), m_xpr(xpr)
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol)),
+        m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
     {
       init();
     }
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr,
           Index startRow, Index startCol,
           Index blockRows, Index blockCols)
-      : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol)), blockRows, blockCols),
-        m_xpr(xpr)
+      : Base(xpr.data()+xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol), blockRows, blockCols),
+        m_xpr(xpr), m_startRow(startRow), m_startCol(startCol)
     {
       init();
     }
 
-    const typename internal::remove_all<typename XprType::Nested>::type& nestedExpression() const 
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<XprTypeNested>::type& nestedExpression() const
     { 
       return m_xpr; 
     }
+
+    EIGEN_DEVICE_FUNC
+    XprType& nestedExpression() { return m_xpr; }
       
     /** \sa MapBase::innerStride() */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return internal::traits<BlockType>::HasSameStorageOrderAsXprType
@@ -367,11 +396,24 @@
     }
 
     /** \sa MapBase::outerStride() */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return m_outerStride;
     }
 
+    EIGEN_DEVICE_FUNC
+    StorageIndex startRow() const
+    {
+      return m_startRow.value();
+    }
+
+    EIGEN_DEVICE_FUNC
+    StorageIndex startCol() const
+    {
+      return m_startCol.value();
+    }
+
   #ifndef __SUNPRO_CC
   // FIXME sunstudio is not friendly with the above friend...
   // META-FIXME there is no 'friend' keyword around here. Is this obsolete?
@@ -380,6 +422,7 @@
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal used by allowAligned() */
+    EIGEN_DEVICE_FUNC
     inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols)
       : Base(data, blockRows, blockCols), m_xpr(xpr)
     {
@@ -388,6 +431,7 @@
     #endif
 
   protected:
+    EIGEN_DEVICE_FUNC
     void init()
     {
       m_outerStride = internal::traits<BlockType>::HasSameStorageOrderAsXprType
@@ -395,7 +439,9 @@
                     : m_xpr.innerStride();
     }
 
-    typename XprType::Nested m_xpr;
+    XprTypeNested m_xpr;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
+    const internal::variable_if_dynamic<StorageIndex, (XprType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
     Index m_outerStride;
 };
 
diff --git a/Eigen/src/Core/BooleanRedux.h b/Eigen/src/Core/BooleanRedux.h
index be9f48a..8409d87 100644
--- a/Eigen/src/Core/BooleanRedux.h
+++ b/Eigen/src/Core/BooleanRedux.h
@@ -17,9 +17,10 @@
 template<typename Derived, int UnrollCount>
 struct all_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
   };
 
   static inline bool run(const Derived &mat)
@@ -43,11 +44,12 @@
 template<typename Derived, int UnrollCount>
 struct any_unroller
 {
+  typedef typename Derived::ExpressionTraits Traits;
   enum {
-    col = (UnrollCount-1) / Derived::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived::RowsAtCompileTime
+    col = (UnrollCount-1) / Traits::RowsAtCompileTime,
+    row = (UnrollCount-1) % Traits::RowsAtCompileTime
   };
-
+  
   static inline bool run(const Derived &mat)
   {
     return any_unroller<Derived, UnrollCount-1>::run(mat) || mat.coeff(row, col);
@@ -78,19 +80,19 @@
 template<typename Derived>
 inline bool DenseBase<Derived>::all() const
 {
+  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && CoeffReadCost != Dynamic
-          && NumTraits<Scalar>::AddCost != Dynamic
-          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
+  Evaluator evaluator(derived());
   if(unroll)
-    return internal::all_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
+    return internal::all_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
   else
   {
     for(Index j = 0; j < cols(); ++j)
       for(Index i = 0; i < rows(); ++i)
-        if (!coeff(i, j)) return false;
+        if (!evaluator.coeff(i, j)) return false;
     return true;
   }
 }
@@ -102,19 +104,19 @@
 template<typename Derived>
 inline bool DenseBase<Derived>::any() const
 {
+  typedef internal::evaluator<Derived> Evaluator;
   enum {
     unroll = SizeAtCompileTime != Dynamic
-          && CoeffReadCost != Dynamic
-          && NumTraits<Scalar>::AddCost != Dynamic
-          && SizeAtCompileTime * (CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
+          && SizeAtCompileTime * (Evaluator::CoeffReadCost + NumTraits<Scalar>::AddCost) <= EIGEN_UNROLLING_LIMIT
   };
+  Evaluator evaluator(derived());
   if(unroll)
-    return internal::any_unroller<Derived, unroll ? int(SizeAtCompileTime) : Dynamic>::run(derived());
+    return internal::any_unroller<Evaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(evaluator);
   else
   {
     for(Index j = 0; j < cols(); ++j)
       for(Index i = 0; i < rows(); ++i)
-        if (coeff(i, j)) return true;
+        if (evaluator.coeff(i, j)) return true;
     return false;
   }
 }
@@ -124,7 +126,7 @@
   * \sa all(), any()
   */
 template<typename Derived>
-inline typename DenseBase<Derived>::Index DenseBase<Derived>::count() const
+inline Eigen::Index DenseBase<Derived>::count() const
 {
   return derived().template cast<bool>().template cast<Index>().sum();
 }
@@ -136,7 +138,11 @@
 template<typename Derived>
 inline bool DenseBase<Derived>::hasNaN() const
 {
+#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
+  return derived().array().isNaN().any();
+#else
   return !((derived().array()==derived().array()).all());
+#endif
 }
 
 /** \returns true if \c *this contains only finite numbers, i.e., no NaN and no +/-INF values.
@@ -146,7 +152,11 @@
 template<typename Derived>
 inline bool DenseBase<Derived>::allFinite() const
 {
+#if EIGEN_COMP_MSVC || (defined __FAST_MATH__)
+  return derived().array().isFinite().all();
+#else
   return !((derived()-derived()).hasNaN());
+#endif
 }
     
 } // end namespace Eigen
diff --git a/Eigen/src/Core/CMakeLists.txt b/Eigen/src/Core/CMakeLists.txt
deleted file mode 100644
index 2346fc2..0000000
--- a/Eigen/src/Core/CMakeLists.txt
+++ /dev/null
@@ -1,10 +0,0 @@
-FILE(GLOB Eigen_Core_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Core_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core COMPONENT Devel
-  )
-
-ADD_SUBDIRECTORY(products)
-ADD_SUBDIRECTORY(util)
-ADD_SUBDIRECTORY(arch)
diff --git a/Eigen/src/Core/CommaInitializer.h b/Eigen/src/Core/CommaInitializer.h
index a036d8c..d218e98 100644
--- a/Eigen/src/Core/CommaInitializer.h
+++ b/Eigen/src/Core/CommaInitializer.h
@@ -22,14 +22,14 @@
   * the return type of MatrixBase::operator<<, and most of the time this is the only
   * way it is used.
   *
-  * \sa \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
+  * \sa \blank \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished()
   */
 template<typename XprType>
 struct CommaInitializer
 {
   typedef typename XprType::Scalar Scalar;
-  typedef typename XprType::Index Index;
 
+  EIGEN_DEVICE_FUNC
   inline CommaInitializer(XprType& xpr, const Scalar& s)
     : m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1)
   {
@@ -37,6 +37,7 @@
   }
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline CommaInitializer(XprType& xpr, const DenseBase<OtherDerived>& other)
     : m_xpr(xpr), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows())
   {
@@ -46,6 +47,7 @@
   /* Copy/Move constructor which transfers ownership. This is crucial in 
    * absence of return value optimization to avoid assertions during destruction. */
   // FIXME in C++11 mode this could be replaced by a proper RValue constructor
+  EIGEN_DEVICE_FUNC
   inline CommaInitializer(const CommaInitializer& o)
   : m_xpr(o.m_xpr), m_row(o.m_row), m_col(o.m_col), m_currentBlockRows(o.m_currentBlockRows) {
     // Mark original object as finished. In absence of R-value references we need to const_cast:
@@ -55,6 +57,7 @@
   }
 
   /* inserts a scalar value in the target matrix */
+  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const Scalar& s)
   {
     if (m_col==m_xpr.cols())
@@ -74,11 +77,10 @@
 
   /* inserts a matrix expression in the target matrix */
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   CommaInitializer& operator,(const DenseBase<OtherDerived>& other)
   {
-    if(other.cols()==0 || other.rows()==0)
-      return *this;
-    if (m_col==m_xpr.cols())
+    if (m_col==m_xpr.cols() && (other.cols()!=0 || other.rows()!=m_currentBlockRows))
     {
       m_row+=m_currentBlockRows;
       m_col = 0;
@@ -86,24 +88,22 @@
       eigen_assert(m_row+m_currentBlockRows<=m_xpr.rows()
         && "Too many rows passed to comma initializer (operator<<)");
     }
-    eigen_assert(m_col<m_xpr.cols()
+    eigen_assert((m_col + other.cols() <= m_xpr.cols())
       && "Too many coefficients passed to comma initializer (operator<<)");
     eigen_assert(m_currentBlockRows==other.rows());
-    if (OtherDerived::SizeAtCompileTime != Dynamic)
-      m_xpr.template block<OtherDerived::RowsAtCompileTime != Dynamic ? OtherDerived::RowsAtCompileTime : 1,
-                              OtherDerived::ColsAtCompileTime != Dynamic ? OtherDerived::ColsAtCompileTime : 1>
-                    (m_row, m_col) = other;
-    else
-      m_xpr.block(m_row, m_col, other.rows(), other.cols()) = other;
+    m_xpr.template block<OtherDerived::RowsAtCompileTime, OtherDerived::ColsAtCompileTime>
+                    (m_row, m_col, other.rows(), other.cols()) = other;
     m_col += other.cols();
     return *this;
   }
 
+  EIGEN_DEVICE_FUNC
   inline ~CommaInitializer()
+#if defined VERIFY_RAISES_ASSERT && (!defined EIGEN_NO_ASSERTION_CHECKING) && defined EIGEN_EXCEPTIONS
+  EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception)
+#endif
   {
-    eigen_assert((m_row+m_currentBlockRows) == m_xpr.rows()
-         && m_col == m_xpr.cols()
-         && "Too few coefficients passed to comma initializer (operator<<)");
+      finished();
   }
 
   /** \returns the built matrix once all its coefficients have been set.
@@ -113,9 +113,15 @@
     * quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished());
     * \endcode
     */
-  inline XprType& finished() { return m_xpr; }
+  EIGEN_DEVICE_FUNC
+  inline XprType& finished() {
+      eigen_assert(((m_row+m_currentBlockRows) == m_xpr.rows() || m_xpr.cols() == 0)
+           && m_col == m_xpr.cols()
+           && "Too few coefficients passed to comma initializer (operator<<)");
+      return m_xpr;
+  }
 
-  XprType& m_xpr;   // target expression
+  XprType& m_xpr;           // target expression
   Index m_row;              // current row id
   Index m_col;              // current col id
   Index m_currentBlockRows; // current block height
diff --git a/Eigen/src/Core/ConditionEstimator.h b/Eigen/src/Core/ConditionEstimator.h
new file mode 100644
index 0000000..51a2e5f
--- /dev/null
+++ b/Eigen/src/Core/ConditionEstimator.h
@@ -0,0 +1,175 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Rasmus Munk Larsen (rmlarsen@google.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CONDITIONESTIMATOR_H
+#define EIGEN_CONDITIONESTIMATOR_H
+
+namespace Eigen {
+
+namespace internal {
+
+template <typename Vector, typename RealVector, bool IsComplex>
+struct rcond_compute_sign {
+  static inline Vector run(const Vector& v) {
+    const RealVector v_abs = v.cwiseAbs();
+    return (v_abs.array() == static_cast<typename Vector::RealScalar>(0))
+            .select(Vector::Ones(v.size()), v.cwiseQuotient(v_abs));
+  }
+};
+
+// Partial specialization to avoid elementwise division for real vectors.
+template <typename Vector>
+struct rcond_compute_sign<Vector, Vector, false> {
+  static inline Vector run(const Vector& v) {
+    return (v.array() < static_cast<typename Vector::RealScalar>(0))
+           .select(-Vector::Ones(v.size()), Vector::Ones(v.size()));
+  }
+};
+
+/**
+  * \returns an estimate of ||inv(matrix)||_1 given a decomposition of
+  * \a matrix that implements .solve() and .adjoint().solve() methods.
+  *
+  * This function implements Algorithms 4.1 and 5.1 from
+  *   http://www.maths.manchester.ac.uk/~higham/narep/narep135.pdf
+  * which also forms the basis for the condition number estimators in
+  * LAPACK. Since at most 10 calls to the solve method of dec are
+  * performed, the total cost is O(dims^2), as opposed to O(dims^3)
+  * needed to compute the inverse matrix explicitly.
+  *
+  * The most common usage is in estimating the condition number
+  * ||matrix||_1 * ||inv(matrix)||_1. The first term ||matrix||_1 can be
+  * computed directly in O(n^2) operations.
+  *
+  * Supports the following decompositions: FullPivLU, PartialPivLU, LDLT, and
+  * LLT.
+  *
+  * \sa FullPivLU, PartialPivLU, LDLT, LLT.
+  */
+template <typename Decomposition>
+typename Decomposition::RealScalar rcond_invmatrix_L1_norm_estimate(const Decomposition& dec)
+{
+  typedef typename Decomposition::MatrixType MatrixType;
+  typedef typename Decomposition::Scalar Scalar;
+  typedef typename Decomposition::RealScalar RealScalar;
+  typedef typename internal::plain_col_type<MatrixType>::type Vector;
+  typedef typename internal::plain_col_type<MatrixType, RealScalar>::type RealVector;
+  const bool is_complex = (NumTraits<Scalar>::IsComplex != 0);
+
+  eigen_assert(dec.rows() == dec.cols());
+  const Index n = dec.rows();
+  if (n == 0)
+    return 0;
+
+  // Disable Index to float conversion warning
+#ifdef __INTEL_COMPILER
+  #pragma warning push
+  #pragma warning ( disable : 2259 )
+#endif
+  Vector v = dec.solve(Vector::Ones(n) / Scalar(n));
+#ifdef __INTEL_COMPILER
+  #pragma warning pop
+#endif
+
+  // lower_bound is a lower bound on
+  //   ||inv(matrix)||_1  = sup_v ||inv(matrix) v||_1 / ||v||_1
+  // and is the objective maximized by the ("super-") gradient ascent
+  // algorithm below.
+  RealScalar lower_bound = v.template lpNorm<1>();
+  if (n == 1)
+    return lower_bound;
+
+  // Gradient ascent algorithm follows: We know that the optimum is achieved at
+  // one of the simplices v = e_i, so in each iteration we follow a
+  // super-gradient to move towards the optimal one.
+  RealScalar old_lower_bound = lower_bound;
+  Vector sign_vector(n);
+  Vector old_sign_vector;
+  Index v_max_abs_index = -1;
+  Index old_v_max_abs_index = v_max_abs_index;
+  for (int k = 0; k < 4; ++k)
+  {
+    sign_vector = internal::rcond_compute_sign<Vector, RealVector, is_complex>::run(v);
+    if (k > 0 && !is_complex && sign_vector == old_sign_vector) {
+      // Break if the solution stagnated.
+      break;
+    }
+    // v_max_abs_index = argmax |real( inv(matrix)^T * sign_vector )|
+    v = dec.adjoint().solve(sign_vector);
+    v.real().cwiseAbs().maxCoeff(&v_max_abs_index);
+    if (v_max_abs_index == old_v_max_abs_index) {
+      // Break if the solution stagnated.
+      break;
+    }
+    // Move to the new simplex e_j, where j = v_max_abs_index.
+    v = dec.solve(Vector::Unit(n, v_max_abs_index));  // v = inv(matrix) * e_j.
+    lower_bound = v.template lpNorm<1>();
+    if (lower_bound <= old_lower_bound) {
+      // Break if the gradient step did not increase the lower_bound.
+      break;
+    }
+    if (!is_complex) {
+      old_sign_vector = sign_vector;
+    }
+    old_v_max_abs_index = v_max_abs_index;
+    old_lower_bound = lower_bound;
+  }
+  // The following calculates an independent estimate of ||matrix||_1 by
+  // multiplying matrix by a vector with entries of slowly increasing
+  // magnitude and alternating sign:
+  //   v_i = (-1)^{i} (1 + (i / (dim-1))), i = 0,...,dim-1.
+  // This improvement to Hager's algorithm above is due to Higham. It was
+  // added to make the algorithm more robust in certain corner cases where
+  // large elements in the matrix might otherwise escape detection due to
+  // exact cancellation (especially when op and op_adjoint correspond to a
+  // sequence of backsubstitutions and permutations), which could cause
+  // Hager's algorithm to vastly underestimate ||matrix||_1.
+  Scalar alternating_sign(RealScalar(1));
+  for (Index i = 0; i < n; ++i) {
+    // The static_cast is needed when Scalar is a complex and RealScalar implements expression templates
+    v[i] = alternating_sign * static_cast<RealScalar>(RealScalar(1) + (RealScalar(i) / (RealScalar(n - 1))));
+    alternating_sign = -alternating_sign;
+  }
+  v = dec.solve(v);
+  const RealScalar alternate_lower_bound = (2 * v.template lpNorm<1>()) / (3 * RealScalar(n));
+  return numext::maxi(lower_bound, alternate_lower_bound);
+}
+
+/** \brief Reciprocal condition number estimator.
+  *
+  * Computing a decomposition of a dense matrix takes O(n^3) operations, while
+  * this method estimates the condition number quickly and reliably in O(n^2)
+  * operations.
+  *
+  * \returns an estimate of the reciprocal condition number
+  * (1 / (||matrix||_1 * ||inv(matrix)||_1)) of matrix, given ||matrix||_1 and
+  * its decomposition. Supports the following decompositions: FullPivLU,
+  * PartialPivLU, LDLT, and LLT.
+  *
+  * \sa FullPivLU, PartialPivLU, LDLT, LLT.
+  */
+template <typename Decomposition>
+typename Decomposition::RealScalar
+rcond_estimate_helper(typename Decomposition::RealScalar matrix_norm, const Decomposition& dec)
+{
+  typedef typename Decomposition::RealScalar RealScalar;
+  eigen_assert(dec.rows() == dec.cols());
+  if (dec.rows() == 0)              return NumTraits<RealScalar>::infinity();
+  if (matrix_norm == RealScalar(0)) return RealScalar(0);
+  if (dec.rows() == 1)              return RealScalar(1);
+  const RealScalar inverse_matrix_norm = rcond_invmatrix_L1_norm_estimate(dec);
+  return (inverse_matrix_norm == RealScalar(0) ? RealScalar(0)
+                                               : (RealScalar(1) / inverse_matrix_norm) / matrix_norm);
+}
+
+}  // namespace internal
+
+}  // namespace Eigen
+
+#endif
diff --git a/Eigen/src/Core/CoreEvaluators.h b/Eigen/src/Core/CoreEvaluators.h
new file mode 100644
index 0000000..910889e
--- /dev/null
+++ b/Eigen/src/Core/CoreEvaluators.h
@@ -0,0 +1,1688 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+
+#ifndef EIGEN_COREEVALUATORS_H
+#define EIGEN_COREEVALUATORS_H
+
+namespace Eigen {
+  
+namespace internal {
+
+// This class returns the evaluator kind from the expression storage kind.
+// Default assumes index based accessors
+template<typename StorageKind>
+struct storage_kind_to_evaluator_kind {
+  typedef IndexBased Kind;
+};
+
+// This class returns the evaluator shape from the expression storage kind.
+// It can be Dense, Sparse, Triangular, Diagonal, SelfAdjoint, Band, etc.
+template<typename StorageKind> struct storage_kind_to_shape;
+
+template<> struct storage_kind_to_shape<Dense>                  { typedef DenseShape Shape;           };
+template<> struct storage_kind_to_shape<SolverStorage>          { typedef SolverShape Shape;           };
+template<> struct storage_kind_to_shape<PermutationStorage>     { typedef PermutationShape Shape;     };
+template<> struct storage_kind_to_shape<TranspositionsStorage>  { typedef TranspositionsShape Shape;  };
+
+// Evaluators have to be specialized with respect to various criteria such as:
+//  - storage/structure/shape
+//  - scalar type
+//  - etc.
+// Therefore, we need specialization of evaluator providing additional template arguments for each kind of evaluators.
+// We currently distinguish the following kind of evaluators:
+// - unary_evaluator    for expressions taking only one arguments (CwiseUnaryOp, CwiseUnaryView, Transpose, MatrixWrapper, ArrayWrapper, Reverse, Replicate)
+// - binary_evaluator   for expression taking two arguments (CwiseBinaryOp)
+// - ternary_evaluator   for expression taking three arguments (CwiseTernaryOp)
+// - product_evaluator  for linear algebra products (Product); special case of binary_evaluator because it requires additional tags for dispatching.
+// - mapbase_evaluator  for Map, Block, Ref
+// - block_evaluator    for Block (special dispatching to a mapbase_evaluator or unary_evaluator)
+
+template< typename T,
+          typename Arg1Kind   = typename evaluator_traits<typename T::Arg1>::Kind,
+          typename Arg2Kind   = typename evaluator_traits<typename T::Arg2>::Kind,
+          typename Arg3Kind   = typename evaluator_traits<typename T::Arg3>::Kind,
+          typename Arg1Scalar = typename traits<typename T::Arg1>::Scalar,
+          typename Arg2Scalar = typename traits<typename T::Arg2>::Scalar,
+          typename Arg3Scalar = typename traits<typename T::Arg3>::Scalar> struct ternary_evaluator;
+
+template< typename T,
+          typename LhsKind   = typename evaluator_traits<typename T::Lhs>::Kind,
+          typename RhsKind   = typename evaluator_traits<typename T::Rhs>::Kind,
+          typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
+          typename RhsScalar = typename traits<typename T::Rhs>::Scalar> struct binary_evaluator;
+
+template< typename T,
+          typename Kind   = typename evaluator_traits<typename T::NestedExpression>::Kind,
+          typename Scalar = typename T::Scalar> struct unary_evaluator;
+          
+// evaluator_traits<T> contains traits for evaluator<T> 
+
+template<typename T>
+struct evaluator_traits_base
+{
+  // by default, get evaluator kind and shape from storage
+  typedef typename storage_kind_to_evaluator_kind<typename traits<T>::StorageKind>::Kind Kind;
+  typedef typename storage_kind_to_shape<typename traits<T>::StorageKind>::Shape Shape;
+};
+
+// Default evaluator traits
+template<typename T>
+struct evaluator_traits : public evaluator_traits_base<T>
+{
+};
+
+template<typename T, typename Shape = typename evaluator_traits<T>::Shape >
+struct evaluator_assume_aliasing {
+  static const bool value = false;
+};
+
+// By default, we assume a unary expression:
+template<typename T>
+struct evaluator : public unary_evaluator<T>
+{
+  typedef unary_evaluator<T> Base;
+  EIGEN_DEVICE_FUNC explicit evaluator(const T& xpr) : Base(xpr) {}
+};
+
+
+// TODO: Think about const-correctness
+template<typename T>
+struct evaluator<const T>
+  : evaluator<T>
+{
+  EIGEN_DEVICE_FUNC
+  explicit evaluator(const T& xpr) : evaluator<T>(xpr) {}
+};
+
+// ---------- base class for all evaluators ----------
+
+template<typename ExpressionType>
+struct evaluator_base : public noncopyable
+{
+  // TODO that's not very nice to have to propagate all these traits. They are currently only needed to handle outer,inner indices.
+  typedef traits<ExpressionType> ExpressionTraits;
+  
+  enum {
+    Alignment = 0
+  };
+};
+
+// -------------------- Matrix and Array --------------------
+//
+// evaluator<PlainObjectBase> is a common base class for the
+// Matrix and Array evaluators.
+// Here we directly specialize evaluator. This is not really a unary expression, and it is, by definition, dense,
+// so no need for more sophisticated dispatching.
+
+template<typename Derived>
+struct evaluator<PlainObjectBase<Derived> >
+  : evaluator_base<Derived>
+{
+  typedef PlainObjectBase<Derived> PlainObjectType;
+  typedef typename PlainObjectType::Scalar Scalar;
+  typedef typename PlainObjectType::CoeffReturnType CoeffReturnType;
+
+  enum {
+    IsRowMajor = PlainObjectType::IsRowMajor,
+    IsVectorAtCompileTime = PlainObjectType::IsVectorAtCompileTime,
+    RowsAtCompileTime = PlainObjectType::RowsAtCompileTime,
+    ColsAtCompileTime = PlainObjectType::ColsAtCompileTime,
+    
+    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = traits<Derived>::EvaluatorFlags,
+    Alignment = traits<Derived>::Alignment
+  };
+  
+  EIGEN_DEVICE_FUNC evaluator()
+    : m_data(0),
+      m_outerStride(IsVectorAtCompileTime  ? 0 
+                                           : int(IsRowMajor) ? ColsAtCompileTime 
+                                           : RowsAtCompileTime)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const PlainObjectType& m)
+    : m_data(m.data()), m_outerStride(IsVectorAtCompileTime ? 0 : m.outerStride()) 
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    if (IsRowMajor)
+      return m_data[row * m_outerStride.value() + col];
+    else
+      return m_data[row + col * m_outerStride.value()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_data[index];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    if (IsRowMajor)
+      return const_cast<Scalar*>(m_data)[row * m_outerStride.value() + col];
+    else
+      return const_cast<Scalar*>(m_data)[row + col * m_outerStride.value()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return const_cast<Scalar*>(m_data)[index];
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    if (IsRowMajor)
+      return ploadt<PacketType, LoadMode>(m_data + row * m_outerStride.value() + col);
+    else
+      return ploadt<PacketType, LoadMode>(m_data + row + col * m_outerStride.value());
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return ploadt<PacketType, LoadMode>(m_data + index);
+  }
+
+  template<int StoreMode,typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    if (IsRowMajor)
+      return pstoret<Scalar, PacketType, StoreMode>
+	            (const_cast<Scalar*>(m_data) + row * m_outerStride.value() + col, x);
+    else
+      return pstoret<Scalar, PacketType, StoreMode>
+                    (const_cast<Scalar*>(m_data) + row + col * m_outerStride.value(), x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    return pstoret<Scalar, PacketType, StoreMode>(const_cast<Scalar*>(m_data) + index, x);
+  }
+
+protected:
+  const Scalar *m_data;
+
+  // We do not need to know the outer stride for vectors
+  variable_if_dynamic<Index, IsVectorAtCompileTime  ? 0 
+                                                    : int(IsRowMajor) ? ColsAtCompileTime 
+                                                    : RowsAtCompileTime> m_outerStride;
+};
+
+template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
+struct evaluator<Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
+  : evaluator<PlainObjectBase<Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> > >
+{
+  typedef Matrix<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
+  
+  EIGEN_DEVICE_FUNC evaluator() {}
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
+    : evaluator<PlainObjectBase<XprType> >(m) 
+  { }
+};
+
+template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
+struct evaluator<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> >
+  : evaluator<PlainObjectBase<Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> > >
+{
+  typedef Array<Scalar, Rows, Cols, Options, MaxRows, MaxCols> XprType;
+
+  EIGEN_DEVICE_FUNC evaluator() {}
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& m)
+    : evaluator<PlainObjectBase<XprType> >(m) 
+  { }
+};
+
+// -------------------- Transpose --------------------
+
+template<typename ArgType>
+struct unary_evaluator<Transpose<ArgType>, IndexBased>
+  : evaluator_base<Transpose<ArgType> >
+{
+  typedef Transpose<ArgType> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,    
+    Flags = evaluator<ArgType>::Flags ^ RowMajorBit,
+    Alignment = evaluator<ArgType>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& t) : m_argImpl(t.nestedExpression()) {}
+
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_argImpl.coeff(col, row);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(index);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_argImpl.coeffRef(col, row);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  typename XprType::Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(index);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(col, row);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(index);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode,PacketType>(col, row, x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode,PacketType>(index, x);
+  }
+
+protected:
+  evaluator<ArgType> m_argImpl;
+};
+
+// -------------------- CwiseNullaryOp --------------------
+// Like Matrix and Array, this is not really a unary expression, so we directly specialize evaluator.
+// Likewise, there is not need to more sophisticated dispatching here.
+
+template<typename Scalar,typename NullaryOp,
+         bool has_nullary = has_nullary_operator<NullaryOp>::value,
+         bool has_unary   = has_unary_operator<NullaryOp>::value,
+         bool has_binary  = has_binary_operator<NullaryOp>::value>
+struct nullary_wrapper
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const { return op(i,j); }
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); }
+
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const { return op.template packetOp<T>(i,j); }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp<T>(i); }
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,true,false,false>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType=0, IndexType=0) const { return op(); }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType=0, IndexType=0) const { return op.template packetOp<T>(); }
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,false,false,true>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j=0) const { return op(i,j); }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j=0) const { return op.template packetOp<T>(i,j); }
+};
+
+// We need the following specialization for vector-only functors assigned to a runtime vector,
+// for instance, using linspace and assigning a RowVectorXd to a MatrixXd or even a row of a MatrixXd.
+// In this case, i==0 and j is used for the actual iteration.
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,false,true,false>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const {
+    eigen_assert(i==0 || j==0);
+    return op(i+j);
+  }
+  template <typename T, typename IndexType> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const {
+    eigen_assert(i==0 || j==0);
+    return op.template packetOp<T>(i+j);
+  }
+
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); }
+  template <typename T, typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp<T>(i); }
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,false,false,false> {};
+
+#if 0 && EIGEN_COMP_MSVC>0
+// Disable this ugly workaround. This is now handled in traits<Ref>::match,
+// but this piece of code might still become handly if some other weird compilation
+// erros pop up again.
+
+// MSVC exhibits a weird compilation error when
+// compiling:
+//    Eigen::MatrixXf A = MatrixXf::Random(3,3);
+//    Ref<const MatrixXf> R = 2.f*A;
+// and that has_*ary_operator<scalar_constant_op<float>> have not been instantiated yet.
+// The "problem" is that evaluator<2.f*A> is instantiated by traits<Ref>::match<2.f*A>
+// and at that time has_*ary_operator<T> returns true regardless of T.
+// Then nullary_wrapper is badly instantiated as nullary_wrapper<.,.,true,true,true>.
+// The trick is thus to defer the proper instantiation of nullary_wrapper when coeff(),
+// and packet() are really instantiated as implemented below:
+
+// This is a simple wrapper around Index to enforce the re-instantiation of
+// has_*ary_operator when needed.
+template<typename T> struct nullary_wrapper_workaround_msvc {
+  nullary_wrapper_workaround_msvc(const T&);
+  operator T()const;
+};
+
+template<typename Scalar,typename NullaryOp>
+struct nullary_wrapper<Scalar,NullaryOp,true,true,true>
+{
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().operator()(op,i,j);
+  }
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().operator()(op,i);
+  }
+
+  template <typename T, typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().template packetOp<T>(op,i,j);
+  }
+  template <typename T, typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const {
+    return nullary_wrapper<Scalar,NullaryOp,
+    has_nullary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_unary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value,
+    has_binary_operator<NullaryOp,nullary_wrapper_workaround_msvc<IndexType> >::value>().template packetOp<T>(op,i);
+  }
+};
+#endif // MSVC workaround
+
+template<typename NullaryOp, typename PlainObjectType>
+struct evaluator<CwiseNullaryOp<NullaryOp,PlainObjectType> >
+  : evaluator_base<CwiseNullaryOp<NullaryOp,PlainObjectType> >
+{
+  typedef CwiseNullaryOp<NullaryOp,PlainObjectType> XprType;
+  typedef typename internal::remove_all<PlainObjectType>::type PlainObjectTypeCleaned;
+  
+  enum {
+    CoeffReadCost = internal::functor_traits<NullaryOp>::Cost,
+    
+    Flags = (evaluator<PlainObjectTypeCleaned>::Flags
+          &  (  HereditaryBits
+              | (functor_has_linear_access<NullaryOp>::ret  ? LinearAccessBit : 0)
+              | (functor_traits<NullaryOp>::PacketAccess    ? PacketAccessBit : 0)))
+          | (functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
+    Alignment = AlignedMax
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& n)
+    : m_functor(n.functor()), m_wrapper()
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(IndexType row, IndexType col) const
+  {
+    return m_wrapper(m_functor, row, col);
+  }
+
+  template <typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(IndexType index) const
+  {
+    return m_wrapper(m_functor,index);
+  }
+
+  template<int LoadMode, typename PacketType, typename IndexType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(IndexType row, IndexType col) const
+  {
+    return m_wrapper.template packetOp<PacketType>(m_functor, row, col);
+  }
+
+  template<int LoadMode, typename PacketType, typename IndexType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(IndexType index) const
+  {
+    return m_wrapper.template packetOp<PacketType>(m_functor, index);
+  }
+
+protected:
+  const NullaryOp m_functor;
+  const internal::nullary_wrapper<CoeffReturnType,NullaryOp> m_wrapper;
+};
+
+// -------------------- CwiseUnaryOp --------------------
+
+template<typename UnaryOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryOp<UnaryOp, ArgType>, IndexBased >
+  : evaluator_base<CwiseUnaryOp<UnaryOp, ArgType> >
+{
+  typedef CwiseUnaryOp<UnaryOp, ArgType> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
+    
+    Flags = evaluator<ArgType>::Flags
+          & (HereditaryBits | LinearAccessBit | (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
+    Alignment = evaluator<ArgType>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  explicit unary_evaluator(const XprType& op)
+    : m_functor(op.functor()), 
+      m_argImpl(op.nestedExpression()) 
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_functor(m_argImpl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_functor(m_argImpl.coeff(index));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(row, col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_functor.packetOp(m_argImpl.template packet<LoadMode, PacketType>(index));
+  }
+
+protected:
+  const UnaryOp m_functor;
+  evaluator<ArgType> m_argImpl;
+};
+
+// -------------------- CwiseTernaryOp --------------------
+
+// this is a ternary expression
+template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
+struct evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
+  : public ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
+{
+  typedef CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> XprType;
+  typedef ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> > Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
+struct ternary_evaluator<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3>, IndexBased, IndexBased>
+  : evaluator_base<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >
+{
+  typedef CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<Arg1>::CoeffReadCost + evaluator<Arg2>::CoeffReadCost + evaluator<Arg3>::CoeffReadCost + functor_traits<TernaryOp>::Cost,
+    
+    Arg1Flags = evaluator<Arg1>::Flags,
+    Arg2Flags = evaluator<Arg2>::Flags,
+    Arg3Flags = evaluator<Arg3>::Flags,
+    SameType = is_same<typename Arg1::Scalar,typename Arg2::Scalar>::value && is_same<typename Arg1::Scalar,typename Arg3::Scalar>::value,
+    StorageOrdersAgree = (int(Arg1Flags)&RowMajorBit)==(int(Arg2Flags)&RowMajorBit) && (int(Arg1Flags)&RowMajorBit)==(int(Arg3Flags)&RowMajorBit),
+    Flags0 = (int(Arg1Flags) | int(Arg2Flags) | int(Arg3Flags)) & (
+        HereditaryBits
+        | (int(Arg1Flags) & int(Arg2Flags) & int(Arg3Flags) &
+           ( (StorageOrdersAgree ? LinearAccessBit : 0)
+           | (functor_traits<TernaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
+           )
+        )
+     ),
+    Flags = (Flags0 & ~RowMajorBit) | (Arg1Flags & RowMajorBit),
+    Alignment = EIGEN_PLAIN_ENUM_MIN(
+        EIGEN_PLAIN_ENUM_MIN(evaluator<Arg1>::Alignment, evaluator<Arg2>::Alignment),
+        evaluator<Arg3>::Alignment)
+  };
+
+  EIGEN_DEVICE_FUNC explicit ternary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_arg1Impl(xpr.arg1()), 
+      m_arg2Impl(xpr.arg2()), 
+      m_arg3Impl(xpr.arg3())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<TernaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_functor(m_arg1Impl.coeff(row, col), m_arg2Impl.coeff(row, col), m_arg3Impl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_functor(m_arg1Impl.coeff(index), m_arg2Impl.coeff(index), m_arg3Impl.coeff(index));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(row, col),
+                              m_arg2Impl.template packet<LoadMode,PacketType>(row, col),
+                              m_arg3Impl.template packet<LoadMode,PacketType>(row, col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_functor.packetOp(m_arg1Impl.template packet<LoadMode,PacketType>(index),
+                              m_arg2Impl.template packet<LoadMode,PacketType>(index),
+                              m_arg3Impl.template packet<LoadMode,PacketType>(index));
+  }
+
+protected:
+  const TernaryOp m_functor;
+  evaluator<Arg1> m_arg1Impl;
+  evaluator<Arg2> m_arg2Impl;
+  evaluator<Arg3> m_arg3Impl;
+};
+
+// -------------------- CwiseBinaryOp --------------------
+
+// this is a binary expression
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+  : public binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs> > Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IndexBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    
+    LhsFlags = evaluator<Lhs>::Flags,
+    RhsFlags = evaluator<Rhs>::Flags,
+    SameType = is_same<typename Lhs::Scalar,typename Rhs::Scalar>::value,
+    StorageOrdersAgree = (int(LhsFlags)&RowMajorBit)==(int(RhsFlags)&RowMajorBit),
+    Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
+        HereditaryBits
+      | (int(LhsFlags) & int(RhsFlags) &
+           ( (StorageOrdersAgree ? LinearAccessBit : 0)
+           | (functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
+           )
+        )
+     ),
+    Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
+    Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<Lhs>::Alignment,evaluator<Rhs>::Alignment)
+  };
+
+  EIGEN_DEVICE_FUNC explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_functor(m_lhsImpl.coeff(row, col), m_rhsImpl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_functor(m_lhsImpl.coeff(index), m_rhsImpl.coeff(index));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(row, col),
+                              m_rhsImpl.template packet<LoadMode,PacketType>(row, col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_functor.packetOp(m_lhsImpl.template packet<LoadMode,PacketType>(index),
+                              m_rhsImpl.template packet<LoadMode,PacketType>(index));
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
+};
+
+// -------------------- CwiseUnaryView --------------------
+
+template<typename UnaryOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryView<UnaryOp, ArgType>, IndexBased>
+  : evaluator_base<CwiseUnaryView<UnaryOp, ArgType> >
+{
+  typedef CwiseUnaryView<UnaryOp, ArgType> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
+    
+    Flags = (evaluator<ArgType>::Flags & (HereditaryBits | LinearAccessBit | DirectAccessBit)),
+    
+    Alignment = 0 // FIXME it is not very clear why alignment is necessarily lost...
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op)
+    : m_unaryOp(op.functor()), 
+      m_argImpl(op.nestedExpression()) 
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_unaryOp(m_argImpl.coeff(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_unaryOp(m_argImpl.coeff(index));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_unaryOp(m_argImpl.coeffRef(row, col));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_unaryOp(m_argImpl.coeffRef(index));
+  }
+
+protected:
+  const UnaryOp m_unaryOp;
+  evaluator<ArgType> m_argImpl;
+};
+
+// -------------------- Map --------------------
+
+// FIXME perhaps the PlainObjectType could be provided by Derived::PlainObject ?
+// but that might complicate template specialization
+template<typename Derived, typename PlainObjectType>
+struct mapbase_evaluator;
+
+template<typename Derived, typename PlainObjectType>
+struct mapbase_evaluator : evaluator_base<Derived>
+{
+  typedef Derived  XprType;
+  typedef typename XprType::PointerType PointerType;
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  
+  enum {
+    IsRowMajor = XprType::RowsAtCompileTime,
+    ColsAtCompileTime = XprType::ColsAtCompileTime,
+    CoeffReadCost = NumTraits<Scalar>::ReadCost
+  };
+
+  EIGEN_DEVICE_FUNC explicit mapbase_evaluator(const XprType& map)
+    : m_data(const_cast<PointerType>(map.data())),
+      m_innerStride(map.innerStride()),
+      m_outerStride(map.outerStride())
+  {
+    EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(evaluator<Derived>::Flags&PacketAccessBit, internal::inner_stride_at_compile_time<Derived>::ret==1),
+                        PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_data[col * colStride() + row * rowStride()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_data[index * m_innerStride.value()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_data[col * colStride() + row * rowStride()];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_data[index * m_innerStride.value()];
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    PointerType ptr = m_data + row * rowStride() + col * colStride();
+    return internal::ploadt<PacketType, LoadMode>(ptr);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return internal::ploadt<PacketType, LoadMode>(m_data + index * m_innerStride.value());
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    PointerType ptr = m_data + row * rowStride() + col * colStride();
+    return internal::pstoret<Scalar, PacketType, StoreMode>(ptr, x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    internal::pstoret<Scalar, PacketType, StoreMode>(m_data + index * m_innerStride.value(), x);
+  }
+protected:
+  EIGEN_DEVICE_FUNC
+  inline Index rowStride() const { return XprType::IsRowMajor ? m_outerStride.value() : m_innerStride.value(); }
+  EIGEN_DEVICE_FUNC
+  inline Index colStride() const { return XprType::IsRowMajor ? m_innerStride.value() : m_outerStride.value(); }
+
+  PointerType m_data;
+  const internal::variable_if_dynamic<Index, XprType::InnerStrideAtCompileTime> m_innerStride;
+  const internal::variable_if_dynamic<Index, XprType::OuterStrideAtCompileTime> m_outerStride;
+};
+
+template<typename PlainObjectType, int MapOptions, typename StrideType> 
+struct evaluator<Map<PlainObjectType, MapOptions, StrideType> >
+  : public mapbase_evaluator<Map<PlainObjectType, MapOptions, StrideType>, PlainObjectType>
+{
+  typedef Map<PlainObjectType, MapOptions, StrideType> XprType;
+  typedef typename XprType::Scalar Scalar;
+  // TODO: should check for smaller packet types once we can handle multi-sized packet types
+  typedef typename packet_traits<Scalar>::type PacketScalar;
+  
+  enum {
+    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
+                             ? int(PlainObjectType::InnerStrideAtCompileTime)
+                             : int(StrideType::InnerStrideAtCompileTime),
+    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
+                             ? int(PlainObjectType::OuterStrideAtCompileTime)
+                             : int(StrideType::OuterStrideAtCompileTime),
+    HasNoInnerStride = InnerStrideAtCompileTime == 1,
+    HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
+    HasNoStride = HasNoInnerStride && HasNoOuterStride,
+    IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic,
+    
+    PacketAccessMask = bool(HasNoInnerStride) ? ~int(0) : ~int(PacketAccessBit),
+    LinearAccessMask = bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime) ? ~int(0) : ~int(LinearAccessBit),
+    Flags = int( evaluator<PlainObjectType>::Flags) & (LinearAccessMask&PacketAccessMask),
+    
+    Alignment = int(MapOptions)&int(AlignedMask)
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& map)
+    : mapbase_evaluator<XprType, PlainObjectType>(map) 
+  { }
+};
+
+// -------------------- Ref --------------------
+
+template<typename PlainObjectType, int RefOptions, typename StrideType> 
+struct evaluator<Ref<PlainObjectType, RefOptions, StrideType> >
+  : public mapbase_evaluator<Ref<PlainObjectType, RefOptions, StrideType>, PlainObjectType>
+{
+  typedef Ref<PlainObjectType, RefOptions, StrideType> XprType;
+  
+  enum {
+    Flags = evaluator<Map<PlainObjectType, RefOptions, StrideType> >::Flags,
+    Alignment = evaluator<Map<PlainObjectType, RefOptions, StrideType> >::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& ref)
+    : mapbase_evaluator<XprType, PlainObjectType>(ref) 
+  { }
+};
+
+// -------------------- Block --------------------
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel,
+         bool HasDirectAccess = internal::has_direct_access<ArgType>::ret> struct block_evaluator;
+         
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel> 
+struct evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
+  : block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel>
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+  typedef typename XprType::Scalar Scalar;
+  // TODO: should check for smaller packet types once we can handle multi-sized packet types
+  typedef typename packet_traits<Scalar>::type PacketScalar;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    
+    RowsAtCompileTime = traits<XprType>::RowsAtCompileTime,
+    ColsAtCompileTime = traits<XprType>::ColsAtCompileTime,
+    MaxRowsAtCompileTime = traits<XprType>::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = traits<XprType>::MaxColsAtCompileTime,
+    
+    ArgTypeIsRowMajor = (int(evaluator<ArgType>::Flags)&RowMajorBit) != 0,
+    IsRowMajor = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? 1
+               : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
+               : ArgTypeIsRowMajor,
+    HasSameStorageOrderAsArgType = (IsRowMajor == ArgTypeIsRowMajor),
+    InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
+    InnerStrideAtCompileTime = HasSameStorageOrderAsArgType
+                             ? int(inner_stride_at_compile_time<ArgType>::ret)
+                             : int(outer_stride_at_compile_time<ArgType>::ret),
+    OuterStrideAtCompileTime = HasSameStorageOrderAsArgType
+                             ? int(outer_stride_at_compile_time<ArgType>::ret)
+                             : int(inner_stride_at_compile_time<ArgType>::ret),
+    MaskPacketAccessBit = (InnerStrideAtCompileTime == 1 || HasSameStorageOrderAsArgType) ? PacketAccessBit : 0,
+    
+    FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (evaluator<ArgType>::Flags&LinearAccessBit))) ? LinearAccessBit : 0,    
+    FlagsRowMajorBit = XprType::Flags&RowMajorBit,
+    Flags0 = evaluator<ArgType>::Flags & ( (HereditaryBits & ~RowMajorBit) |
+                                           DirectAccessBit |
+                                           MaskPacketAccessBit),
+    Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit,
+    
+    PacketAlignment = unpacket_traits<PacketScalar>::alignment,
+    Alignment0 = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic)
+                             && (OuterStrideAtCompileTime!=0)
+                             && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % int(PacketAlignment)) == 0)) ? int(PacketAlignment) : 0,
+    Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ArgType>::Alignment, Alignment0)
+  };
+  typedef block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel> block_evaluator_type;
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& block) : block_evaluator_type(block)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+};
+
+// no direct-access => dispatch to a unary evaluator
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /*HasDirectAccess*/ false>
+  : unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+
+  EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
+    : unary_evaluator<XprType>(block) 
+  {}
+};
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+struct unary_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>, IndexBased>
+  : evaluator_base<Block<ArgType, BlockRows, BlockCols, InnerPanel> >
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& block)
+    : m_argImpl(block.nestedExpression()), 
+      m_startRow(block.startRow()), 
+      m_startCol(block.startCol()),
+      m_linear_offset(InnerPanel?(XprType::IsRowMajor ? block.startRow()*block.cols() : block.startCol()*block.rows()):0)
+  { }
+ 
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  enum {
+    RowsAtCompileTime = XprType::RowsAtCompileTime,
+    ForwardLinearAccess = InnerPanel && bool(evaluator<ArgType>::Flags&LinearAccessBit)
+  };
+ 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  { 
+    return m_argImpl.coeff(m_startRow.value() + row, m_startCol.value() + col); 
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  { 
+    if (ForwardLinearAccess)
+      return m_argImpl.coeff(m_linear_offset.value() + index); 
+    else
+      return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  { 
+    return m_argImpl.coeffRef(m_startRow.value() + row, m_startCol.value() + col); 
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  { 
+    if (ForwardLinearAccess)
+      return m_argImpl.coeffRef(m_linear_offset.value() + index); 
+    else
+      return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0);
+  }
+ 
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const 
+  { 
+    return m_argImpl.template packet<LoadMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col); 
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const 
+  { 
+    if (ForwardLinearAccess)
+      return m_argImpl.template packet<LoadMode,PacketType>(m_linear_offset.value() + index);
+    else
+      return packet<LoadMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
+                                         RowsAtCompileTime == 1 ? index : 0);
+  }
+  
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x) 
+  {
+    return m_argImpl.template writePacket<StoreMode,PacketType>(m_startRow.value() + row, m_startCol.value() + col, x); 
+  }
+  
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x) 
+  {
+    if (ForwardLinearAccess)
+      return m_argImpl.template writePacket<StoreMode,PacketType>(m_linear_offset.value() + index, x);
+    else
+      return writePacket<StoreMode,PacketType>(RowsAtCompileTime == 1 ? 0 : index,
+                                              RowsAtCompileTime == 1 ? index : 0,
+                                              x);
+  }
+ 
+protected:
+  evaluator<ArgType> m_argImpl;
+  const variable_if_dynamic<Index, (ArgType::RowsAtCompileTime == 1 && BlockRows==1) ? 0 : Dynamic> m_startRow;
+  const variable_if_dynamic<Index, (ArgType::ColsAtCompileTime == 1 && BlockCols==1) ? 0 : Dynamic> m_startCol;
+  const variable_if_dynamic<Index, InnerPanel ? Dynamic : 0> m_linear_offset;
+};
+
+// TODO: This evaluator does not actually use the child evaluator; 
+// all action is via the data() as returned by the Block expression.
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel> 
+struct block_evaluator<ArgType, BlockRows, BlockCols, InnerPanel, /* HasDirectAccess */ true>
+  : mapbase_evaluator<Block<ArgType, BlockRows, BlockCols, InnerPanel>,
+                      typename Block<ArgType, BlockRows, BlockCols, InnerPanel>::PlainObject>
+{
+  typedef Block<ArgType, BlockRows, BlockCols, InnerPanel> XprType;
+  typedef typename XprType::Scalar Scalar;
+
+  EIGEN_DEVICE_FUNC explicit block_evaluator(const XprType& block)
+    : mapbase_evaluator<XprType, typename XprType::PlainObject>(block) 
+  {
+    // TODO: for the 3.3 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime
+    eigen_assert(((internal::UIntPtr(block.data()) % EIGEN_PLAIN_ENUM_MAX(1,evaluator<XprType>::Alignment)) == 0) && "data is not aligned");
+  }
+};
+
+
+// -------------------- Select --------------------
+// NOTE shall we introduce a ternary_evaluator?
+
+// TODO enable vectorization for Select
+template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
+struct evaluator<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
+  : evaluator_base<Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >
+{
+  typedef Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> XprType;
+  enum {
+    CoeffReadCost = evaluator<ConditionMatrixType>::CoeffReadCost
+                  + EIGEN_PLAIN_ENUM_MAX(evaluator<ThenMatrixType>::CoeffReadCost,
+                                         evaluator<ElseMatrixType>::CoeffReadCost),
+
+    Flags = (unsigned int)evaluator<ThenMatrixType>::Flags & evaluator<ElseMatrixType>::Flags & HereditaryBits,
+    
+    Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator<ThenMatrixType>::Alignment, evaluator<ElseMatrixType>::Alignment)
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& select)
+    : m_conditionImpl(select.conditionMatrix()),
+      m_thenImpl(select.thenMatrix()),
+      m_elseImpl(select.elseMatrix())
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+ 
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    if (m_conditionImpl.coeff(row, col))
+      return m_thenImpl.coeff(row, col);
+    else
+      return m_elseImpl.coeff(row, col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    if (m_conditionImpl.coeff(index))
+      return m_thenImpl.coeff(index);
+    else
+      return m_elseImpl.coeff(index);
+  }
+ 
+protected:
+  evaluator<ConditionMatrixType> m_conditionImpl;
+  evaluator<ThenMatrixType> m_thenImpl;
+  evaluator<ElseMatrixType> m_elseImpl;
+};
+
+
+// -------------------- Replicate --------------------
+
+template<typename ArgType, int RowFactor, int ColFactor> 
+struct unary_evaluator<Replicate<ArgType, RowFactor, ColFactor> >
+  : evaluator_base<Replicate<ArgType, RowFactor, ColFactor> >
+{
+  typedef Replicate<ArgType, RowFactor, ColFactor> XprType;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  enum {
+    Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor
+  };
+  typedef typename internal::nested_eval<ArgType,Factor>::type ArgTypeNested;
+  typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgTypeNestedCleaned>::CoeffReadCost,
+    LinearAccessMask = XprType::IsVectorAtCompileTime ? LinearAccessBit : 0,
+    Flags = (evaluator<ArgTypeNestedCleaned>::Flags & (HereditaryBits|LinearAccessMask) & ~RowMajorBit) | (traits<XprType>::Flags & RowMajorBit),
+    
+    Alignment = evaluator<ArgTypeNestedCleaned>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& replicate)
+    : m_arg(replicate.nestedExpression()),
+      m_argImpl(m_arg),
+      m_rows(replicate.nestedExpression().rows()),
+      m_cols(replicate.nestedExpression().cols())
+  {}
+ 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    // try to avoid using modulo; this is a pure optimization strategy
+    const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
+                           : RowFactor==1 ? row
+                           : row % m_rows.value();
+    const Index actual_col = internal::traits<XprType>::ColsAtCompileTime==1 ? 0
+                           : ColFactor==1 ? col
+                           : col % m_cols.value();
+    
+    return m_argImpl.coeff(actual_row, actual_col);
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    // try to avoid using modulo; this is a pure optimization strategy
+    const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
+                                  ? (ColFactor==1 ?  index : index%m_cols.value())
+                                  : (RowFactor==1 ?  index : index%m_rows.value());
+    
+    return m_argImpl.coeff(actual_index);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    const Index actual_row = internal::traits<XprType>::RowsAtCompileTime==1 ? 0
+                           : RowFactor==1 ? row
+                           : row % m_rows.value();
+    const Index actual_col = internal::traits<XprType>::ColsAtCompileTime==1 ? 0
+                           : ColFactor==1 ? col
+                           : col % m_cols.value();
+
+    return m_argImpl.template packet<LoadMode,PacketType>(actual_row, actual_col);
+  }
+  
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    const Index actual_index = internal::traits<XprType>::RowsAtCompileTime==1
+                                  ? (ColFactor==1 ?  index : index%m_cols.value())
+                                  : (RowFactor==1 ?  index : index%m_rows.value());
+
+    return m_argImpl.template packet<LoadMode,PacketType>(actual_index);
+  }
+ 
+protected:
+  const ArgTypeNested m_arg;
+  evaluator<ArgTypeNestedCleaned> m_argImpl;
+  const variable_if_dynamic<Index, ArgType::RowsAtCompileTime> m_rows;
+  const variable_if_dynamic<Index, ArgType::ColsAtCompileTime> m_cols;
+};
+
+
+// -------------------- PartialReduxExpr --------------------
+
+template< typename ArgType, typename MemberOp, int Direction>
+struct evaluator<PartialReduxExpr<ArgType, MemberOp, Direction> >
+  : evaluator_base<PartialReduxExpr<ArgType, MemberOp, Direction> >
+{
+  typedef PartialReduxExpr<ArgType, MemberOp, Direction> XprType;
+  typedef typename internal::nested_eval<ArgType,1>::type ArgTypeNested;
+  typedef typename internal::remove_all<ArgTypeNested>::type ArgTypeNestedCleaned;
+  typedef typename ArgType::Scalar InputScalar;
+  typedef typename XprType::Scalar Scalar;
+  enum {
+    TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) :  int(ArgType::ColsAtCompileTime)
+  };
+  typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
+  enum {
+    CoeffReadCost = TraversalSize==Dynamic ? HugeCost
+                  : TraversalSize * evaluator<ArgType>::CoeffReadCost + int(CostOpType::value),
+    
+    Flags = (traits<XprType>::Flags&RowMajorBit) | (evaluator<ArgType>::Flags&(HereditaryBits&(~RowMajorBit))) | LinearAccessBit,
+    
+    Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr)
+    : m_arg(xpr.nestedExpression()), m_functor(xpr.functor())
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : int(CostOpType::value));
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index i, Index j) const
+  {
+    if (Direction==Vertical)
+      return m_functor(m_arg.col(j));
+    else
+      return m_functor(m_arg.row(i));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar coeff(Index index) const
+  {
+    if (Direction==Vertical)
+      return m_functor(m_arg.col(index));
+    else
+      return m_functor(m_arg.row(index));
+  }
+
+protected:
+  typename internal::add_const_on_value_type<ArgTypeNested>::type m_arg;
+  const MemberOp m_functor;
+};
+
+
+// -------------------- MatrixWrapper and ArrayWrapper --------------------
+//
+// evaluator_wrapper_base<T> is a common base class for the
+// MatrixWrapper and ArrayWrapper evaluators.
+
+template<typename XprType>
+struct evaluator_wrapper_base
+  : evaluator_base<XprType>
+{
+  typedef typename remove_all<typename XprType::NestedExpressionType>::type ArgType;
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    Flags = evaluator<ArgType>::Flags,
+    Alignment = evaluator<ArgType>::Alignment
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator_wrapper_base(const ArgType& arg) : m_argImpl(arg) {}
+
+  typedef typename ArgType::Scalar Scalar;
+  typedef typename ArgType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_argImpl.coeff(row, col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(index);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_argImpl.coeffRef(row, col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(index);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(row, col);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    return m_argImpl.template packet<LoadMode,PacketType>(index);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode>(row, col, x);
+  }
+
+  template<int StoreMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    m_argImpl.template writePacket<StoreMode>(index, x);
+  }
+
+protected:
+  evaluator<ArgType> m_argImpl;
+};
+
+template<typename TArgType>
+struct unary_evaluator<MatrixWrapper<TArgType> >
+  : evaluator_wrapper_base<MatrixWrapper<TArgType> >
+{
+  typedef MatrixWrapper<TArgType> XprType;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
+    : evaluator_wrapper_base<MatrixWrapper<TArgType> >(wrapper.nestedExpression())
+  { }
+};
+
+template<typename TArgType>
+struct unary_evaluator<ArrayWrapper<TArgType> >
+  : evaluator_wrapper_base<ArrayWrapper<TArgType> >
+{
+  typedef ArrayWrapper<TArgType> XprType;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& wrapper)
+    : evaluator_wrapper_base<ArrayWrapper<TArgType> >(wrapper.nestedExpression())
+  { }
+};
+
+
+// -------------------- Reverse --------------------
+
+// defined in Reverse.h:
+template<typename PacketType, bool ReversePacket> struct reverse_packet_cond;
+
+template<typename ArgType, int Direction>
+struct unary_evaluator<Reverse<ArgType, Direction> >
+  : evaluator_base<Reverse<ArgType, Direction> >
+{
+  typedef Reverse<ArgType, Direction> XprType;
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  enum {
+    IsRowMajor = XprType::IsRowMajor,
+    IsColMajor = !IsRowMajor,
+    ReverseRow = (Direction == Vertical)   || (Direction == BothDirections),
+    ReverseCol = (Direction == Horizontal) || (Direction == BothDirections),
+    ReversePacket = (Direction == BothDirections)
+                    || ((Direction == Vertical)   && IsColMajor)
+                    || ((Direction == Horizontal) && IsRowMajor),
+                    
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    
+    // let's enable LinearAccess only with vectorization because of the product overhead
+    // FIXME enable DirectAccess with negative strides?
+    Flags0 = evaluator<ArgType>::Flags,
+    LinearAccess = ( (Direction==BothDirections) && (int(Flags0)&PacketAccessBit) )
+                  || ((ReverseRow && XprType::ColsAtCompileTime==1) || (ReverseCol && XprType::RowsAtCompileTime==1))
+                 ? LinearAccessBit : 0,
+
+    Flags = int(Flags0) & (HereditaryBits | PacketAccessBit | LinearAccess),
+    
+    Alignment = 0 // FIXME in some rare cases, Alignment could be preserved, like a Vector4f.
+  };
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& reverse)
+    : m_argImpl(reverse.nestedExpression()),
+      m_rows(ReverseRow ? reverse.nestedExpression().rows() : 1),
+      m_cols(ReverseCol ? reverse.nestedExpression().cols() : 1)
+  { }
+ 
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index col) const
+  {
+    return m_argImpl.coeff(ReverseRow ? m_rows.value() - row - 1 : row,
+                           ReverseCol ? m_cols.value() - col - 1 : col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(m_rows.value() * m_cols.value() - index - 1);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index col)
+  {
+    return m_argImpl.coeffRef(ReverseRow ? m_rows.value() - row - 1 : row,
+                              ReverseCol ? m_cols.value() - col - 1 : col);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(m_rows.value() * m_cols.value() - index - 1);
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index row, Index col) const
+  {
+    enum {
+      PacketSize = unpacket_traits<PacketType>::size,
+      OffsetRow  = ReverseRow && IsColMajor ? PacketSize : 1,
+      OffsetCol  = ReverseCol && IsRowMajor ? PacketSize : 1
+    };
+    typedef internal::reverse_packet_cond<PacketType,ReversePacket> reverse_packet;
+    return reverse_packet::run(m_argImpl.template packet<LoadMode,PacketType>(
+                                  ReverseRow ? m_rows.value() - row - OffsetRow : row,
+                                  ReverseCol ? m_cols.value() - col - OffsetCol : col));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  PacketType packet(Index index) const
+  {
+    enum { PacketSize = unpacket_traits<PacketType>::size };
+    return preverse(m_argImpl.template packet<LoadMode,PacketType>(m_rows.value() * m_cols.value() - index - PacketSize));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index row, Index col, const PacketType& x)
+  {
+    // FIXME we could factorize some code with packet(i,j)
+    enum {
+      PacketSize = unpacket_traits<PacketType>::size,
+      OffsetRow  = ReverseRow && IsColMajor ? PacketSize : 1,
+      OffsetCol  = ReverseCol && IsRowMajor ? PacketSize : 1
+    };
+    typedef internal::reverse_packet_cond<PacketType,ReversePacket> reverse_packet;
+    m_argImpl.template writePacket<LoadMode>(
+                                  ReverseRow ? m_rows.value() - row - OffsetRow : row,
+                                  ReverseCol ? m_cols.value() - col - OffsetCol : col,
+                                  reverse_packet::run(x));
+  }
+
+  template<int LoadMode, typename PacketType>
+  EIGEN_STRONG_INLINE
+  void writePacket(Index index, const PacketType& x)
+  {
+    enum { PacketSize = unpacket_traits<PacketType>::size };
+    m_argImpl.template writePacket<LoadMode>
+      (m_rows.value() * m_cols.value() - index - PacketSize, preverse(x));
+  }
+ 
+protected:
+  evaluator<ArgType> m_argImpl;
+
+  // If we do not reverse rows, then we do not need to know the number of rows; same for columns
+  // Nonetheless, in this case it is important to set to 1 such that the coeff(index) method works fine for vectors.
+  const variable_if_dynamic<Index, ReverseRow ? ArgType::RowsAtCompileTime : 1> m_rows;
+  const variable_if_dynamic<Index, ReverseCol ? ArgType::ColsAtCompileTime : 1> m_cols;
+};
+
+
+// -------------------- Diagonal --------------------
+
+template<typename ArgType, int DiagIndex>
+struct evaluator<Diagonal<ArgType, DiagIndex> >
+  : evaluator_base<Diagonal<ArgType, DiagIndex> >
+{
+  typedef Diagonal<ArgType, DiagIndex> XprType;
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    
+    Flags = (unsigned int)(evaluator<ArgType>::Flags & (HereditaryBits | DirectAccessBit) & ~RowMajorBit) | LinearAccessBit,
+    
+    Alignment = 0
+  };
+
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& diagonal)
+    : m_argImpl(diagonal.nestedExpression()),
+      m_index(diagonal.index())
+  { }
+ 
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index row, Index) const
+  {
+    return m_argImpl.coeff(row + rowOffset(), row + colOffset());
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  CoeffReturnType coeff(Index index) const
+  {
+    return m_argImpl.coeff(index + rowOffset(), index + colOffset());
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index row, Index)
+  {
+    return m_argImpl.coeffRef(row + rowOffset(), row + colOffset());
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  Scalar& coeffRef(Index index)
+  {
+    return m_argImpl.coeffRef(index + rowOffset(), index + colOffset());
+  }
+
+protected:
+  evaluator<ArgType> m_argImpl;
+  const internal::variable_if_dynamicindex<Index, XprType::DiagIndex> m_index;
+
+private:
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value() > 0 ? 0 : -m_index.value(); }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value() > 0 ? m_index.value() : 0; }
+};
+
+
+//----------------------------------------------------------------------
+// deprecated code
+//----------------------------------------------------------------------
+
+// -------------------- EvalToTemp --------------------
+
+// expression class for evaluating nested expression to a temporary
+
+template<typename ArgType> class EvalToTemp;
+
+template<typename ArgType>
+struct traits<EvalToTemp<ArgType> >
+  : public traits<ArgType>
+{ };
+
+template<typename ArgType>
+class EvalToTemp
+  : public dense_xpr_base<EvalToTemp<ArgType> >::type
+{
+ public:
+ 
+  typedef typename dense_xpr_base<EvalToTemp>::type Base;
+  EIGEN_GENERIC_PUBLIC_INTERFACE(EvalToTemp)
+ 
+  explicit EvalToTemp(const ArgType& arg)
+    : m_arg(arg)
+  { }
+ 
+  const ArgType& arg() const
+  {
+    return m_arg;
+  }
+
+  Index rows() const 
+  {
+    return m_arg.rows();
+  }
+
+  Index cols() const 
+  {
+    return m_arg.cols();
+  }
+
+ private:
+  const ArgType& m_arg;
+};
+ 
+template<typename ArgType>
+struct evaluator<EvalToTemp<ArgType> >
+  : public evaluator<typename ArgType::PlainObject>
+{
+  typedef EvalToTemp<ArgType>                   XprType;
+  typedef typename ArgType::PlainObject         PlainObject;
+  typedef evaluator<PlainObject> Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
+    : m_result(xpr.arg())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+  }
+
+  // This constructor is used when nesting an EvalTo evaluator in another evaluator
+  EIGEN_DEVICE_FUNC evaluator(const ArgType& arg)
+    : m_result(arg)
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+} // namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COREEVALUATORS_H
diff --git a/Eigen/src/Core/CoreIterators.h b/Eigen/src/Core/CoreIterators.h
index 6da4683..4eb42b9 100644
--- a/Eigen/src/Core/CoreIterators.h
+++ b/Eigen/src/Core/CoreIterators.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -15,47 +15,113 @@
 /* This file contains the respective InnerIterator definition of the expressions defined in Eigen/Core
  */
 
-/** \ingroup SparseCore_Module
-  * \class InnerIterator
-  * \brief An InnerIterator allows to loop over the element of a sparse (or dense) matrix or expression
-  *
-  * todo
+namespace internal {
+
+template<typename XprType, typename EvaluatorKind>
+class inner_iterator_selector;
+
+}
+
+/** \class InnerIterator
+  * \brief An InnerIterator allows to loop over the element of any matrix expression.
+  * 
+  * \warning To be used with care because an evaluator is constructed every time an InnerIterator iterator is constructed.
+  * 
+  * TODO: add a usage example
   */
-
-// generic version for dense matrix and expressions
-template<typename Derived> class DenseBase<Derived>::InnerIterator
+template<typename XprType>
+class InnerIterator
 {
-  protected:
-    typedef typename Derived::Scalar Scalar;
-    typedef typename Derived::Index Index;
-
-    enum { IsRowMajor = (Derived::Flags&RowMajorBit)==RowMajorBit };
-  public:
-    EIGEN_STRONG_INLINE InnerIterator(const Derived& expr, Index outer)
-      : m_expression(expr), m_inner(0), m_outer(outer), m_end(expr.innerSize())
-    {}
-
-    EIGEN_STRONG_INLINE Scalar value() const
-    {
-      return (IsRowMajor) ? m_expression.coeff(m_outer, m_inner)
-                          : m_expression.coeff(m_inner, m_outer);
-    }
-
-    EIGEN_STRONG_INLINE InnerIterator& operator++() { m_inner++; return *this; }
-
-    EIGEN_STRONG_INLINE Index index() const { return m_inner; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
-
-  protected:
-    const Derived& m_expression;
-    Index m_inner;
-    const Index m_outer;
-    const Index m_end;
+protected:
+  typedef internal::inner_iterator_selector<XprType, typename internal::evaluator_traits<XprType>::Kind> IteratorType;
+  typedef internal::evaluator<XprType> EvaluatorType;
+  typedef typename internal::traits<XprType>::Scalar Scalar;
+public:
+  /** Construct an iterator over the \a outerId -th row or column of \a xpr */
+  InnerIterator(const XprType &xpr, const Index &outerId)
+    : m_eval(xpr), m_iter(m_eval, outerId, xpr.innerSize())
+  {}
+  
+  /// \returns the value of the current coefficient.
+  EIGEN_STRONG_INLINE Scalar value() const          { return m_iter.value(); }
+  /** Increment the iterator \c *this to the next non-zero coefficient.
+    * Explicit zeros are not skipped over. To skip explicit zeros, see class SparseView
+    */
+  EIGEN_STRONG_INLINE InnerIterator& operator++()   { m_iter.operator++(); return *this; }
+  /// \returns the column or row index of the current coefficient.
+  EIGEN_STRONG_INLINE Index index() const           { return m_iter.index(); }
+  /// \returns the row index of the current coefficient.
+  EIGEN_STRONG_INLINE Index row() const             { return m_iter.row(); }
+  /// \returns the column index of the current coefficient.
+  EIGEN_STRONG_INLINE Index col() const             { return m_iter.col(); }
+  /// \returns \c true if the iterator \c *this still references a valid coefficient.
+  EIGEN_STRONG_INLINE operator bool() const         { return m_iter; }
+  
+protected:
+  EvaluatorType m_eval;
+  IteratorType m_iter;
+private:
+  // If you get here, then you're not using the right InnerIterator type, e.g.:
+  //   SparseMatrix<double,RowMajor> A;
+  //   SparseMatrix<double>::InnerIterator it(A,0);
+  template<typename T> InnerIterator(const EigenBase<T>&,Index outer);
 };
 
+namespace internal {
+
+// Generic inner iterator implementation for dense objects
+template<typename XprType>
+class inner_iterator_selector<XprType, IndexBased>
+{
+protected:
+  typedef evaluator<XprType> EvaluatorType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit };
+  
+public:
+  EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &innerSize)
+    : m_eval(eval), m_inner(0), m_outer(outerId), m_end(innerSize)
+  {}
+
+  EIGEN_STRONG_INLINE Scalar value() const
+  {
+    return (IsRowMajor) ? m_eval.coeff(m_outer, m_inner)
+                        : m_eval.coeff(m_inner, m_outer);
+  }
+
+  EIGEN_STRONG_INLINE inner_iterator_selector& operator++() { m_inner++; return *this; }
+
+  EIGEN_STRONG_INLINE Index index() const { return m_inner; }
+  inline Index row() const { return IsRowMajor ? m_outer : index(); }
+  inline Index col() const { return IsRowMajor ? index() : m_outer; }
+
+  EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
+
+protected:
+  const EvaluatorType& m_eval;
+  Index m_inner;
+  const Index m_outer;
+  const Index m_end;
+};
+
+// For iterator-based evaluator, inner-iterator is already implemented as
+// evaluator<>::InnerIterator
+template<typename XprType>
+class inner_iterator_selector<XprType, IteratorBased>
+ : public evaluator<XprType>::InnerIterator
+{
+protected:
+  typedef typename evaluator<XprType>::InnerIterator Base;
+  typedef evaluator<XprType> EvaluatorType;
+  
+public:
+  EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &/*innerSize*/)
+    : Base(eval, outerId)
+  {}  
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_COREITERATORS_H
diff --git a/Eigen/src/Core/CwiseBinaryOp.h b/Eigen/src/Core/CwiseBinaryOp.h
index 586f77a..a36765e 100644
--- a/Eigen/src/Core/CwiseBinaryOp.h
+++ b/Eigen/src/Core/CwiseBinaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -13,26 +13,6 @@
 
 namespace Eigen {
 
-/** \class CwiseBinaryOp
-  * \ingroup Core_Module
-  *
-  * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
-  *
-  * \param BinaryOp template functor implementing the operator
-  * \param Lhs the type of the left-hand side
-  * \param Rhs the type of the right-hand side
-  *
-  * This class represents an expression  where a coefficient-wise binary operator is applied to two expressions.
-  * It is the return type of binary operators, by which we mean only those binary operators where
-  * both the left-hand side and the right-hand side are Eigen expressions.
-  * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
-  *
-  * Most of the time, this is the only way that it is used, so you typically don't have to name
-  * CwiseBinaryOp types explicitly.
-  *
-  * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
-  */
-
 namespace internal {
 template<typename BinaryOp, typename Lhs, typename Rhs>
 struct traits<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
@@ -52,76 +32,75 @@
   // we still want to handle the case when the result type is different.
   typedef typename result_of<
                      BinaryOp(
-                       typename Lhs::Scalar,
-                       typename Rhs::Scalar
+                       const typename Lhs::Scalar&,
+                       const typename Rhs::Scalar&
                      )
                    >::type Scalar;
-  typedef typename promote_storage_type<typename traits<Lhs>::StorageKind,
-                                           typename traits<Rhs>::StorageKind>::ret StorageKind;
-  typedef typename promote_index_type<typename traits<Lhs>::Index,
-                                         typename traits<Rhs>::Index>::type Index;
+  typedef typename cwise_promote_storage_type<typename traits<Lhs>::StorageKind,
+                                              typename traits<Rhs>::StorageKind,
+                                              BinaryOp>::ret StorageKind;
+  typedef typename promote_index_type<typename traits<Lhs>::StorageIndex,
+                                      typename traits<Rhs>::StorageIndex>::type StorageIndex;
   typedef typename Lhs::Nested LhsNested;
   typedef typename Rhs::Nested RhsNested;
   typedef typename remove_reference<LhsNested>::type _LhsNested;
   typedef typename remove_reference<RhsNested>::type _RhsNested;
   enum {
-    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-    SameType = is_same<typename _LhsNested::Scalar,typename _RhsNested::Scalar>::value,
-    StorageOrdersAgree = (int(Lhs::Flags)&RowMajorBit)==(int(Rhs::Flags)&RowMajorBit),
-    Flags0 = (int(LhsFlags) | int(RhsFlags)) & (
-        HereditaryBits
-      | (int(LhsFlags) & int(RhsFlags) &
-           ( AlignedBit
-           | (StorageOrdersAgree ? LinearAccessBit : 0)
-           | (functor_traits<BinaryOp>::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0)
-           )
-        )
-     ),
-    Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit),
-    CoeffReadCost = LhsCoeffReadCost + RhsCoeffReadCost + functor_traits<BinaryOp>::Cost
+    Flags = cwise_promote_storage_order<typename traits<Lhs>::StorageKind,typename traits<Rhs>::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value
   };
 };
 } // end namespace internal
 
-// we require Lhs and Rhs to have the same scalar type. Currently there is no example of a binary functor
-// that would take two operands of different types. If there were such an example, then this check should be
-// moved to the BinaryOp functors, on a per-case basis. This would however require a change in the BinaryOp functors, as
-// currently they take only one typename Scalar template parameter.
-// It is tempting to always allow mixing different types but remember that this is often impossible in the vectorized paths.
-// So allowing mixing different types gives very unexpected errors when enabling vectorization, when the user tries to
-// add together a float matrix and a double matrix.
-#define EIGEN_CHECK_BINARY_COMPATIBILIY(BINOP,LHS,RHS) \
-  EIGEN_STATIC_ASSERT((internal::functor_is_product_like<BINOP>::ret \
-                        ? int(internal::scalar_product_traits<LHS, RHS>::Defined) \
-                        : int(internal::is_same<LHS, RHS>::value)), \
-    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
 template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
 class CwiseBinaryOpImpl;
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOp : internal::no_assignment_operator,
+/** \class CwiseBinaryOp
+  * \ingroup Core_Module
+  *
+  * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions
+  *
+  * \tparam BinaryOp template functor implementing the operator
+  * \tparam LhsType the type of the left-hand side
+  * \tparam RhsType the type of the right-hand side
+  *
+  * This class represents an expression  where a coefficient-wise binary operator is applied to two expressions.
+  * It is the return type of binary operators, by which we mean only those binary operators where
+  * both the left-hand side and the right-hand side are Eigen expressions.
+  * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically don't have to name
+  * CwiseBinaryOp types explicitly.
+  *
+  * \sa MatrixBase::binaryExpr(const MatrixBase<OtherDerived> &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp
+  */
+template<typename BinaryOp, typename LhsType, typename RhsType>
+class CwiseBinaryOp : 
   public CwiseBinaryOpImpl<
-          BinaryOp, Lhs, Rhs,
-          typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
-                                           typename internal::traits<Rhs>::StorageKind>::ret>
+          BinaryOp, LhsType, RhsType,
+          typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
+                                                        typename internal::traits<RhsType>::StorageKind,
+                                                        BinaryOp>::ret>,
+  internal::no_assignment_operator
 {
   public:
+    
+    typedef typename internal::remove_all<BinaryOp>::type Functor;
+    typedef typename internal::remove_all<LhsType>::type Lhs;
+    typedef typename internal::remove_all<RhsType>::type Rhs;
 
     typedef typename CwiseBinaryOpImpl<
-        BinaryOp, Lhs, Rhs,
-        typename internal::promote_storage_type<typename internal::traits<Lhs>::StorageKind,
-                                         typename internal::traits<Rhs>::StorageKind>::ret>::Base Base;
+        BinaryOp, LhsType, RhsType,
+        typename internal::cwise_promote_storage_type<typename internal::traits<LhsType>::StorageKind,
+                                                      typename internal::traits<Rhs>::StorageKind,
+                                                      BinaryOp>::ret>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseBinaryOp)
 
-    typedef typename internal::nested<Lhs>::type LhsNested;
-    typedef typename internal::nested<Rhs>::type RhsNested;
+    typedef typename internal::ref_selector<LhsType>::type LhsNested;
+    typedef typename internal::ref_selector<RhsType>::type RhsNested;
     typedef typename internal::remove_reference<LhsNested>::type _LhsNested;
     typedef typename internal::remove_reference<RhsNested>::type _RhsNested;
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp())
       : m_lhs(aLhs), m_rhs(aRhs), m_functor(func)
     {
@@ -131,6 +110,7 @@
       eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols());
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const {
       // return the fixed size type if available to enable compile time optimizations
       if (internal::traits<typename internal::remove_all<LhsNested>::type>::RowsAtCompileTime==Dynamic)
@@ -138,6 +118,7 @@
       else
         return m_lhs.rows();
     }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const {
       // return the fixed size type if available to enable compile time optimizations
       if (internal::traits<typename internal::remove_all<LhsNested>::type>::ColsAtCompileTime==Dynamic)
@@ -147,10 +128,13 @@
     }
 
     /** \returns the left hand side nested expression */
+    EIGEN_DEVICE_FUNC
     const _LhsNested& lhs() const { return m_lhs; }
     /** \returns the right hand side nested expression */
+    EIGEN_DEVICE_FUNC
     const _RhsNested& rhs() const { return m_rhs; }
     /** \returns the functor representing the binary operation */
+    EIGEN_DEVICE_FUNC
     const BinaryOp& functor() const { return m_functor; }
 
   protected:
@@ -159,41 +143,13 @@
     const BinaryOp m_functor;
 };
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Dense>
-  : public internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
+// Generic API dispatcher
+template<typename BinaryOp, typename Lhs, typename Rhs, typename StorageKind>
+class CwiseBinaryOpImpl
+  : public internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
 {
-    typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> Derived;
-  public:
-
-    typedef typename internal::dense_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE( Derived )
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
-    {
-      return derived().functor()(derived().lhs().coeff(rowId, colId),
-                                 derived().rhs().coeff(rowId, colId));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
-    {
-      return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(rowId, colId),
-                                          derived().rhs().template packet<LoadMode>(rowId, colId));
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return derived().functor()(derived().lhs().coeff(index),
-                                 derived().rhs().coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return derived().functor().packetOp(derived().lhs().template packet<LoadMode>(index),
-                                          derived().rhs().template packet<LoadMode>(index));
-    }
+public:
+  typedef typename internal::generic_xpr_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type Base;
 };
 
 /** replaces \c *this by \c *this - \a other.
@@ -205,8 +161,7 @@
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator-=(const MatrixBase<OtherDerived> &other)
 {
-  SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
@@ -219,11 +174,11 @@
 EIGEN_STRONG_INLINE Derived &
 MatrixBase<Derived>::operator+=(const MatrixBase<OtherDerived>& other)
 {
-  SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, Derived, OtherDerived> tmp(derived());
-  tmp = other.derived();
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
 } // end namespace Eigen
 
 #endif // EIGEN_CWISE_BINARY_OP_H
+
diff --git a/Eigen/src/Core/CwiseNullaryOp.h b/Eigen/src/Core/CwiseNullaryOp.h
index a93bab2..ddd607e 100644
--- a/Eigen/src/Core/CwiseNullaryOp.h
+++ b/Eigen/src/Core/CwiseNullaryOp.h
@@ -12,13 +12,24 @@
 
 namespace Eigen {
 
+namespace internal {
+template<typename NullaryOp, typename PlainObjectType>
+struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
+{
+  enum {
+    Flags = traits<PlainObjectType>::Flags & RowMajorBit
+  };
+};
+
+} // namespace internal
+
 /** \class CwiseNullaryOp
   * \ingroup Core_Module
   *
   * \brief Generic expression of a matrix where all coefficients are defined by a functor
   *
-  * \param NullaryOp template functor implementing the operator
-  * \param PlainObjectType the underlying plain matrix/array type
+  * \tparam NullaryOp template functor implementing the operator
+  * \tparam PlainObjectType the underlying plain matrix/array type
   *
   * This class represents an expression of a generic nullary operator.
   * It is the return type of the Ones(), Zero(), Constant(), Identity() and Random() methods,
@@ -27,68 +38,49 @@
   * However, if you want to write a function returning such an expression, you
   * will need to use this class.
   *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr()
+  * The functor NullaryOp must expose one of the following method:
+    <table class="manual">
+    <tr            ><td>\c operator()() </td><td>if the procedural generation does not depend on the coefficient entries (e.g., random numbers)</td></tr>
+    <tr class="alt"><td>\c operator()(Index i)</td><td>if the procedural generation makes sense for vectors only and that it depends on the coefficient index \c i (e.g., linspace) </td></tr>
+    <tr            ><td>\c operator()(Index i,Index j)</td><td>if the procedural generation depends on the matrix coordinates \c i, \c j (e.g., to generate a checkerboard with 0 and 1)</td></tr>
+    </table>
+  * It is also possible to expose the last two operators if the generation makes sense for matrices but can be optimized for vectors.
+  *
+  * See DenseBase::NullaryExpr(Index,const CustomNullaryOp&) for an example binding
+  * C++11 random number generators.
+  *
+  * A nullary expression can also be used to implement custom sophisticated matrix manipulations
+  * that cannot be covered by the existing set of natively supported matrix manipulations.
+  * See this \ref TopicCustomizing_NullaryExpr "page" for some examples and additional explanations
+  * on the behavior of CwiseNullaryOp.
+  *
+  * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr
   */
-
-namespace internal {
 template<typename NullaryOp, typename PlainObjectType>
-struct traits<CwiseNullaryOp<NullaryOp, PlainObjectType> > : traits<PlainObjectType>
-{
-  enum {
-    Flags = (traits<PlainObjectType>::Flags
-      & (  HereditaryBits
-         | (functor_has_linear_access<NullaryOp>::ret ? LinearAccessBit : 0)
-         | (functor_traits<NullaryOp>::PacketAccess ? PacketAccessBit : 0)))
-      | (functor_traits<NullaryOp>::IsRepeatable ? 0 : EvalBeforeNestingBit),
-    CoeffReadCost = functor_traits<NullaryOp>::Cost
-  };
-};
-}
-
-template<typename NullaryOp, typename PlainObjectType>
-class CwiseNullaryOp : internal::no_assignment_operator,
-  public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type
+class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp<NullaryOp, PlainObjectType> >::type, internal::no_assignment_operator
 {
   public:
 
     typedef typename internal::dense_xpr_base<CwiseNullaryOp>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(CwiseNullaryOp)
 
-    CwiseNullaryOp(Index nbRows, Index nbCols, const NullaryOp& func = NullaryOp())
-      : m_rows(nbRows), m_cols(nbCols), m_functor(func)
+    EIGEN_DEVICE_FUNC
+    CwiseNullaryOp(Index rows, Index cols, const NullaryOp& func = NullaryOp())
+      : m_rows(rows), m_cols(cols), m_functor(func)
     {
-      eigen_assert(nbRows >= 0
-            && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows)
-            &&  nbCols >= 0
-            && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols));
+      eigen_assert(rows >= 0
+            && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
+            &&  cols >= 0
+            && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols));
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const { return m_rows.value(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); }
 
-    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
-    {
-      return m_functor(rowId, colId);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
-    {
-      return m_functor.packetOp(rowId, colId);
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return m_functor(index);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return m_functor.packetOp(index);
-    }
-
     /** \returns the functor representing the nullary operation */
+    EIGEN_DEVICE_FUNC
     const NullaryOp& functor() const { return m_functor; }
 
   protected:
@@ -113,10 +105,10 @@
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, Derived>(rows, cols, func);
+  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(rows, cols, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
@@ -132,16 +124,19 @@
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
   *
+  * Here is an example with C++11 random generators: \include random_cpp11.cpp
+  * Output: \verbinclude random_cpp11.out
+  * 
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(Index size, const CustomNullaryOp& func)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, Derived>(1, size, func);
-  else return CwiseNullaryOp<CustomNullaryOp, Derived>(size, 1, func);
+  if(RowsAtCompileTime == 1) return CwiseNullaryOp<CustomNullaryOp, PlainObject>(1, size, func);
+  else return CwiseNullaryOp<CustomNullaryOp, PlainObject>(size, 1, func);
 }
 
 /** \returns an expression of a matrix defined by a custom functor \a func
@@ -155,19 +150,19 @@
   */
 template<typename Derived>
 template<typename CustomNullaryOp>
-EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseNullaryOp<CustomNullaryOp, typename DenseBase<Derived>::PlainObject>
 DenseBase<Derived>::NullaryExpr(const CustomNullaryOp& func)
 {
-  return CwiseNullaryOp<CustomNullaryOp, Derived>(RowsAtCompileTime, ColsAtCompileTime, func);
+  return CwiseNullaryOp<CustomNullaryOp, PlainObject>(RowsAtCompileTime, ColsAtCompileTime, func);
 }
 
 /** \returns an expression of a constant matrix of value \a value
   *
-  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
+  * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this DenseBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
-  * it is redundant to pass \a nbRows and \a nbCols as arguments, so Zero() should be used
+  * it is redundant to pass \a rows and \a cols as arguments, so Zero() should be used
   * instead.
   *
   * The template parameter \a CustomNullaryOp is the type of the functor.
@@ -176,9 +171,9 @@
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Constant(Index nbRows, Index nbCols, const Scalar& value)
+DenseBase<Derived>::Constant(Index rows, Index cols, const Scalar& value)
 {
-  return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_constant_op<Scalar>(value));
+  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_constant_op<Scalar>(value));
 }
 
 /** \returns an expression of a constant matrix of value \a value
@@ -197,7 +192,7 @@
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(Index size, const Scalar& value)
 {
   return DenseBase<Derived>::NullaryExpr(size, internal::scalar_constant_op<Scalar>(value));
@@ -213,53 +208,40 @@
   * \sa class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Constant(const Scalar& value)
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
   return DenseBase<Derived>::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_constant_op<Scalar>(value));
 }
 
-/**
-  * \brief Sets a linearly space vector.
+/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(Index,const Scalar&,const Scalar&)
   *
-  * The function generates 'size' equally spaced values in the closed interval [low,high].
-  * This particular version of LinSpaced() uses sequential access, i.e. vector access is
-  * assumed to be a(0), a(1), ..., a(size). This assumption allows for better vectorization
-  * and yields faster code than the random access version.
-  *
-  * When size is set to 1, a vector of length 1 containing 'high' is returned.
-  *
-  * \only_for_vectors
-  *
-  * Example: \include DenseBase_LinSpaced_seq.cpp
-  * Output: \verbinclude DenseBase_LinSpaced_seq.out
-  *
-  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Index,Scalar,Scalar), CwiseNullaryOp
+  * \sa LinSpaced(Index,Scalar,Scalar), setLinSpaced(Index,const Scalar&,const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,false>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
 }
 
-/**
-  * \copydoc DenseBase::LinSpaced(Sequential_t, Index, const Scalar&, const Scalar&)
-  * Special version for fixed size types which does not require the size parameter.
+/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(const Scalar&,const Scalar&)
+  *
+  * \sa LinSpaced(Scalar,Scalar)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::SequentialLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,false>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
 }
 
 /**
-  * \brief Sets a linearly space vector.
+  * \brief Sets a linearly spaced vector.
   *
   * The function generates 'size' equally spaced values in the closed interval [low,high].
   * When size is set to 1, a vector of length 1 containing 'high' is returned.
@@ -269,14 +251,24 @@
   * Example: \include DenseBase_LinSpaced.cpp
   * Output: \verbinclude DenseBase_LinSpaced.out
   *
-  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Sequential_t,Index,const Scalar&,const Scalar&,Index), CwiseNullaryOp
+  * For integer scalar types, an even spacing is possible if and only if the length of the range,
+  * i.e., \c high-low is a scalar multiple of \c size-1, or if \c size is a scalar multiple of the
+  * number of values \c high-low+1 (meaning each value can be repeated the same number of time).
+  * If one of these two considions is not satisfied, then \c high is lowered to the largest value
+  * satisfying one of this constraint.
+  * Here are some examples:
+  *
+  * Example: \include DenseBase_LinSpacedInt.cpp
+  * Output: \verbinclude DenseBase_LinSpacedInt.out
+  *
+  * \sa setLinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(Index size, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,true>(low,high,size));
+  return DenseBase<Derived>::NullaryExpr(size, internal::linspaced_op<Scalar,PacketScalar>(low,high,size));
 }
 
 /**
@@ -284,22 +276,23 @@
   * Special version for fixed size types which does not require the size parameter.
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::RandomAccessLinSpacedReturnType
 DenseBase<Derived>::LinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,true>(low,high,Derived::SizeAtCompileTime));
+  return DenseBase<Derived>::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op<Scalar,PacketScalar>(low,high,Derived::SizeAtCompileTime));
 }
 
 /** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */
 template<typename Derived>
-bool DenseBase<Derived>::isApproxToConstant
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isApproxToConstant
 (const Scalar& val, const RealScalar& prec) const
 {
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
     for(Index i = 0; i < rows(); ++i)
-      if(!internal::isApprox(this->coeff(i, j), val, prec))
+      if(!internal::isApprox(self.coeff(i, j), val, prec))
         return false;
   return true;
 }
@@ -308,7 +301,7 @@
   *
   * \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */
 template<typename Derived>
-bool DenseBase<Derived>::isConstant
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isConstant
 (const Scalar& val, const RealScalar& prec) const
 {
   return isApproxToConstant(val, prec);
@@ -319,22 +312,22 @@
   * \sa setConstant(), Constant(), class CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void DenseBase<Derived>::fill(const Scalar& val)
 {
   setConstant(val);
 }
 
-/** Sets all coefficients in this expression to \a value.
+/** Sets all coefficients in this expression to value \a val.
   *
   * \sa fill(), setConstant(Index,const Scalar&), setConstant(Index,Index,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setConstant(const Scalar& val)
 {
   return derived() = Constant(rows(), cols(), val);
 }
 
-/** Resizes to the given \a size, and sets all coefficients in this expression to the given \a value.
+/** Resizes to the given \a size, and sets all coefficients in this expression to the given value \a val.
   *
   * \only_for_vectors
   *
@@ -344,17 +337,17 @@
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setConstant(Index size, const Scalar& val)
 {
   resize(size);
   return setConstant(val);
 }
 
-/** Resizes to the given size, and sets all coefficients in this expression to the given \a value.
+/** Resizes to the given size, and sets all coefficients in this expression to the given value \a val.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   * \param val the value to which all coefficients are set
   *
   * Example: \include Matrix_setConstant_int_int.cpp
@@ -363,15 +356,15 @@
   * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setConstant(Index nbRows, Index nbCols, const Scalar& val)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setConstant(Index rows, Index cols, const Scalar& val)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setConstant(val);
 }
 
 /**
-  * \brief Sets a linearly space vector.
+  * \brief Sets a linearly spaced vector.
   *
   * The function generates 'size' equally spaced values in the closed interval [low,high].
   * When size is set to 1, a vector of length 1 containing 'high' is returned.
@@ -381,27 +374,33 @@
   * Example: \include DenseBase_setLinSpaced.cpp
   * Output: \verbinclude DenseBase_setLinSpaced.out
   *
-  * \sa CwiseNullaryOp
+  * For integer scalar types, do not miss the explanations on the definition
+  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
+  *
+  * \sa LinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,false>(low,high,newSize));
+  return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op<Scalar,PacketScalar>(low,high,newSize));
 }
 
 /**
-  * \brief Sets a linearly space vector.
+  * \brief Sets a linearly spaced vector.
   *
-  * The function fill *this with equally spaced values in the closed interval [low,high].
+  * The function fills \c *this with equally spaced values in the closed interval [low,high].
   * When size is set to 1, a vector of length 1 containing 'high' is returned.
   *
   * \only_for_vectors
   *
-  * \sa setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
+  * For integer scalar types, do not miss the explanations on the definition
+  * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink.
+  *
+  * \sa LinSpaced(Index,const Scalar&,const Scalar&), setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setLinSpaced(const Scalar& low, const Scalar& high)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return setLinSpaced(size(), low, high);
@@ -424,10 +423,10 @@
   * \sa Zero(), Zero(Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Zero(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Zero(Index rows, Index cols)
 {
-  return Constant(nbRows, nbCols, Scalar(0));
+  return Constant(rows, cols, Scalar(0));
 }
 
 /** \returns an expression of a zero vector.
@@ -447,7 +446,7 @@
   * \sa Zero(), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero(Index size)
 {
   return Constant(size, Scalar(0));
@@ -464,7 +463,7 @@
   * \sa Zero(Index), Zero(Index,Index)
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Zero()
 {
   return Constant(Scalar(0));
@@ -479,11 +478,12 @@
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-bool DenseBase<Derived>::isZero(const RealScalar& prec) const
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isZero(const RealScalar& prec) const
 {
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
     for(Index i = 0; i < rows(); ++i)
-      if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<Scalar>(1), prec))
+      if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<Scalar>(1), prec))
         return false;
   return true;
 }
@@ -496,7 +496,7 @@
   * \sa class CwiseNullaryOp, Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setZero()
 {
   return setConstant(Scalar(0));
 }
@@ -511,7 +511,7 @@
   * \sa DenseBase::setZero(), setZero(Index,Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setZero(Index newSize)
 {
   resize(newSize);
@@ -520,8 +520,8 @@
 
 /** Resizes to the given size, and sets all coefficients in this expression to zero.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setZero_int_int.cpp
   * Output: \verbinclude Matrix_setZero_int_int.out
@@ -529,10 +529,10 @@
   * \sa DenseBase::setZero(), setZero(Index), class CwiseNullaryOp, DenseBase::Zero()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setZero(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setZero(Index rows, Index cols)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setConstant(Scalar(0));
 }
 
@@ -540,7 +540,7 @@
 
 /** \returns an expression of a matrix where all coefficients equal one.
   *
-  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
+  * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
@@ -553,10 +553,10 @@
   * \sa Ones(), Ones(Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
-DenseBase<Derived>::Ones(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+DenseBase<Derived>::Ones(Index rows, Index cols)
 {
-  return Constant(nbRows, nbCols, Scalar(1));
+  return Constant(rows, cols, Scalar(1));
 }
 
 /** \returns an expression of a vector where all coefficients equal one.
@@ -576,7 +576,7 @@
   * \sa Ones(), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones(Index newSize)
 {
   return Constant(newSize, Scalar(1));
@@ -593,7 +593,7 @@
   * \sa Ones(Index), Ones(Index,Index), isOnes(), class Ones
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase<Derived>::ConstantReturnType
 DenseBase<Derived>::Ones()
 {
   return Constant(Scalar(1));
@@ -608,7 +608,7 @@
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-bool DenseBase<Derived>::isOnes
+EIGEN_DEVICE_FUNC bool DenseBase<Derived>::isOnes
 (const RealScalar& prec) const
 {
   return isApproxToConstant(Scalar(1), prec);
@@ -622,7 +622,7 @@
   * \sa class CwiseNullaryOp, Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::setOnes()
 {
   return setConstant(Scalar(1));
 }
@@ -637,7 +637,7 @@
   * \sa MatrixBase::setOnes(), setOnes(Index,Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
 PlainObjectBase<Derived>::setOnes(Index newSize)
 {
   resize(newSize);
@@ -646,8 +646,8 @@
 
 /** Resizes to the given size, and sets all coefficients in this expression to one.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setOnes_int_int.cpp
   * Output: \verbinclude Matrix_setOnes_int_int.out
@@ -655,10 +655,10 @@
   * \sa MatrixBase::setOnes(), setOnes(Index), class CwiseNullaryOp, MatrixBase::Ones()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setOnes(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived&
+PlainObjectBase<Derived>::setOnes(Index rows, Index cols)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setConstant(Scalar(1));
 }
 
@@ -666,7 +666,7 @@
 
 /** \returns an expression of the identity matrix (not necessarily square).
   *
-  * The parameters \a nbRows and \a nbCols are the number of rows and of columns of
+  * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
@@ -679,10 +679,10 @@
   * \sa Identity(), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
-MatrixBase<Derived>::Identity(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+MatrixBase<Derived>::Identity(Index rows, Index cols)
 {
-  return DenseBase<Derived>::NullaryExpr(nbRows, nbCols, internal::scalar_identity_op<Scalar>());
+  return DenseBase<Derived>::NullaryExpr(rows, cols, internal::scalar_identity_op<Scalar>());
 }
 
 /** \returns an expression of the identity matrix (not necessarily square).
@@ -696,7 +696,7 @@
   * \sa Identity(Index,Index), setIdentity(), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::IdentityReturnType
 MatrixBase<Derived>::Identity()
 {
   EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
@@ -716,18 +716,19 @@
 bool MatrixBase<Derived>::isIdentity
 (const RealScalar& prec) const
 {
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index j = 0; j < cols(); ++j)
   {
     for(Index i = 0; i < rows(); ++i)
     {
       if(i == j)
       {
-        if(!internal::isApprox(this->coeff(i, j), static_cast<Scalar>(1), prec))
+        if(!internal::isApprox(self.coeff(i, j), static_cast<Scalar>(1), prec))
           return false;
       }
       else
       {
-        if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast<RealScalar>(1), prec))
+        if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast<RealScalar>(1), prec))
           return false;
       }
     }
@@ -740,6 +741,7 @@
 template<typename Derived, bool Big = (Derived::SizeAtCompileTime>=16)>
 struct setIdentity_impl
 {
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     return m = Derived::Identity(m.rows(), m.cols());
@@ -749,11 +751,11 @@
 template<typename Derived>
 struct setIdentity_impl<Derived, true>
 {
-  typedef typename Derived::Index Index;
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Derived& run(Derived& m)
   {
     m.setZero();
-    const Index size = (std::min)(m.rows(), m.cols());
+    const Index size = numext::mini(m.rows(), m.cols());
     for(Index i = 0; i < size; ++i) m.coeffRef(i,i) = typename Derived::Scalar(1);
     return m;
   }
@@ -769,15 +771,15 @@
   * \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), isIdentity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity()
 {
   return internal::setIdentity_impl<Derived>::run(derived());
 }
 
 /** \brief Resizes to the given size, and writes the identity expression (not necessarily square) into *this.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setIdentity_int_int.cpp
   * Output: \verbinclude Matrix_setIdentity_int_int.out
@@ -785,9 +787,9 @@
   * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Identity()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index nbRows, Index nbCols)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase<Derived>::setIdentity(Index rows, Index cols)
 {
-  derived().resize(nbRows, nbCols);
+  derived().resize(rows, cols);
   return setIdentity();
 }
 
@@ -798,7 +800,7 @@
   * \sa MatrixBase::Unit(Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index newSize, Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(newSize,newSize), i);
@@ -813,7 +815,7 @@
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::Unit(Index i)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return BasisReturnType(SquareMatrixType::Identity(),i);
@@ -826,7 +828,7 @@
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitX()
 { return Derived::Unit(0); }
 
 /** \returns an expression of the Y axis unit vector (0,1{,0}^*)
@@ -836,7 +838,7 @@
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitY()
 { return Derived::Unit(1); }
 
 /** \returns an expression of the Z axis unit vector (0,0,1{,0}^*)
@@ -846,7 +848,7 @@
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitZ()
 { return Derived::Unit(2); }
 
 /** \returns an expression of the W axis unit vector (0,0,0,1)
@@ -856,7 +858,7 @@
   * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW()
   */
 template<typename Derived>
-EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::BasisReturnType MatrixBase<Derived>::UnitW()
 { return Derived::Unit(3); }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/CwiseTernaryOp.h b/Eigen/src/Core/CwiseTernaryOp.h
new file mode 100644
index 0000000..9f3576f
--- /dev/null
+++ b/Eigen/src/Core/CwiseTernaryOp.h
@@ -0,0 +1,197 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2016 Eugene Brevdo <ebrevdo@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_CWISE_TERNARY_OP_H
+#define EIGEN_CWISE_TERNARY_OP_H
+
+namespace Eigen {
+
+namespace internal {
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>
+struct traits<CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> > {
+  // we must not inherit from traits<Arg1> since it has
+  // the potential to cause problems with MSVC
+  typedef typename remove_all<Arg1>::type Ancestor;
+  typedef typename traits<Ancestor>::XprKind XprKind;
+  enum {
+    RowsAtCompileTime = traits<Ancestor>::RowsAtCompileTime,
+    ColsAtCompileTime = traits<Ancestor>::ColsAtCompileTime,
+    MaxRowsAtCompileTime = traits<Ancestor>::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = traits<Ancestor>::MaxColsAtCompileTime
+  };
+
+  // even though we require Arg1, Arg2, and Arg3 to have the same scalar type
+  // (see CwiseTernaryOp constructor),
+  // we still want to handle the case when the result type is different.
+  typedef typename result_of<TernaryOp(
+      const typename Arg1::Scalar&, const typename Arg2::Scalar&,
+      const typename Arg3::Scalar&)>::type Scalar;
+
+  typedef typename internal::traits<Arg1>::StorageKind StorageKind;
+  typedef typename internal::traits<Arg1>::StorageIndex StorageIndex;
+
+  typedef typename Arg1::Nested Arg1Nested;
+  typedef typename Arg2::Nested Arg2Nested;
+  typedef typename Arg3::Nested Arg3Nested;
+  typedef typename remove_reference<Arg1Nested>::type _Arg1Nested;
+  typedef typename remove_reference<Arg2Nested>::type _Arg2Nested;
+  typedef typename remove_reference<Arg3Nested>::type _Arg3Nested;
+  enum { Flags = _Arg1Nested::Flags & RowMajorBit };
+};
+}  // end namespace internal
+
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
+          typename StorageKind>
+class CwiseTernaryOpImpl;
+
+/** \class CwiseTernaryOp
+  * \ingroup Core_Module
+  *
+  * \brief Generic expression where a coefficient-wise ternary operator is
+ * applied to two expressions
+  *
+  * \tparam TernaryOp template functor implementing the operator
+  * \tparam Arg1Type the type of the first argument
+  * \tparam Arg2Type the type of the second argument
+  * \tparam Arg3Type the type of the third argument
+  *
+  * This class represents an expression where a coefficient-wise ternary
+ * operator is applied to three expressions.
+  * It is the return type of ternary operators, by which we mean only those
+ * ternary operators where
+  * all three arguments are Eigen expressions.
+  * For example, the return type of betainc(matrix1, matrix2, matrix3) is a
+ * CwiseTernaryOp.
+  *
+  * Most of the time, this is the only way that it is used, so you typically
+ * don't have to name
+  * CwiseTernaryOp types explicitly.
+  *
+  * \sa MatrixBase::ternaryExpr(const MatrixBase<Argument2> &, const
+ * MatrixBase<Argument3> &, const CustomTernaryOp &) const, class CwiseBinaryOp,
+ * class CwiseUnaryOp, class CwiseNullaryOp
+  */
+template <typename TernaryOp, typename Arg1Type, typename Arg2Type,
+          typename Arg3Type>
+class CwiseTernaryOp : public CwiseTernaryOpImpl<
+                           TernaryOp, Arg1Type, Arg2Type, Arg3Type,
+                           typename internal::traits<Arg1Type>::StorageKind>,
+                       internal::no_assignment_operator
+{
+ public:
+  typedef typename internal::remove_all<Arg1Type>::type Arg1;
+  typedef typename internal::remove_all<Arg2Type>::type Arg2;
+  typedef typename internal::remove_all<Arg3Type>::type Arg3;
+
+  typedef typename CwiseTernaryOpImpl<
+      TernaryOp, Arg1Type, Arg2Type, Arg3Type,
+      typename internal::traits<Arg1Type>::StorageKind>::Base Base;
+  EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseTernaryOp)
+
+  typedef typename internal::ref_selector<Arg1Type>::type Arg1Nested;
+  typedef typename internal::ref_selector<Arg2Type>::type Arg2Nested;
+  typedef typename internal::ref_selector<Arg3Type>::type Arg3Nested;
+  typedef typename internal::remove_reference<Arg1Nested>::type _Arg1Nested;
+  typedef typename internal::remove_reference<Arg2Nested>::type _Arg2Nested;
+  typedef typename internal::remove_reference<Arg3Nested>::type _Arg3Nested;
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE CwiseTernaryOp(const Arg1& a1, const Arg2& a2,
+                                     const Arg3& a3,
+                                     const TernaryOp& func = TernaryOp())
+      : m_arg1(a1), m_arg2(a2), m_arg3(a3), m_functor(func) {
+    // require the sizes to match
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg2)
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg3)
+
+    // The index types should match
+    EIGEN_STATIC_ASSERT((internal::is_same<
+                         typename internal::traits<Arg1Type>::StorageKind,
+                         typename internal::traits<Arg2Type>::StorageKind>::value),
+                        STORAGE_KIND_MUST_MATCH)
+    EIGEN_STATIC_ASSERT((internal::is_same<
+                         typename internal::traits<Arg1Type>::StorageKind,
+                         typename internal::traits<Arg3Type>::StorageKind>::value),
+                        STORAGE_KIND_MUST_MATCH)
+
+    eigen_assert(a1.rows() == a2.rows() && a1.cols() == a2.cols() &&
+                 a1.rows() == a3.rows() && a1.cols() == a3.cols());
+  }
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE Index rows() const {
+    // return the fixed size type if available to enable compile time
+    // optimizations
+    if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                RowsAtCompileTime == Dynamic &&
+        internal::traits<typename internal::remove_all<Arg2Nested>::type>::
+                RowsAtCompileTime == Dynamic)
+      return m_arg3.rows();
+    else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                     RowsAtCompileTime == Dynamic &&
+             internal::traits<typename internal::remove_all<Arg3Nested>::type>::
+                     RowsAtCompileTime == Dynamic)
+      return m_arg2.rows();
+    else
+      return m_arg1.rows();
+  }
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE Index cols() const {
+    // return the fixed size type if available to enable compile time
+    // optimizations
+    if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                ColsAtCompileTime == Dynamic &&
+        internal::traits<typename internal::remove_all<Arg2Nested>::type>::
+                ColsAtCompileTime == Dynamic)
+      return m_arg3.cols();
+    else if (internal::traits<typename internal::remove_all<Arg1Nested>::type>::
+                     ColsAtCompileTime == Dynamic &&
+             internal::traits<typename internal::remove_all<Arg3Nested>::type>::
+                     ColsAtCompileTime == Dynamic)
+      return m_arg2.cols();
+    else
+      return m_arg1.cols();
+  }
+
+  /** \returns the first argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg1Nested& arg1() const { return m_arg1; }
+  /** \returns the first argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg2Nested& arg2() const { return m_arg2; }
+  /** \returns the third argument nested expression */
+  EIGEN_DEVICE_FUNC
+  const _Arg3Nested& arg3() const { return m_arg3; }
+  /** \returns the functor representing the ternary operation */
+  EIGEN_DEVICE_FUNC
+  const TernaryOp& functor() const { return m_functor; }
+
+ protected:
+  Arg1Nested m_arg1;
+  Arg2Nested m_arg2;
+  Arg3Nested m_arg3;
+  const TernaryOp m_functor;
+};
+
+// Generic API dispatcher
+template <typename TernaryOp, typename Arg1, typename Arg2, typename Arg3,
+          typename StorageKind>
+class CwiseTernaryOpImpl
+    : public internal::generic_xpr_base<
+          CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type {
+ public:
+  typedef typename internal::generic_xpr_base<
+      CwiseTernaryOp<TernaryOp, Arg1, Arg2, Arg3> >::type Base;
+};
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_CWISE_TERNARY_OP_H
diff --git a/Eigen/src/Core/CwiseUnaryOp.h b/Eigen/src/Core/CwiseUnaryOp.h
index f2de749..1d2dd19 100644
--- a/Eigen/src/Core/CwiseUnaryOp.h
+++ b/Eigen/src/Core/CwiseUnaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -13,13 +13,32 @@
 
 namespace Eigen { 
 
+namespace internal {
+template<typename UnaryOp, typename XprType>
+struct traits<CwiseUnaryOp<UnaryOp, XprType> >
+ : traits<XprType>
+{
+  typedef typename result_of<
+                     UnaryOp(const typename XprType::Scalar&)
+                   >::type Scalar;
+  typedef typename XprType::Nested XprTypeNested;
+  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
+  enum {
+    Flags = _XprTypeNested::Flags & RowMajorBit 
+  };
+};
+}
+
+template<typename UnaryOp, typename XprType, typename StorageKind>
+class CwiseUnaryOpImpl;
+
 /** \class CwiseUnaryOp
   * \ingroup Core_Module
   *
   * \brief Generic expression where a coefficient-wise unary operator is applied to an expression
   *
-  * \param UnaryOp template functor implementing the operator
-  * \param XprType the type of the expression to which we are applying the unary operator
+  * \tparam UnaryOp template functor implementing the operator
+  * \tparam XprType the type of the expression to which we are applying the unary operator
   *
   * This class represents an expression where a unary operator is applied to an expression.
   * It is the return type of all operations taking exactly 1 input expression, regardless of the
@@ -32,93 +51,51 @@
   *
   * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp
   */
-
-namespace internal {
 template<typename UnaryOp, typename XprType>
-struct traits<CwiseUnaryOp<UnaryOp, XprType> >
- : traits<XprType>
-{
-  typedef typename result_of<
-                     UnaryOp(typename XprType::Scalar)
-                   >::type Scalar;
-  typedef typename XprType::Nested XprTypeNested;
-  typedef typename remove_reference<XprTypeNested>::type _XprTypeNested;
-  enum {
-    Flags = _XprTypeNested::Flags & (
-      HereditaryBits | LinearAccessBit | AlignedBit
-      | (functor_traits<UnaryOp>::PacketAccess ? PacketAccessBit : 0)),
-    CoeffReadCost = _XprTypeNested::CoeffReadCost + functor_traits<UnaryOp>::Cost
-  };
-};
-}
-
-template<typename UnaryOp, typename XprType, typename StorageKind>
-class CwiseUnaryOpImpl;
-
-template<typename UnaryOp, typename XprType>
-class CwiseUnaryOp : internal::no_assignment_operator,
-  public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>
+class CwiseUnaryOp : public CwiseUnaryOpImpl<UnaryOp, XprType, typename internal::traits<XprType>::StorageKind>, internal::no_assignment_operator
 {
   public:
 
     typedef typename CwiseUnaryOpImpl<UnaryOp, XprType,typename internal::traits<XprType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp)
+    typedef typename internal::ref_selector<XprType>::type XprTypeNested;
+    typedef typename internal::remove_all<XprType>::type NestedExpression;
 
-    inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    explicit CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp())
       : m_xpr(xpr), m_functor(func) {}
 
-    EIGEN_STRONG_INLINE Index rows() const { return m_xpr.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_xpr.cols(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index rows() const { return m_xpr.rows(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Index cols() const { return m_xpr.cols(); }
 
     /** \returns the functor representing the unary operation */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     const UnaryOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    const typename internal::remove_all<typename XprType::Nested>::type&
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    const typename internal::remove_all<XprTypeNested>::type&
     nestedExpression() const { return m_xpr; }
 
     /** \returns the nested expression */
-    typename internal::remove_all<typename XprType::Nested>::type&
-    nestedExpression() { return m_xpr.const_cast_derived(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    typename internal::remove_all<XprTypeNested>::type&
+    nestedExpression() { return m_xpr; }
 
   protected:
-    typename XprType::Nested m_xpr;
+    XprTypeNested m_xpr;
     const UnaryOp m_functor;
 };
 
-// This is the generic implementation for dense storage.
-// It can be used for any expression types implementing the dense concept.
-template<typename UnaryOp, typename XprType>
-class CwiseUnaryOpImpl<UnaryOp,XprType,Dense>
-  : public internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
+// Generic API dispatcher
+template<typename UnaryOp, typename XprType, typename StorageKind>
+class CwiseUnaryOpImpl
+  : public internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type
 {
-  public:
-
-    typedef CwiseUnaryOp<UnaryOp, XprType> Derived;
-    typedef typename internal::dense_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(rowId, colId));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const
-    {
-      return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(rowId, colId));
-    }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(index));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index index) const
-    {
-      return derived().functor().packetOp(derived().nestedExpression().template packet<LoadMode>(index));
-    }
+public:
+  typedef typename internal::generic_xpr_base<CwiseUnaryOp<UnaryOp, XprType> >::type Base;
 };
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/CwiseUnaryView.h b/Eigen/src/Core/CwiseUnaryView.h
index b2638d3..2710330 100644
--- a/Eigen/src/Core/CwiseUnaryView.h
+++ b/Eigen/src/Core/CwiseUnaryView.h
@@ -12,33 +12,19 @@
 
 namespace Eigen {
 
-/** \class CwiseUnaryView
-  * \ingroup Core_Module
-  *
-  * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
-  *
-  * \param ViewOp template functor implementing the view
-  * \param MatrixType the type of the matrix we are applying the unary operator
-  *
-  * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
-  * It is the return type of real() and imag(), and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
-  */
-
 namespace internal {
 template<typename ViewOp, typename MatrixType>
 struct traits<CwiseUnaryView<ViewOp, MatrixType> >
  : traits<MatrixType>
 {
   typedef typename result_of<
-                     ViewOp(typename traits<MatrixType>::Scalar)
+                     ViewOp(const typename traits<MatrixType>::Scalar&)
                    >::type Scalar;
   typedef typename MatrixType::Nested MatrixTypeNested;
   typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
-    Flags = (traits<_MatrixTypeNested>::Flags & (HereditaryBits | LvalueBit | LinearAccessBit | DirectAccessBit)),
-    CoeffReadCost = traits<_MatrixTypeNested>::CoeffReadCost + functor_traits<ViewOp>::Cost,
+    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags = traits<_MatrixTypeNested>::Flags & (RowMajorBit | FlagsLvalueBit | DirectAccessBit), // FIXME DirectAccessBit should not be handled by expressions
     MatrixTypeInnerStride =  inner_stride_at_compile_time<MatrixType>::ret,
     // need to cast the sizeof's from size_t to int explicitly, otherwise:
     // "error: no integral type can represent all of the enumerator values
@@ -55,6 +41,19 @@
 template<typename ViewOp, typename MatrixType, typename StorageKind>
 class CwiseUnaryViewImpl;
 
+/** \class CwiseUnaryView
+  * \ingroup Core_Module
+  *
+  * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector
+  *
+  * \tparam ViewOp template functor implementing the view
+  * \tparam MatrixType the type of the matrix we are applying the unary operator
+  *
+  * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector.
+  * It is the return type of real() and imag(), and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp
+  */
 template<typename ViewOp, typename MatrixType>
 class CwiseUnaryView : public CwiseUnaryViewImpl<ViewOp, MatrixType, typename internal::traits<MatrixType>::StorageKind>
 {
@@ -62,8 +61,10 @@
 
     typedef typename CwiseUnaryViewImpl<ViewOp, MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView)
+    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-    inline CwiseUnaryView(const MatrixType& mat, const ViewOp& func = ViewOp())
+    explicit inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp())
       : m_matrix(mat), m_functor(func) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView)
@@ -75,19 +76,27 @@
     const ViewOp& functor() const { return m_functor; }
 
     /** \returns the nested expression */
-    const typename internal::remove_all<typename MatrixType::Nested>::type&
+    const typename internal::remove_all<MatrixTypeNested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    typename internal::remove_all<typename MatrixType::Nested>::type&
+    typename internal::remove_reference<MatrixTypeNested>::type&
     nestedExpression() { return m_matrix.const_cast_derived(); }
 
   protected:
-    // FIXME changed from MatrixType::Nested because of a weird compilation error with sun CC
-    typename internal::nested<MatrixType>::type m_matrix;
+    MatrixTypeNested m_matrix;
     ViewOp m_functor;
 };
 
+// Generic API dispatcher
+template<typename ViewOp, typename XprType, typename StorageKind>
+class CwiseUnaryViewImpl
+  : public internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<CwiseUnaryView<ViewOp, XprType> >::type Base;
+};
+
 template<typename ViewOp, typename MatrixType>
 class CwiseUnaryViewImpl<ViewOp,MatrixType,Dense>
   : public internal::dense_xpr_base< CwiseUnaryView<ViewOp, MatrixType> >::type
@@ -100,38 +109,18 @@
     EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl)
     
-    inline Scalar* data() { return &coeffRef(0); }
-    inline const Scalar* data() const { return &coeff(0); }
+    EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); }
 
-    inline Index innerStride() const
+    EIGEN_DEVICE_FUNC inline Index innerStride() const
     {
       return derived().nestedExpression().innerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
 
-    inline Index outerStride() const
+    EIGEN_DEVICE_FUNC inline Index outerStride() const
     {
       return derived().nestedExpression().outerStride() * sizeof(typename internal::traits<MatrixType>::Scalar) / sizeof(Scalar);
     }
-
-    EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(row, col));
-    }
-
-    EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const
-    {
-      return derived().functor()(derived().nestedExpression().coeff(index));
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
-    {
-      return derived().functor()(const_cast_derived().nestedExpression().coeffRef(row, col));
-    }
-
-    EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
-    {
-      return derived().functor()(const_cast_derived().nestedExpression().coeffRef(index));
-    }
 };
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/DenseBase.h b/Eigen/src/Core/DenseBase.h
index 32f60d8..90066ae 100644
--- a/Eigen/src/Core/DenseBase.h
+++ b/Eigen/src/Core/DenseBase.h
@@ -34,37 +34,45 @@
   * \tparam Derived is the derived type, e.g., a matrix type or an expression.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN.
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived> class DenseBase
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-  : public internal::special_scalar_op_base<Derived,typename internal::traits<Derived>::Scalar,
-                                     typename NumTraits<typename internal::traits<Derived>::Scalar>::Real>
-#else
   : public DenseCoeffsBase<Derived>
+#else
+  : public DenseCoeffsBase<Derived,DirectWriteAccessors>
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 {
   public:
-    using internal::special_scalar_op_base<Derived,typename internal::traits<Derived>::Scalar,
-                typename NumTraits<typename internal::traits<Derived>::Scalar>::Real>::operator*;
 
-    class InnerIterator;
+    /** Inner iterator type to iterate over the coefficients of a row or column.
+      * \sa class InnerIterator
+      */
+    typedef Eigen::InnerIterator<Derived> InnerIterator;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
 
-    /** \brief The type of indices 
-      * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
-      * \sa \ref TopicPreprocessorDirectives.
-      */
-    typedef typename internal::traits<Derived>::Index Index; 
+    /**
+      * \brief The type used to store indices
+      * \details This typedef is relevant for types that store multiple indices such as
+      *          PermutationMatrix or Transpositions, otherwise it defaults to Eigen::Index
+      * \sa \blank \ref TopicPreprocessorDirectives, Eigen::Index, SparseMatrixBase.
+     */
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
 
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc. */
     typedef typename internal::traits<Derived>::Scalar Scalar;
-    typedef typename internal::packet_traits<Scalar>::type PacketScalar;
+    
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc.
+      *
+      * It is an alias for the Scalar type */
+    typedef Scalar value_type;
+    
     typedef typename NumTraits<Scalar>::Real RealScalar;
-
     typedef DenseCoeffsBase<Derived> Base;
+
     using Base::derived;
     using Base::const_cast_derived;
     using Base::rows;
@@ -74,16 +82,6 @@
     using Base::colIndexByOuterInner;
     using Base::coeff;
     using Base::coeffByOuterInner;
-    using Base::packet;
-    using Base::packetByOuterInner;
-    using Base::writePacket;
-    using Base::writePacketByOuterInner;
-    using Base::coeffRef;
-    using Base::coeffRefByOuterInner;
-    using Base::copyCoeff;
-    using Base::copyCoeffByOuterInner;
-    using Base::copyPacket;
-    using Base::copyPacketByOuterInner;
     using Base::operator();
     using Base::operator[];
     using Base::x;
@@ -169,30 +167,54 @@
       InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
                              : int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime),
 
-      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
-        /**< This is a rough measure of how expensive it is to read one coefficient from
-          * this expression.
-          */
-
       InnerStrideAtCompileTime = internal::inner_stride_at_compile_time<Derived>::ret,
       OuterStrideAtCompileTime = internal::outer_stride_at_compile_time<Derived>::ret
     };
+    
+    typedef typename internal::find_best_packet<Scalar,SizeAtCompileTime>::type PacketScalar;
 
-    enum { ThisConstantIsPrivateInPlainObjectBase };
+    enum { IsPlainObjectBase = 0 };
+    
+    /** The plain matrix type corresponding to this expression.
+      * \sa PlainObject */
+    typedef Matrix<typename internal::traits<Derived>::Scalar,
+                internal::traits<Derived>::RowsAtCompileTime,
+                internal::traits<Derived>::ColsAtCompileTime,
+                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                internal::traits<Derived>::MaxRowsAtCompileTime,
+                internal::traits<Derived>::MaxColsAtCompileTime
+          > PlainMatrix;
+    
+    /** The plain array type corresponding to this expression.
+      * \sa PlainObject */
+    typedef Array<typename internal::traits<Derived>::Scalar,
+                internal::traits<Derived>::RowsAtCompileTime,
+                internal::traits<Derived>::ColsAtCompileTime,
+                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                internal::traits<Derived>::MaxRowsAtCompileTime,
+                internal::traits<Derived>::MaxColsAtCompileTime
+          > PlainArray;
+
+    /** \brief The plain matrix or array type corresponding to this expression.
+      *
+      * This is not necessarily exactly the return type of eval(). In the case of plain matrices,
+      * the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed
+      * that the return type of eval() is either PlainObject or const PlainObject&.
+      */
+    typedef typename internal::conditional<internal::is_same<typename internal::traits<Derived>::XprKind,MatrixXpr >::value,
+                                 PlainMatrix, PlainArray>::type PlainObject;
 
     /** \returns the number of nonzero coefficients which is in practice the number
       * of stored coefficients. */
+    EIGEN_DEVICE_FUNC
     inline Index nonZeros() const { return size(); }
-    /** \returns true if either the number of rows or the number of columns is equal to 1.
-      * In other words, this function returns
-      * \code rows()==1 || cols()==1 \endcode
-      * \sa rows(), cols(), IsVectorAtCompileTime. */
 
     /** \returns the outer size.
       *
       * \note For a vector, this returns just 1. For a matrix (non-vector), this is the major dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of columns for a
       * column-major matrix, and the number of rows for a row-major matrix. */
+    EIGEN_DEVICE_FUNC
     Index outerSize() const
     {
       return IsVectorAtCompileTime ? 1
@@ -204,6 +226,7 @@
       * \note For a vector, this is just the size. For a matrix (non-vector), this is the minor dimension
       * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a 
       * column-major matrix, and the number of columns for a row-major matrix. */
+    EIGEN_DEVICE_FUNC
     Index innerSize() const
     {
       return IsVectorAtCompileTime ? this->size()
@@ -214,6 +237,7 @@
       * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
       * nothing else.
       */
+    EIGEN_DEVICE_FUNC
     void resize(Index newSize)
     {
       EIGEN_ONLY_USED_FOR_DEBUG(newSize);
@@ -224,22 +248,22 @@
       * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does
       * nothing else.
       */
-    void resize(Index nbRows, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    void resize(Index rows, Index cols)
     {
-      EIGEN_ONLY_USED_FOR_DEBUG(nbRows);
-      EIGEN_ONLY_USED_FOR_DEBUG(nbCols);
-      eigen_assert(nbRows == this->rows() && nbCols == this->cols()
+      EIGEN_ONLY_USED_FOR_DEBUG(rows);
+      EIGEN_ONLY_USED_FOR_DEBUG(cols);
+      eigen_assert(rows == this->rows() && cols == this->cols()
                 && "DenseBase::resize() does not actually allow to resize.");
     }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
-
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
-    /** \internal Represents a vector with linearly spaced coefficients that allows sequential access only. */
-    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,false>,Derived> SequentialLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
+    /** \internal \deprecated Represents a vector with linearly spaced coefficients that allows sequential access only. */
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> SequentialLinSpacedReturnType;
     /** \internal Represents a vector with linearly spaced coefficients that allows random access. */
-    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,true>,Derived> RandomAccessLinSpacedReturnType;
+    typedef CwiseNullaryOp<internal::linspaced_op<Scalar,PacketScalar>,PlainObject> RandomAccessLinSpacedReturnType;
     /** \internal the return type of MatrixBase::eigenvalues() */
     typedef Matrix<typename NumTraits<typename internal::traits<Derived>::Scalar>::Real, internal::traits<Derived>::ColsAtCompileTime, 1> EigenvaluesReturnType;
 
@@ -247,120 +271,133 @@
 
     /** Copies \a other into *this. \returns a reference to *this. */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase<OtherDerived>& other);
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator+=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator-=(const EigenBase<OtherDerived> &other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const ReturnByValue<OtherDerived>& func);
 
-    /** \internal Copies \a other into *this without evaluating other. \returns a reference to *this. */
+    /** \internal
+      * Copies \a other into *this without evaluating other. \returns a reference to *this.
+      * \deprecated */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& lazyAssign(const DenseBase<OtherDerived>& other);
 
-    /** \internal Evaluates \a other into *this. \returns a reference to *this. */
-    template<typename OtherDerived>
-    Derived& lazyAssign(const ReturnByValue<OtherDerived>& other);
-
+    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const Scalar& s);
 
+    /** \deprecated it now returns \c *this */
     template<unsigned int Added,unsigned int Removed>
-    const Flagged<Derived, Added, Removed> flagged() const;
+    EIGEN_DEPRECATED
+    const Derived& flagged() const
+    { return derived(); }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     CommaInitializer<Derived> operator<< (const DenseBase<OtherDerived>& other);
 
-    Eigen::Transpose<Derived> transpose();
-	typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
+    typedef Transpose<Derived> TransposeReturnType;
+    EIGEN_DEVICE_FUNC
+    TransposeReturnType transpose();
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
+    EIGEN_DEVICE_FUNC
     ConstTransposeReturnType transpose() const;
+    EIGEN_DEVICE_FUNC
     void transposeInPlace();
-#ifndef EIGEN_NO_DEBUG
-  protected:
-    template<typename OtherDerived>
-    void checkTransposeAliasing(const OtherDerived& other) const;
-  public:
-#endif
 
-
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(Index rows, Index cols, const Scalar& value);
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(Index size, const Scalar& value);
-    static const ConstantReturnType
+    EIGEN_DEVICE_FUNC static const ConstantReturnType
     Constant(const Scalar& value);
 
-    static const SequentialLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high);
-    static const RandomAccessLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(Index size, const Scalar& low, const Scalar& high);
-    static const SequentialLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const SequentialLinSpacedReturnType
     LinSpaced(Sequential_t, const Scalar& low, const Scalar& high);
-    static const RandomAccessLinSpacedReturnType
+    EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType
     LinSpaced(const Scalar& low, const Scalar& high);
 
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(Index size, const CustomNullaryOp& func);
-    template<typename CustomNullaryOp>
-    static const CwiseNullaryOp<CustomNullaryOp, Derived>
+    template<typename CustomNullaryOp> EIGEN_DEVICE_FUNC
+    static const CwiseNullaryOp<CustomNullaryOp, PlainObject>
     NullaryExpr(const CustomNullaryOp& func);
 
-    static const ConstantReturnType Zero(Index rows, Index cols);
-    static const ConstantReturnType Zero(Index size);
-    static const ConstantReturnType Zero();
-    static const ConstantReturnType Ones(Index rows, Index cols);
-    static const ConstantReturnType Ones(Index size);
-    static const ConstantReturnType Ones();
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index size);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Zero();
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index size);
+    EIGEN_DEVICE_FUNC static const ConstantReturnType Ones();
 
-    void fill(const Scalar& value);
-    Derived& setConstant(const Scalar& value);
-    Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high);
-    Derived& setLinSpaced(const Scalar& low, const Scalar& high);
-    Derived& setZero();
-    Derived& setOnes();
-    Derived& setRandom();
+    EIGEN_DEVICE_FUNC void fill(const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setConstant(const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC Derived& setLinSpaced(const Scalar& low, const Scalar& high);
+    EIGEN_DEVICE_FUNC Derived& setZero();
+    EIGEN_DEVICE_FUNC Derived& setOnes();
+    EIGEN_DEVICE_FUNC Derived& setRandom();
 
-    template<typename OtherDerived>
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
     bool isApprox(const DenseBase<OtherDerived>& other,
                   const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC 
     bool isMuchSmallerThan(const RealScalar& other,
                            const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    template<typename OtherDerived>
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC
     bool isMuchSmallerThan(const DenseBase<OtherDerived>& other,
                            const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
-    bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
-    bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isZero(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
+    EIGEN_DEVICE_FUNC bool isOnes(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
     
     inline bool hasNaN() const;
     inline bool allFinite() const;
 
-    inline Derived& operator*=(const Scalar& other);
-    inline Derived& operator/=(const Scalar& other);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator*=(const Scalar& other);
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+    Derived& operator/=(const Scalar& other);
 
     typedef typename internal::add_const_on_value_type<typename internal::eval<Derived>::type>::type EvalReturnType;
     /** \returns the matrix or vector obtained by evaluating this expression.
       *
       * Notice that in the case of a plain matrix or vector (not an expression) this function just returns
       * a const reference, in order to avoid a useless copy.
+      * 
+      * \warning Be carefull with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE EvalReturnType eval() const
     {
       // Even though MSVC does not honor strong inlining when the return type
@@ -368,61 +405,78 @@
       // size types on MSVC.
       return typename internal::eval<Derived>::type(derived());
     }
-
+    
     /** swaps *this with the expression \a other.
       *
       */
     template<typename OtherDerived>
-    void swap(const DenseBase<OtherDerived>& other,
-              int = OtherDerived::ThisConstantIsPrivateInPlainObjectBase)
+    EIGEN_DEVICE_FUNC
+    void swap(const DenseBase<OtherDerived>& other)
     {
-      SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
+      EIGEN_STATIC_ASSERT(!OtherDerived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      eigen_assert(rows()==other.rows() && cols()==other.cols());
+      call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
     }
 
     /** swaps *this with the matrix or array \a other.
       *
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     void swap(PlainObjectBase<OtherDerived>& other)
     {
-      SwapWrapper<Derived>(derived()).lazyAssign(other.derived());
+      eigen_assert(rows()==other.rows() && cols()==other.cols());
+      call_assignment(derived(), other.derived(), internal::swap_assign_op<Scalar>());
     }
 
+    EIGEN_DEVICE_FUNC inline const NestByValue<Derived> nestByValue() const;
+    EIGEN_DEVICE_FUNC inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
+    EIGEN_DEVICE_FUNC inline ForceAlignedAccess<Derived> forceAlignedAccess();
+    template<bool Enable> EIGEN_DEVICE_FUNC
+    inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
+    template<bool Enable> EIGEN_DEVICE_FUNC
+    inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
 
-    inline const NestByValue<Derived> nestByValue() const;
-    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
-    inline ForceAlignedAccess<Derived> forceAlignedAccess();
-    template<bool Enable> inline const typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf() const;
-    template<bool Enable> inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
+    EIGEN_DEVICE_FUNC Scalar sum() const;
+    EIGEN_DEVICE_FUNC Scalar mean() const;
+    EIGEN_DEVICE_FUNC Scalar trace() const;
 
-    Scalar sum() const;
-    Scalar mean() const;
-    Scalar trace() const;
+    EIGEN_DEVICE_FUNC Scalar prod() const;
 
-    Scalar prod() const;
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar minCoeff() const;
+    EIGEN_DEVICE_FUNC typename internal::traits<Derived>::Scalar maxCoeff() const;
 
-    typename internal::traits<Derived>::Scalar minCoeff() const;
-    typename internal::traits<Derived>::Scalar maxCoeff() const;
-
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* row, IndexType* col) const;
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* row, IndexType* col) const;
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar minCoeff(IndexType* index) const;
-    template<typename IndexType>
+    template<typename IndexType> EIGEN_DEVICE_FUNC
     typename internal::traits<Derived>::Scalar maxCoeff(IndexType* index) const;
 
     template<typename BinaryOp>
-    typename internal::result_of<BinaryOp(typename internal::traits<Derived>::Scalar)>::type
-    redux(const BinaryOp& func) const;
+    EIGEN_DEVICE_FUNC
+    Scalar redux(const BinaryOp& func) const;
 
     template<typename Visitor>
+    EIGEN_DEVICE_FUNC
     void visit(Visitor& func) const;
 
-    inline const WithFormat<Derived> format(const IOFormat& fmt) const;
+    /** \returns a WithFormat proxy object allowing to print a matrix the with given
+      * format \a fmt.
+      *
+      * See class IOFormat for some examples.
+      *
+      * \sa class IOFormat, class WithFormat
+      */
+    inline const WithFormat<Derived> format(const IOFormat& fmt) const
+    {
+      return WithFormat<Derived>(derived(), fmt);
+    }
 
     /** \returns the unique coefficient of a 1x1 expression */
+    EIGEN_DEVICE_FUNC
     CoeffReturnType value() const
     {
       EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
@@ -430,23 +484,44 @@
       return derived().coeff(0,0);
     }
 
-    bool all(void) const;
-    bool any(void) const;
-    Index count() const;
+    EIGEN_DEVICE_FUNC bool all() const;
+    EIGEN_DEVICE_FUNC bool any() const;
+    EIGEN_DEVICE_FUNC Index count() const;
 
     typedef VectorwiseOp<Derived, Horizontal> RowwiseReturnType;
     typedef const VectorwiseOp<const Derived, Horizontal> ConstRowwiseReturnType;
     typedef VectorwiseOp<Derived, Vertical> ColwiseReturnType;
     typedef const VectorwiseOp<const Derived, Vertical> ConstColwiseReturnType;
 
-    ConstRowwiseReturnType rowwise() const;
-    RowwiseReturnType rowwise();
-    ConstColwiseReturnType colwise() const;
-    ColwiseReturnType colwise();
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
+    *
+    * Example: \include MatrixBase_rowwise.cpp
+    * Output: \verbinclude MatrixBase_rowwise.out
+    *
+    * \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
+    */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC inline ConstRowwiseReturnType rowwise() const {
+      return ConstRowwiseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC RowwiseReturnType rowwise();
 
-    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(Index rows, Index cols);
-    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random(Index size);
-    static const CwiseNullaryOp<internal::scalar_random_op<Scalar>,Derived> Random();
+    /** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
+    *
+    * Example: \include MatrixBase_colwise.cpp
+    * Output: \verbinclude MatrixBase_colwise.out
+    *
+    * \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
+    */
+    EIGEN_DEVICE_FUNC inline ConstColwiseReturnType colwise() const {
+      return ConstColwiseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC ColwiseReturnType colwise();
+
+    typedef CwiseNullaryOp<internal::scalar_random_op<Scalar>,PlainObject> RandomReturnType;
+    static const RandomReturnType Random(Index rows, Index cols);
+    static const RandomReturnType Random(Index size);
+    static const RandomReturnType Random();
 
     template<typename ThenDerived,typename ElseDerived>
     const Select<Derived,ThenDerived,ElseDerived>
@@ -464,45 +539,56 @@
     template<int p> RealScalar lpNorm() const;
 
     template<int RowFactor, int ColFactor>
-    inline const Replicate<Derived,RowFactor,ColFactor> replicate() const;
-    
-    typedef Replicate<Derived,Dynamic,Dynamic> ReplicateReturnType;
-    inline const ReplicateReturnType replicate(Index rowFacor,Index colFactor) const;
+    EIGEN_DEVICE_FUNC
+    const Replicate<Derived,RowFactor,ColFactor> replicate() const;
+    /**
+    * \return an expression of the replication of \c *this
+    *
+    * Example: \include MatrixBase_replicate_int_int.cpp
+    * Output: \verbinclude MatrixBase_replicate_int_int.out
+    *
+    * \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
+    */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC
+    const Replicate<Derived, Dynamic, Dynamic> replicate(Index rowFactor, Index colFactor) const
+    {
+      return Replicate<Derived, Dynamic, Dynamic>(derived(), rowFactor, colFactor);
+    }
 
     typedef Reverse<Derived, BothDirections> ReverseReturnType;
     typedef const Reverse<const Derived, BothDirections> ConstReverseReturnType;
-    ReverseReturnType reverse();
-    ConstReverseReturnType reverse() const;
-    void reverseInPlace();
+    EIGEN_DEVICE_FUNC ReverseReturnType reverse();
+    /** This is the const version of reverse(). */
+    //Code moved here due to a CUDA compiler bug
+    EIGEN_DEVICE_FUNC ConstReverseReturnType reverse() const
+    {
+      return ConstReverseReturnType(derived());
+    }
+    EIGEN_DEVICE_FUNC void reverseInPlace();
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::DenseBase
+#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
 #   include "../plugins/BlockMethods.h"
 #   ifdef EIGEN_DENSEBASE_PLUGIN
 #     include EIGEN_DENSEBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
-
-#ifdef EIGEN2_SUPPORT
-
-    Block<Derived> corner(CornerType type, Index cRows, Index cCols);
-    const Block<Derived> corner(CornerType type, Index cRows, Index cCols) const;
-    template<int CRows, int CCols>
-    Block<Derived, CRows, CCols> corner(CornerType type);
-    template<int CRows, int CCols>
-    const Block<Derived, CRows, CCols> corner(CornerType type) const;
-
-#endif // EIGEN2_SUPPORT
-
+#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
 
     // disable the use of evalTo for dense objects with a nice compilation error
-    template<typename Dest> inline void evalTo(Dest& ) const
+    template<typename Dest>
+    EIGEN_DEVICE_FUNC
+    inline void evalTo(Dest& ) const
     {
       EIGEN_STATIC_ASSERT((internal::is_same<Dest,void>::value),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS);
     }
 
   protected:
     /** Default constructor. Do nothing. */
-    DenseBase()
+    EIGEN_DEVICE_FUNC DenseBase()
     {
       /* Just checks for self-consistency of the flags.
        * Only do it when debugging Eigen, as this borders on paranoiac and could slow compilation down
@@ -515,9 +601,9 @@
     }
 
   private:
-    explicit DenseBase(int);
-    DenseBase(int,int);
-    template<typename OtherDerived> explicit DenseBase(const DenseBase<OtherDerived>&);
+    EIGEN_DEVICE_FUNC explicit DenseBase(int);
+    EIGEN_DEVICE_FUNC DenseBase(int,int);
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit DenseBase(const DenseBase<OtherDerived>&);
 };
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/DenseCoeffsBase.h b/Eigen/src/Core/DenseCoeffsBase.h
index 3c890f2..c4af48a 100644
--- a/Eigen/src/Core/DenseCoeffsBase.h
+++ b/Eigen/src/Core/DenseCoeffsBase.h
@@ -35,7 +35,6 @@
 {
   public:
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
 
@@ -61,6 +60,7 @@
     using Base::size;
     using Base::derived;
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner) const
     {
       return int(Derived::RowsAtCompileTime) == 1 ? 0
@@ -69,6 +69,7 @@
           : inner;
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner) const
     {
       return int(Derived::ColsAtCompileTime) == 1 ? 0
@@ -91,13 +92,15 @@
       *
       * \sa operator()(Index,Index) const, coeffRef(Index,Index), coeff(Index) const
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const
     {
       eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      return derived().coeff(row, col);
+                         && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).coeff(row,col);
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
     {
       return coeff(rowIndexByOuterInner(outer, inner),
@@ -108,11 +111,12 @@
       *
       * \sa operator()(Index,Index), operator[](Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType operator()(Index row, Index col) const
     {
       eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return derived().coeff(row, col);
+      return coeff(row, col);
     }
 
     /** Short version: don't use this function, use
@@ -130,11 +134,14 @@
       * \sa operator[](Index) const, coeffRef(Index), coeff(Index,Index) const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     coeff(Index index) const
     {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
       eigen_internal_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      return internal::evaluator<Derived>(derived()).coeff(index);
     }
 
 
@@ -146,15 +153,14 @@
       * z() const, w() const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     operator[](Index index) const
     {
-      #ifndef EIGEN2_SUPPORT
       EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
                           THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
-      #endif
       eigen_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      return coeff(index);
     }
 
     /** \returns the coefficient at given index.
@@ -167,32 +173,49 @@
       * z() const, w() const
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     operator()(Index index) const
     {
       eigen_assert(index >= 0 && index < size());
-      return derived().coeff(index);
+      return coeff(index);
     }
 
     /** equivalent to operator[](0).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
     x() const { return (*this)[0]; }
 
     /** equivalent to operator[](1).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    y() const { return (*this)[1]; }
+    y() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
+      return (*this)[1];
+    }
 
     /** equivalent to operator[](2).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    z() const { return (*this)[2]; }
+    z() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
+      return (*this)[2];
+    }
 
     /** equivalent to operator[](3).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE CoeffReturnType
-    w() const { return (*this)[3]; }
+    w() const
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
+      return (*this)[3];
+    }
 
     /** \internal
       * \returns the packet of coefficients starting at the given row and column. It is your responsibility
@@ -207,9 +230,9 @@
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const
     {
-      eigen_internal_assert(row >= 0 && row < rows()
-                      && col >= 0 && col < cols());
-      return derived().template packet<LoadMode>(row,col);
+      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
+      eigen_internal_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(row,col);
     }
 
 
@@ -234,8 +257,11 @@
     template<int LoadMode>
     EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const
     {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
+      typedef typename internal::packet_traits<Scalar>::type DefaultPacketType;
       eigen_internal_assert(index >= 0 && index < size());
-      return derived().template packet<LoadMode>(index);
+      return internal::evaluator<Derived>(derived()).template packet<LoadMode,DefaultPacketType>(index);
     }
 
   protected:
@@ -278,7 +304,6 @@
     typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -311,13 +336,15 @@
       *
       * \sa operator()(Index,Index), coeff(Index, Index) const, coeffRef(Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col)
     {
       eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      return derived().coeffRef(row, col);
+                         && col >= 0 && col < cols());
+      return internal::evaluator<Derived>(derived()).coeffRef(row,col);
     }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     coeffRefByOuterInner(Index outer, Index inner)
     {
@@ -330,12 +357,13 @@
       * \sa operator[](Index)
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator()(Index row, Index col)
     {
       eigen_assert(row >= 0 && row < rows()
           && col >= 0 && col < cols());
-      return derived().coeffRef(row, col);
+      return coeffRef(row, col);
     }
 
 
@@ -354,11 +382,14 @@
       * \sa operator[](Index), coeff(Index) const, coeffRef(Index,Index)
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     coeffRef(Index index)
     {
+      EIGEN_STATIC_ASSERT(internal::evaluator<Derived>::Flags & LinearAccessBit,
+                          THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS)
       eigen_internal_assert(index >= 0 && index < size());
-      return derived().coeffRef(index);
+      return internal::evaluator<Derived>(derived()).coeffRef(index);
     }
 
     /** \returns a reference to the coefficient at given index.
@@ -368,15 +399,14 @@
       * \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator[](Index index)
     {
-      #ifndef EIGEN2_SUPPORT
       EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime,
                           THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD)
-      #endif
       eigen_assert(index >= 0 && index < size());
-      return derived().coeffRef(index);
+      return coeffRef(index);
     }
 
     /** \returns a reference to the coefficient at given index.
@@ -388,167 +418,49 @@
       * \sa operator[](Index) const, operator()(Index,Index), x(), y(), z(), w()
       */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     operator()(Index index)
     {
       eigen_assert(index >= 0 && index < size());
-      return derived().coeffRef(index);
+      return coeffRef(index);
     }
 
     /** equivalent to operator[](0).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
     x() { return (*this)[0]; }
 
     /** equivalent to operator[](1).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
-    y() { return (*this)[1]; }
+    y()
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS);
+      return (*this)[1];
+    }
 
     /** equivalent to operator[](2).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
-    z() { return (*this)[2]; }
+    z()
+    {
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS);
+      return (*this)[2];
+    }
 
     /** equivalent to operator[](3).  */
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar&
-    w() { return (*this)[3]; }
-
-    /** \internal
-      * Stores the given packet of coefficients, at the given row and column of this expression. It is your responsibility
-      * to ensure that a packet really starts there. This method is only available on expressions having the
-      * PacketAccessBit.
-      *
-      * The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select
-      * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-      * starting at an address which is a multiple of the packet size.
-      */
-
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket
-    (Index row, Index col, const typename internal::packet_traits<Scalar>::type& val)
+    w()
     {
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      derived().template writePacket<StoreMode>(row,col,val);
+      EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS);
+      return (*this)[3];
     }
-
-
-    /** \internal */
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacketByOuterInner
-    (Index outer, Index inner, const typename internal::packet_traits<Scalar>::type& val)
-    {
-      writePacket<StoreMode>(rowIndexByOuterInner(outer, inner),
-                            colIndexByOuterInner(outer, inner),
-                            val);
-    }
-
-    /** \internal
-      * Stores the given packet of coefficients, at the given index in this expression. It is your responsibility
-      * to ensure that a packet really starts there. This method is only available on expressions having the
-      * PacketAccessBit and the LinearAccessBit.
-      *
-      * The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select
-      * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets
-      * starting at an address which is a multiple of the packet size.
-      */
-    template<int StoreMode>
-    EIGEN_STRONG_INLINE void writePacket
-    (Index index, const typename internal::packet_traits<Scalar>::type& val)
-    {
-      eigen_internal_assert(index >= 0 && index < size());
-      derived().template writePacket<StoreMode>(index,val);
-    }
-
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-
-    /** \internal Copies the coefficient at position (row,col) of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeff(Index row, Index col, const DenseBase<OtherDerived>& other)
-    {
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      derived().coeffRef(row, col) = other.derived().coeff(row, col);
-    }
-
-    /** \internal Copies the coefficient at the given index of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
-    {
-      eigen_internal_assert(index >= 0 && index < size());
-      derived().coeffRef(index) = other.derived().coeff(index);
-    }
-
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void copyCoeffByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
-    {
-      const Index row = rowIndexByOuterInner(outer,inner);
-      const Index col = colIndexByOuterInner(outer,inner);
-      // derived() is important here: copyCoeff() may be reimplemented in Derived!
-      derived().copyCoeff(row, col, other);
-    }
-
-    /** \internal Copies the packet at position (row,col) of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacket(Index row, Index col, const DenseBase<OtherDerived>& other)
-    {
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      derived().template writePacket<StoreMode>(row, col,
-        other.derived().template packet<LoadMode>(row, col));
-    }
-
-    /** \internal Copies the packet at the given index of other into *this.
-      *
-      * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code
-      * with usual assignments.
-      *
-      * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox.
-      */
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacket(Index index, const DenseBase<OtherDerived>& other)
-    {
-      eigen_internal_assert(index >= 0 && index < size());
-      derived().template writePacket<StoreMode>(index,
-        other.derived().template packet<LoadMode>(index));
-    }
-
-    /** \internal */
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    EIGEN_STRONG_INLINE void copyPacketByOuterInner(Index outer, Index inner, const DenseBase<OtherDerived>& other)
-    {
-      const Index row = rowIndexByOuterInner(outer,inner);
-      const Index col = colIndexByOuterInner(outer,inner);
-      // derived() is important here: copyCoeff() may be reimplemented in Derived!
-      derived().template copyPacket< OtherDerived, StoreMode, LoadMode>(row, col, other);
-    }
-#endif
-
 };
 
 /** \brief Base class providing direct read-only coefficient access to matrices and arrays.
@@ -560,7 +472,7 @@
   * inherits DenseCoeffsBase<Derived, ReadOnlyAccessors> which defines functions to access entries read-only using
   * \c operator() .
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived>
 class DenseCoeffsBase<Derived, DirectAccessors> : public DenseCoeffsBase<Derived, ReadOnlyAccessors>
@@ -568,7 +480,6 @@
   public:
 
     typedef DenseCoeffsBase<Derived, ReadOnlyAccessors> Base;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
@@ -581,6 +492,7 @@
       *
       * \sa outerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -591,6 +503,7 @@
       *
       * \sa innerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
@@ -606,6 +519,7 @@
       *
       * \sa innerStride(), outerStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -615,6 +529,7 @@
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -630,7 +545,7 @@
   * inherits DenseCoeffsBase<Derived, WriteAccessors> which defines functions to access entries read/write using
   * \c operator().
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived>
 class DenseCoeffsBase<Derived, DirectWriteAccessors>
@@ -639,7 +554,6 @@
   public:
 
     typedef DenseCoeffsBase<Derived, WriteAccessors> Base;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
@@ -652,6 +566,7 @@
       *
       * \sa outerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return derived().innerStride();
@@ -662,6 +577,7 @@
       *
       * \sa innerStride(), rowStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return derived().outerStride();
@@ -677,6 +593,7 @@
       *
       * \sa innerStride(), outerStride(), colStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index rowStride() const
     {
       return Derived::IsRowMajor ? outerStride() : innerStride();
@@ -686,6 +603,7 @@
       *
       * \sa innerStride(), outerStride(), rowStride()
       */
+    EIGEN_DEVICE_FUNC
     inline Index colStride() const
     {
       return Derived::IsRowMajor ? innerStride() : outerStride();
@@ -694,33 +612,42 @@
 
 namespace internal {
 
-template<typename Derived, bool JustReturnZero>
+template<int Alignment, typename Derived, bool JustReturnZero>
 struct first_aligned_impl
 {
-  static inline typename Derived::Index run(const Derived&)
+  static inline Index run(const Derived&)
   { return 0; }
 };
 
-template<typename Derived>
-struct first_aligned_impl<Derived, false>
+template<int Alignment, typename Derived>
+struct first_aligned_impl<Alignment, Derived, false>
 {
-  static inline typename Derived::Index run(const Derived& m)
+  static inline Index run(const Derived& m)
   {
-    return internal::first_aligned(&m.const_cast_derived().coeffRef(0,0), m.size());
+    return internal::first_aligned<Alignment>(m.data(), m.size());
   }
 };
 
-/** \internal \returns the index of the first element of the array that is well aligned for vectorization.
+/** \internal \returns the index of the first element of the array stored by \a m that is properly aligned with respect to \a Alignment for vectorization.
+  *
+  * \tparam Alignment requested alignment in Bytes.
   *
   * There is also the variant first_aligned(const Scalar*, Integer) defined in Memory.h. See it for more
   * documentation.
   */
-template<typename Derived>
-static inline typename Derived::Index first_aligned(const Derived& m)
+template<int Alignment, typename Derived>
+static inline Index first_aligned(const DenseBase<Derived>& m)
 {
-  return first_aligned_impl
-          <Derived, (Derived::Flags & AlignedBit) || !(Derived::Flags & DirectAccessBit)>
-          ::run(m);
+  enum { ReturnZero = (int(evaluator<Derived>::Alignment) >= Alignment) || !(Derived::Flags & DirectAccessBit) };
+  return first_aligned_impl<Alignment, Derived, ReturnZero>::run(m.derived());
+}
+
+template<typename Derived>
+static inline Index first_default_aligned(const DenseBase<Derived>& m)
+{
+  typedef typename Derived::Scalar Scalar;
+  typedef typename packet_traits<Scalar>::type DefaultPacketType;
+  return internal::first_aligned<int(unpacket_traits<DefaultPacketType>::alignment),Derived>(m);
 }
 
 template<typename Derived, bool HasDirectAccess = has_direct_access<Derived>::ret>
diff --git a/Eigen/src/Core/DenseStorage.h b/Eigen/src/Core/DenseStorage.h
index a72738e..7958fee 100644
--- a/Eigen/src/Core/DenseStorage.h
+++ b/Eigen/src/Core/DenseStorage.h
@@ -3,7 +3,7 @@
 //
 // Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2010 Hauke Heibel <hauke.heibel@gmail.com>
+// Copyright (C) 2010-2013 Hauke Heibel <hauke.heibel@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -13,9 +13,9 @@
 #define EIGEN_MATRIXSTORAGE_H
 
 #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) X; EIGEN_DENSE_STORAGE_CTOR_PLUGIN;
 #else
-  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+  #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X)
 #endif
 
 namespace Eigen {
@@ -24,7 +24,9 @@
 
 struct constructor_without_unaligned_array_assert {};
 
-template<typename T, int Size> void check_static_allocation_size()
+template<typename T, int Size>
+EIGEN_DEVICE_FUNC
+void check_static_allocation_size()
 {
   // if EIGEN_STACK_ALLOCATION_LIMIT is defined to 0, then no limit
   #if EIGEN_STACK_ALLOCATION_LIMIT
@@ -38,18 +40,19 @@
   */
 template <typename T, int Size, int MatrixOrArrayOptions,
           int Alignment = (MatrixOrArrayOptions&DontAlign) ? 0
-                        : (((Size*sizeof(T))%16)==0) ? 16
-                        : 0 >
+                        : compute_default_alignment<T,Size>::value >
 struct plain_array
 {
   T array[Size];
 
-  plain_array() 
+  EIGEN_DEVICE_FUNC
+  plain_array()
   { 
     check_static_allocation_size<T,Size>();
   }
 
-  plain_array(constructor_without_unaligned_array_assert) 
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert)
   { 
     check_static_allocation_size<T,Size>();
   }
@@ -64,29 +67,88 @@
   template<typename PtrType>
   EIGEN_ALWAYS_INLINE PtrType eigen_unaligned_array_assert_workaround_gcc47(PtrType array) { return array; }
   #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
-    eigen_assert((reinterpret_cast<size_t>(eigen_unaligned_array_assert_workaround_gcc47(array)) & sizemask) == 0 \
+    eigen_assert((internal::UIntPtr(eigen_unaligned_array_assert_workaround_gcc47(array)) & (sizemask)) == 0 \
               && "this assertion is explained here: " \
               "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
               " **** READ THIS WEB PAGE !!! ****");
 #else
   #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \
-    eigen_assert((reinterpret_cast<size_t>(array) & sizemask) == 0 \
+    eigen_assert((internal::UIntPtr(array) & (sizemask)) == 0 \
               && "this assertion is explained here: " \
               "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \
               " **** READ THIS WEB PAGE !!! ****");
 #endif
 
 template <typename T, int Size, int MatrixOrArrayOptions>
-struct plain_array<T, Size, MatrixOrArrayOptions, 16>
+struct plain_array<T, Size, MatrixOrArrayOptions, 8>
 {
-  EIGEN_USER_ALIGN16 T array[Size];
+  EIGEN_ALIGN_TO_BOUNDARY(8) T array[Size];
 
+  EIGEN_DEVICE_FUNC
   plain_array() 
-  { 
-    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(0xf);
+  {
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(7);
     check_static_allocation_size<T,Size>();
   }
 
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 16>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(16) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  { 
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(15);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 32>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(32) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  {
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(31);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
+  plain_array(constructor_without_unaligned_array_assert) 
+  { 
+    check_static_allocation_size<T,Size>();
+  }
+};
+
+template <typename T, int Size, int MatrixOrArrayOptions>
+struct plain_array<T, Size, MatrixOrArrayOptions, 64>
+{
+  EIGEN_ALIGN_TO_BOUNDARY(64) T array[Size];
+
+  EIGEN_DEVICE_FUNC
+  plain_array() 
+  { 
+    EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(63);
+    check_static_allocation_size<T,Size>();
+  }
+
+  EIGEN_DEVICE_FUNC
   plain_array(constructor_without_unaligned_array_assert) 
   { 
     check_static_allocation_size<T,Size>();
@@ -96,9 +158,9 @@
 template <typename T, int MatrixOrArrayOptions, int Alignment>
 struct plain_array<T, 0, MatrixOrArrayOptions, Alignment>
 {
-  EIGEN_USER_ALIGN16 T array[1];
-  plain_array() {}
-  plain_array(constructor_without_unaligned_array_assert) {}
+  T array[1];
+  EIGEN_DEVICE_FUNC plain_array() {}
+  EIGEN_DEVICE_FUNC plain_array(constructor_without_unaligned_array_assert) {}
 };
 
 } // end namespace internal
@@ -122,33 +184,54 @@
 {
     internal::plain_array<T,Size,_Options> m_data;
   public:
-    inline DenseStorage() {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
+    EIGEN_DEVICE_FUNC
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()) {}
-    inline DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
-    static inline DenseIndex rows(void) {return _Rows;}
-    static inline DenseIndex cols(void) {return _Cols;}
-    inline void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void resize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline const T *data() const { return m_data.array; }
-    inline T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC 
+    DenseStorage(const DenseStorage& other) : m_data(other.m_data) {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size)
+    }
+    EIGEN_DEVICE_FUNC 
+    DenseStorage& operator=(const DenseStorage& other)
+    { 
+      if (this != &other) m_data = other.m_data;
+      return *this; 
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols);
+      EIGEN_UNUSED_VARIABLE(size);
+      EIGEN_UNUSED_VARIABLE(rows);
+      EIGEN_UNUSED_VARIABLE(cols);
+    }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); }
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // null matrix
 template<typename T, int _Rows, int _Cols, int _Options> class DenseStorage<T, 0, _Rows, _Cols, _Options>
 {
   public:
-    inline DenseStorage() {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert) {}
-    inline DenseStorage(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void swap(DenseStorage& ) {}
-    static inline DenseIndex rows(void) {return _Rows;}
-    static inline DenseIndex cols(void) {return _Cols;}
-    inline void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline void resize(DenseIndex,DenseIndex,DenseIndex) {}
-    inline const T *data() const { return 0; }
-    inline T *data() { return 0; }
+    EIGEN_DEVICE_FUNC DenseStorage() {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) {}
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) { return *this; }
+    EIGEN_DEVICE_FUNC DenseStorage(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& ) {}
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {}
+    EIGEN_DEVICE_FUNC const T *data() const { return 0; }
+    EIGEN_DEVICE_FUNC T *data() { return 0; }
 };
 
 // more specializations for null matrices; these are necessary to resolve ambiguities
@@ -165,86 +248,158 @@
 template<typename T, int Size, int _Options> class DenseStorage<T, Size, Dynamic, Dynamic, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    DenseIndex m_rows;
-    DenseIndex m_cols;
+    Index m_rows;
+    Index m_cols;
   public:
-    inline DenseStorage() : m_rows(0), m_cols(0) {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0), m_cols(0) {}
-    inline DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) : m_rows(nbRows), m_cols(nbCols) {}
-    inline void swap(DenseStorage& other)
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows), m_cols(other.m_cols) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
+    { 
+      if (this != &other)
+      {
+        m_data = other.m_data;
+        m_rows = other.m_rows;
+        m_cols = other.m_cols;
+      }
+      return *this; 
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index cols) : m_rows(rows), m_cols(cols) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
     { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    inline DenseIndex rows() const {return m_rows;}
-    inline DenseIndex cols() const {return m_cols;}
-    inline void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
-    inline void resize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; }
-    inline const T *data() const { return m_data.array; }
-    inline T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC Index rows() const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols() const {return m_cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
+    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // dynamic-size matrix with fixed-size storage and fixed width
 template<typename T, int Size, int _Cols, int _Options> class DenseStorage<T, Size, Dynamic, _Cols, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    DenseIndex m_rows;
+    Index m_rows;
   public:
-    inline DenseStorage() : m_rows(0) {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0) {}
-    inline DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex) : m_rows(nbRows) {}
-    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    inline DenseIndex rows(void) const {return m_rows;}
-    inline DenseIndex cols(void) const {return _Cols;}
-    inline void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; }
-    inline void resize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; }
-    inline const T *data() const { return m_data.array; }
-    inline T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) 
+    {
+      if (this != &other)
+      {
+        m_data = other.m_data;
+        m_rows = other.m_rows;
+      }
+      return *this; 
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index) : m_rows(rows) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return _Cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index) { m_rows = rows; }
+    EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index) { m_rows = rows; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // dynamic-size matrix with fixed-size storage and fixed height
 template<typename T, int Size, int _Rows, int _Options> class DenseStorage<T, Size, _Rows, Dynamic, _Options>
 {
     internal::plain_array<T,Size,_Options> m_data;
-    DenseIndex m_cols;
+    Index m_cols;
   public:
-    inline DenseStorage() : m_cols(0) {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
       : m_data(internal::constructor_without_unaligned_array_assert()), m_cols(0) {}
-    inline DenseStorage(DenseIndex, DenseIndex, DenseIndex nbCols) : m_cols(nbCols) {}
-    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    inline DenseIndex rows(void) const {return _Rows;}
-    inline DenseIndex cols(void) const {return m_cols;}
-    inline void conservativeResize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; }
-    inline void resize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; }
-    inline const T *data() const { return m_data.array; }
-    inline T *data() { return m_data.array; }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_cols(other.m_cols) {}
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        m_data = other.m_data;
+        m_cols = other.m_cols;
+      }
+      return *this;
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(Index, Index, Index cols) : m_cols(cols) {}
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return _Rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    void conservativeResize(Index, Index, Index cols) { m_cols = cols; }
+    void resize(Index, Index, Index cols) { m_cols = cols; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data.array; }
 };
 
 // purely dynamic matrix.
 template<typename T, int _Options> class DenseStorage<T, Dynamic, Dynamic, Dynamic, _Options>
 {
     T *m_data;
-    DenseIndex m_rows;
-    DenseIndex m_cols;
+    Index m_rows;
+    Index m_cols;
   public:
-    inline DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert)
        : m_data(0), m_rows(0), m_cols(0) {}
-    inline DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
-      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(nbRows), m_cols(nbCols)
-    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
-    inline ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
-    inline void swap(DenseStorage& other)
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows), m_cols(cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*other.m_cols))
+      , m_rows(other.m_rows)
+      , m_cols(other.m_cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*m_cols)
+      internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
+      : m_data(std::move(other.m_data))
+      , m_rows(std::move(other.m_rows))
+      , m_cols(std::move(other.m_cols))
+    {
+      other.m_data = nullptr;
+      other.m_rows = 0;
+      other.m_cols = 0;
+    }
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    {
+      using std::swap;
+      swap(m_data, other.m_data);
+      swap(m_rows, other.m_rows);
+      swap(m_cols, other.m_cols);
+      return *this;
+    }
+#endif
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, m_rows*m_cols); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other)
     { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); }
-    inline DenseIndex rows(void) const {return m_rows;}
-    inline DenseIndex cols(void) const {return m_cols;}
-    inline void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    void conservativeResize(Index size, Index rows, Index cols)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*m_cols);
-      m_rows = nbRows;
-      m_cols = nbCols;
+      m_rows = rows;
+      m_cols = cols;
     }
-    void resize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC void resize(Index size, Index rows, Index cols)
     {
       if(size != m_rows*m_cols)
       {
@@ -253,35 +408,73 @@
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
-      m_rows = nbRows;
-      m_cols = nbCols;
+      m_rows = rows;
+      m_cols = cols;
     }
-    inline const T *data() const { return m_data; }
-    inline T *data() { return m_data; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
 };
 
 // matrix with dynamic width and fixed height (so that matrix has dynamic size).
 template<typename T, int _Rows, int _Options> class DenseStorage<T, Dynamic, _Rows, Dynamic, _Options>
 {
     T *m_data;
-    DenseIndex m_cols;
+    Index m_cols;
   public:
-    inline DenseStorage() : m_data(0), m_cols(0) {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
-    inline DenseStorage(DenseIndex size, DenseIndex, DenseIndex nbCols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(nbCols)
-    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
-    inline ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
-    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
-    static inline DenseIndex rows(void) {return _Rows;}
-    inline DenseIndex cols(void) const {return m_cols;}
-    inline void conservativeResize(DenseIndex size, DenseIndex, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_cols(0) {}
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_cols(cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows==_Rows && cols >=0);
+      EIGEN_UNUSED_VARIABLE(rows);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(_Rows*other.m_cols))
+      , m_cols(other.m_cols)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_cols*_Rows)
+      internal::smart_copy(other.m_data, other.m_data+_Rows*m_cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }    
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
+      : m_data(std::move(other.m_data))
+      , m_cols(std::move(other.m_cols))
+    {
+      other.m_data = nullptr;
+      other.m_cols = 0;
+    }
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    {
+      using std::swap;
+      swap(m_data, other.m_data);
+      swap(m_cols, other.m_cols);
+      return *this;
+    }
+#endif
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Rows*m_cols); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); }
+    EIGEN_DEVICE_FUNC static Index rows(void) {return _Rows;}
+    EIGEN_DEVICE_FUNC Index cols(void) const {return m_cols;}
+    EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, _Rows*m_cols);
-      m_cols = nbCols;
+      m_cols = cols;
     }
-    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex, DenseIndex nbCols)
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index, Index cols)
     {
       if(size != _Rows*m_cols)
       {
@@ -290,34 +483,72 @@
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
-      m_cols = nbCols;
+      m_cols = cols;
     }
-    inline const T *data() const { return m_data; }
-    inline T *data() { return m_data; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
 };
 
 // matrix with dynamic height and fixed width (so that matrix has dynamic size).
 template<typename T, int _Cols, int _Options> class DenseStorage<T, Dynamic, Dynamic, _Cols, _Options>
 {
     T *m_data;
-    DenseIndex m_rows;
+    Index m_rows;
   public:
-    inline DenseStorage() : m_data(0), m_rows(0) {}
-    inline DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
-    inline DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(nbRows)
-    { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN }
-    inline ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
-    inline void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
-    inline DenseIndex rows(void) const {return m_rows;}
-    static inline DenseIndex cols(void) {return _Cols;}
-    inline void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex)
+    EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0) {}
+    explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {}
+    EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size)), m_rows(rows)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
+      eigen_internal_assert(size==rows*cols && rows>=0 && cols == _Cols);
+      EIGEN_UNUSED_VARIABLE(cols);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other)
+      : m_data(internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(other.m_rows*_Cols))
+      , m_rows(other.m_rows)
+    {
+      EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*_Cols)
+      internal::smart_copy(other.m_data, other.m_data+other.m_rows*_Cols, m_data);
+    }
+    EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other)
+    {
+      if (this != &other)
+      {
+        DenseStorage tmp(other);
+        this->swap(tmp);
+      }
+      return *this;
+    }    
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT
+      : m_data(std::move(other.m_data))
+      , m_rows(std::move(other.m_rows))
+    {
+      other.m_data = nullptr;
+      other.m_rows = 0;
+    }
+    EIGEN_DEVICE_FUNC
+    DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT
+    {
+      using std::swap;
+      swap(m_data, other.m_data);
+      swap(m_rows, other.m_rows);
+      return *this;
+    }
+#endif
+    EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto<T,(_Options&DontAlign)==0>(m_data, _Cols*m_rows); }
+    EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); }
+    EIGEN_DEVICE_FUNC Index rows(void) const {return m_rows;}
+    EIGEN_DEVICE_FUNC static Index cols(void) {return _Cols;}
+    void conservativeResize(Index size, Index rows, Index)
     {
       m_data = internal::conditional_aligned_realloc_new_auto<T,(_Options&DontAlign)==0>(m_data, size, m_rows*_Cols);
-      m_rows = nbRows;
+      m_rows = rows;
     }
-    EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex nbRows, DenseIndex)
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index rows, Index)
     {
       if(size != m_rows*_Cols)
       {
@@ -326,12 +557,12 @@
           m_data = internal::conditional_aligned_new_auto<T,(_Options&DontAlign)==0>(size);
         else
           m_data = 0;
-        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN
+        EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({})
       }
-      m_rows = nbRows;
+      m_rows = rows;
     }
-    inline const T *data() const { return m_data; }
-    inline T *data() { return m_data; }
+    EIGEN_DEVICE_FUNC const T *data() const { return m_data; }
+    EIGEN_DEVICE_FUNC T *data() { return m_data; }
 };
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/Diagonal.h b/Eigen/src/Core/Diagonal.h
index 68cf6d4..afcaf35 100644
--- a/Eigen/src/Core/Diagonal.h
+++ b/Eigen/src/Core/Diagonal.h
@@ -21,7 +21,7 @@
   * \param MatrixType the type of the object in which we are taking a sub/main/super diagonal
   * \param DiagIndex the index of the sub/super diagonal. The default is 0 and it means the main diagonal.
   *              A positive value means a superdiagonal, a negative value means a subdiagonal.
-  *              You can also use Dynamic so the index can be set at runtime.
+  *              You can also use DynamicIndex so the index can be set at runtime.
   *
   * The matrix is not required to be square.
   *
@@ -37,7 +37,7 @@
 struct traits<Diagonal<MatrixType,DiagIndex> >
  : traits<MatrixType>
 {
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   typedef typename MatrixType::StorageKind StorageKind;
   enum {
@@ -52,8 +52,7 @@
                                                  MatrixType::MaxColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))),
     MaxColsAtCompileTime = 1,
     MaskLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
-    Flags = (unsigned int)_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit,
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost,
+    Flags = (unsigned int)_MatrixTypeNested::Flags & (RowMajorBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, // FIXME DirectAccessBit should not be handled by expressions
     MatrixTypeOuterStride = outer_stride_at_compile_time<MatrixType>::ret,
     InnerStrideAtCompileTime = MatrixTypeOuterStride == Dynamic ? Dynamic : MatrixTypeOuterStride+1,
     OuterStrideAtCompileTime = 0
@@ -70,20 +69,31 @@
     typedef typename internal::dense_xpr_base<Diagonal>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal)
 
-    inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index)
+    {
+      eigen_assert( a_index <= m_matrix.cols() && -a_index <= m_matrix.rows() );
+    }
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Diagonal)
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const
-    { return m_index.value()<0 ? (std::min<Index>)(m_matrix.cols(),m_matrix.rows()+m_index.value()) : (std::min<Index>)(m_matrix.rows(),m_matrix.cols()-m_index.value()); }
+    {
+      return m_index.value()<0 ? numext::mini<Index>(m_matrix.cols(),m_matrix.rows()+m_index.value())
+                               : numext::mini<Index>(m_matrix.rows(),m_matrix.cols()-m_index.value());
+    }
 
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return 1; }
 
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return m_matrix.outerStride() + 1;
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
       return 0;
@@ -95,62 +105,75 @@
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); }
-    inline const Scalar* data() const { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); }
+    EIGEN_DEVICE_FUNC
+    inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
+    EIGEN_DEVICE_FUNC
+    inline const Scalar* data() const { return &(m_matrix.coeffRef(rowOffset(), colOffset())); }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index)
     {
       EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset());
+      return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index row, Index) const
     {
-      return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset());
+      return m_matrix.coeffRef(row+rowOffset(), row+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline CoeffReturnType coeff(Index row, Index) const
     {
       return m_matrix.coeff(row+rowOffset(), row+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index idx)
     {
       EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset());
+      return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index idx) const
     {
-      return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset());
+      return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset());
     }
 
+    EIGEN_DEVICE_FUNC
     inline CoeffReturnType coeff(Index idx) const
     {
       return m_matrix.coeff(idx+rowOffset(), idx+colOffset());
     }
 
-    const typename internal::remove_all<typename MatrixType::Nested>::type& 
+    EIGEN_DEVICE_FUNC
+    inline const typename internal::remove_all<typename MatrixType::Nested>::type& 
     nestedExpression() const 
     {
       return m_matrix;
     }
 
-    int index() const
+    EIGEN_DEVICE_FUNC
+    inline Index index() const
     {
       return m_index.value();
     }
 
   protected:
-    typename MatrixType::Nested m_matrix;
+    typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
     const internal::variable_if_dynamicindex<Index, DiagIndex> m_index;
 
   private:
     // some compilers may fail to optimize std::max etc in case of compile-time constants...
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index absDiagIndex() const { return m_index.value()>0 ? m_index.value() : -m_index.value(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value()>0 ? 0 : -m_index.value(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value()>0 ? m_index.value() : 0; }
-    // triger a compile time error is someone try to call packet
+    // trigger a compile-time error if someone try to call packet
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index) const;
     template<int LoadMode> typename MatrixType::PacketReturnType packet(Index,Index) const;
 };
@@ -167,7 +190,7 @@
 inline typename MatrixBase<Derived>::DiagonalReturnType
 MatrixBase<Derived>::diagonal()
 {
-  return derived();
+  return DiagonalReturnType(derived());
 }
 
 /** This is the const version of diagonal(). */
@@ -216,20 +239,20 @@
   *
   * \sa MatrixBase::diagonal(), class Diagonal */
 template<typename Derived>
-template<int Index>
-inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index>::Type
+template<int Index_>
+inline typename MatrixBase<Derived>::template DiagonalIndexReturnType<Index_>::Type
 MatrixBase<Derived>::diagonal()
 {
-  return derived();
+  return typename DiagonalIndexReturnType<Index_>::Type(derived());
 }
 
 /** This is the const version of diagonal<int>(). */
 template<typename Derived>
-template<int Index>
-inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index>::Type
+template<int Index_>
+inline typename MatrixBase<Derived>::template ConstDiagonalIndexReturnType<Index_>::Type
 MatrixBase<Derived>::diagonal() const
 {
-  return derived();
+  return typename ConstDiagonalIndexReturnType<Index_>::Type(derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/DiagonalMatrix.h b/Eigen/src/Core/DiagonalMatrix.h
index e6c220f..ecfdce8 100644
--- a/Eigen/src/Core/DiagonalMatrix.h
+++ b/Eigen/src/Core/DiagonalMatrix.h
@@ -22,7 +22,7 @@
     typedef typename DiagonalVectorType::Scalar Scalar;
     typedef typename DiagonalVectorType::RealScalar RealScalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
 
     enum {
       RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
@@ -30,79 +30,61 @@
       MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
       IsVectorAtCompileTime = 0,
-      Flags = 0
+      Flags = NoPreferredStorageOrderBit
     };
 
     typedef Matrix<Scalar, RowsAtCompileTime, ColsAtCompileTime, 0, MaxRowsAtCompileTime, MaxColsAtCompileTime> DenseMatrixType;
     typedef DenseMatrixType DenseType;
     typedef DiagonalMatrix<Scalar,DiagonalVectorType::SizeAtCompileTime,DiagonalVectorType::MaxSizeAtCompileTime> PlainObject;
 
+    EIGEN_DEVICE_FUNC
     inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    EIGEN_DEVICE_FUNC
     inline Derived& derived() { return *static_cast<Derived*>(this); }
 
+    EIGEN_DEVICE_FUNC
     DenseMatrixType toDenseMatrix() const { return derived(); }
-    template<typename DenseDerived>
-    void evalTo(MatrixBase<DenseDerived> &other) const;
-    template<typename DenseDerived>
-    void addTo(MatrixBase<DenseDerived> &other) const
-    { other.diagonal() += diagonal(); }
-    template<typename DenseDerived>
-    void subTo(MatrixBase<DenseDerived> &other) const
-    { other.diagonal() -= diagonal(); }
-
+    
+    EIGEN_DEVICE_FUNC
     inline const DiagonalVectorType& diagonal() const { return derived().diagonal(); }
+    EIGEN_DEVICE_FUNC
     inline DiagonalVectorType& diagonal() { return derived().diagonal(); }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return diagonal().size(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return diagonal().size(); }
 
-    /** \returns the diagonal matrix product of \c *this by the matrix \a matrix.
-      */
     template<typename MatrixDerived>
-    const DiagonalProduct<MatrixDerived, Derived, OnTheLeft>
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,MatrixDerived,LazyProduct>
     operator*(const MatrixBase<MatrixDerived> &matrix) const
     {
-      return DiagonalProduct<MatrixDerived, Derived, OnTheLeft>(matrix.derived(), derived());
+      return Product<Derived, MatrixDerived, LazyProduct>(derived(),matrix.derived());
     }
 
-    inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> >
+    typedef DiagonalWrapper<const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const DiagonalVectorType> > InverseReturnType;
+    EIGEN_DEVICE_FUNC
+    inline const InverseReturnType
     inverse() const
     {
-      return diagonal().cwiseInverse();
+      return InverseReturnType(diagonal().cwiseInverse());
     }
     
-    inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const DiagonalVectorType> >
+    EIGEN_DEVICE_FUNC
+    inline const DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >
     operator*(const Scalar& scalar) const
     {
-      return diagonal() * scalar;
+      return DiagonalWrapper<const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(DiagonalVectorType,Scalar,product) >(diagonal() * scalar);
     }
-    friend inline const DiagonalWrapper<const CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const DiagonalVectorType> >
+    EIGEN_DEVICE_FUNC
+    friend inline const DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >
     operator*(const Scalar& scalar, const DiagonalBase& other)
     {
-      return other.diagonal() * scalar;
+      return DiagonalWrapper<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,DiagonalVectorType,product) >(scalar * other.diagonal());
     }
-    
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    bool isApprox(const DiagonalBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return diagonal().isApprox(other.diagonal(), precision);
-    }
-    template<typename OtherDerived>
-    bool isApprox(const MatrixBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return toDenseMatrix().isApprox(other, precision);
-    }
-    #endif
 };
 
-template<typename Derived>
-template<typename DenseDerived>
-void DiagonalBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
-{
-  other.setZero();
-  other.diagonal() = diagonal();
-}
 #endif
 
 /** \class DiagonalMatrix
@@ -124,10 +106,9 @@
  : traits<Matrix<_Scalar,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
 {
   typedef Matrix<_Scalar,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1> DiagonalVectorType;
-  typedef Dense StorageKind;
-  typedef DenseIndex Index;
+  typedef DiagonalShape StorageKind;
   enum {
-    Flags = LvalueBit
+    Flags = LvalueBit | NoPreferredStorageOrderBit
   };
 };
 }
@@ -141,7 +122,7 @@
     typedef const DiagonalMatrix& Nested;
     typedef _Scalar Scalar;
     typedef typename internal::traits<DiagonalMatrix>::StorageKind StorageKind;
-    typedef typename internal::traits<DiagonalMatrix>::Index Index;
+    typedef typename internal::traits<DiagonalMatrix>::StorageIndex StorageIndex;
     #endif
 
   protected:
@@ -151,24 +132,31 @@
   public:
 
     /** const version of diagonal(). */
+    EIGEN_DEVICE_FUNC
     inline const DiagonalVectorType& diagonal() const { return m_diagonal; }
     /** \returns a reference to the stored vector of diagonal coefficients. */
+    EIGEN_DEVICE_FUNC
     inline DiagonalVectorType& diagonal() { return m_diagonal; }
 
     /** Default constructor without initialization */
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix() {}
 
     /** Constructs a diagonal matrix with given dimension  */
-    inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalMatrix(Index dim) : m_diagonal(dim) {}
 
     /** 2D constructor. */
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const Scalar& x, const Scalar& y) : m_diagonal(x,y) {}
 
     /** 3D constructor. */
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const Scalar& x, const Scalar& y, const Scalar& z) : m_diagonal(x,y,z) {}
 
     /** Copy constructor. */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     inline DiagonalMatrix(const DiagonalBase<OtherDerived>& other) : m_diagonal(other.diagonal()) {}
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
@@ -178,11 +166,13 @@
 
     /** generic constructor from expression of the diagonal coefficients */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     explicit inline DiagonalMatrix(const MatrixBase<OtherDerived>& other) : m_diagonal(other)
     {}
 
     /** Copy operator. */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     DiagonalMatrix& operator=(const DiagonalBase<OtherDerived>& other)
     {
       m_diagonal = other.diagonal();
@@ -193,6 +183,7 @@
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
+    EIGEN_DEVICE_FUNC
     DiagonalMatrix& operator=(const DiagonalMatrix& other)
     {
       m_diagonal = other.diagonal();
@@ -201,14 +192,19 @@
     #endif
 
     /** Resizes to given size. */
+    EIGEN_DEVICE_FUNC
     inline void resize(Index size) { m_diagonal.resize(size); }
     /** Sets all coefficients to zero. */
+    EIGEN_DEVICE_FUNC
     inline void setZero() { m_diagonal.setZero(); }
     /** Resizes and sets all coefficients to zero. */
+    EIGEN_DEVICE_FUNC
     inline void setZero(Index size) { m_diagonal.setZero(size); }
     /** Sets this matrix to be the identity matrix of the current size. */
+    EIGEN_DEVICE_FUNC
     inline void setIdentity() { m_diagonal.setOnes(); }
     /** Sets this matrix to be the identity matrix of the given size. */
+    EIGEN_DEVICE_FUNC
     inline void setIdentity(Index size) { m_diagonal.setOnes(size); }
 };
 
@@ -232,14 +228,15 @@
 {
   typedef _DiagonalVectorType DiagonalVectorType;
   typedef typename DiagonalVectorType::Scalar Scalar;
-  typedef typename DiagonalVectorType::Index Index;
-  typedef typename DiagonalVectorType::StorageKind StorageKind;
+  typedef typename DiagonalVectorType::StorageIndex StorageIndex;
+  typedef DiagonalShape StorageKind;
+  typedef typename traits<DiagonalVectorType>::XprKind XprKind;
   enum {
     RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
     ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    MaxColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime,
-    Flags =  traits<DiagonalVectorType>::Flags & LvalueBit
+    MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime,
+    Flags =  (traits<DiagonalVectorType>::Flags & LvalueBit) | NoPreferredStorageOrderBit
   };
 };
 }
@@ -255,9 +252,11 @@
     #endif
 
     /** Constructor from expression of diagonal coefficients to wrap. */
-    inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {}
 
     /** \returns a const reference to the wrapped expression of diagonal coefficients. */
+    EIGEN_DEVICE_FUNC
     const DiagonalVectorType& diagonal() const { return m_diagonal; }
 
   protected:
@@ -277,7 +276,7 @@
 inline const DiagonalWrapper<const Derived>
 MatrixBase<Derived>::asDiagonal() const
 {
-  return derived();
+  return DiagonalWrapper<const Derived>(derived());
 }
 
 /** \returns true if *this is approximately equal to a diagonal matrix,
@@ -291,12 +290,11 @@
 template<typename Derived>
 bool MatrixBase<Derived>::isDiagonal(const RealScalar& prec) const
 {
-  using std::abs;
   if(cols() != rows()) return false;
   RealScalar maxAbsOnDiagonal = static_cast<RealScalar>(-1);
   for(Index j = 0; j < cols(); ++j)
   {
-    RealScalar absOnDiagonal = abs(coeff(j,j));
+    RealScalar absOnDiagonal = numext::abs(coeff(j,j));
     if(absOnDiagonal > maxAbsOnDiagonal) maxAbsOnDiagonal = absOnDiagonal;
   }
   for(Index j = 0; j < cols(); ++j)
@@ -308,6 +306,38 @@
   return true;
 }
 
+namespace internal {
+
+template<> struct storage_kind_to_shape<DiagonalShape> { typedef DiagonalShape Shape; };
+
+struct Diagonal2Dense {};
+
+template<> struct AssignmentKind<DenseShape,DiagonalShape> { typedef Diagonal2Dense Kind; };
+
+// Diagonal matrix to Dense assignment
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Dense>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    dst.setZero();
+    dst.diagonal() = src.diagonal();
+  }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() += src.diagonal(); }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() -= src.diagonal(); }
+};
+
+} // namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_DIAGONALMATRIX_H
diff --git a/Eigen/src/Core/DiagonalProduct.h b/Eigen/src/Core/DiagonalProduct.h
index 00f8f29..d372b93 100644
--- a/Eigen/src/Core/DiagonalProduct.h
+++ b/Eigen/src/Core/DiagonalProduct.h
@@ -13,116 +13,14 @@
 
 namespace Eigen { 
 
-namespace internal {
-template<typename MatrixType, typename DiagonalType, int ProductOrder>
-struct traits<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
- : traits<MatrixType>
-{
-  typedef typename scalar_product_traits<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ReturnType Scalar;
-  enum {
-    RowsAtCompileTime = MatrixType::RowsAtCompileTime,
-    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-
-    _StorageOrder = MatrixType::Flags & RowMajorBit ? RowMajor : ColMajor,
-    _ScalarAccessOnDiag =  !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft)
-                          ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)),
-    _SameTypes = is_same<typename MatrixType::Scalar, typename DiagonalType::Scalar>::value,
-    // FIXME currently we need same types, but in the future the next rule should be the one
-    //_Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))),
-    _Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))),
-    _LinearAccessMask = (RowsAtCompileTime==1 || ColsAtCompileTime==1) ? LinearAccessBit : 0,
-
-    Flags = ((HereditaryBits|_LinearAccessMask|AlignedBit) & (unsigned int)(MatrixType::Flags)) | (_Vectorizable ? PacketAccessBit : 0),//(int(MatrixType::Flags)&int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit),
-    CoeffReadCost = NumTraits<Scalar>::MulCost + MatrixType::CoeffReadCost + DiagonalType::DiagonalVectorType::CoeffReadCost
-  };
-};
-}
-
-template<typename MatrixType, typename DiagonalType, int ProductOrder>
-class DiagonalProduct : internal::no_assignment_operator,
-                        public MatrixBase<DiagonalProduct<MatrixType, DiagonalType, ProductOrder> >
-{
-  public:
-
-    typedef MatrixBase<DiagonalProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(DiagonalProduct)
-
-    inline DiagonalProduct(const MatrixType& matrix, const DiagonalType& diagonal)
-      : m_matrix(matrix), m_diagonal(diagonal)
-    {
-      eigen_assert(diagonal.diagonal().size() == (ProductOrder == OnTheLeft ? matrix.rows() : matrix.cols()));
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
-    {
-      return m_diagonal.diagonal().coeff(ProductOrder == OnTheLeft ? row : col) * m_matrix.coeff(row, col);
-    }
-    
-    EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const
-    {
-      enum {
-        StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
-      };
-      return coeff(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const
-    {
-      enum {
-        StorageOrder = Flags & RowMajorBit ? RowMajor : ColMajor
-      };
-      const Index indexInDiagonalVector = ProductOrder == OnTheLeft ? row : col;
-      return packet_impl<LoadMode>(row,col,indexInDiagonalVector,typename internal::conditional<
-        ((int(StorageOrder) == RowMajor && int(ProductOrder) == OnTheLeft)
-       ||(int(StorageOrder) == ColMajor && int(ProductOrder) == OnTheRight)), internal::true_type, internal::false_type>::type());
-    }
-    
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet(Index idx) const
-    {
-      enum {
-        StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor
-      };
-      return packet<LoadMode>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
-    }
-
-  protected:
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::true_type) const
-    {
-      return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
-                     internal::pset1<PacketScalar>(m_diagonal.diagonal().coeff(id)));
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::false_type) const
-    {
-      enum {
-        InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime,
-        DiagonalVectorPacketLoadMode = (LoadMode == Aligned && (((InnerSize%16) == 0) || (int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit)==AlignedBit) ? Aligned : Unaligned)
-      };
-      return internal::pmul(m_matrix.template packet<LoadMode>(row, col),
-                     m_diagonal.diagonal().template packet<DiagonalVectorPacketLoadMode>(id));
-    }
-
-    typename MatrixType::Nested m_matrix;
-    typename DiagonalType::Nested m_diagonal;
-};
-
 /** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal.
   */
 template<typename Derived>
 template<typename DiagonalDerived>
-inline const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
+inline const Product<Derived, DiagonalDerived, LazyProduct>
 MatrixBase<Derived>::operator*(const DiagonalBase<DiagonalDerived> &a_diagonal) const
 {
-  return DiagonalProduct<Derived, DiagonalDerived, OnTheRight>(derived(), a_diagonal.derived());
+  return Product<Derived, DiagonalDerived, LazyProduct>(derived(),a_diagonal.derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/Dot.h b/Eigen/src/Core/Dot.h
index 9d7651f..1fe7a84 100644
--- a/Eigen/src/Core/Dot.h
+++ b/Eigen/src/Core/Dot.h
@@ -28,26 +28,33 @@
 >
 struct dot_nocheck
 {
-  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
-  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
+  typedef typename conj_prod::result_type ResScalar;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE
+  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.template binaryExpr<scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> >(b).sum();
+    return a.template binaryExpr<conj_prod>(b).sum();
   }
 };
 
 template<typename T, typename U>
 struct dot_nocheck<T, U, true>
 {
-  typedef typename scalar_product_traits<typename traits<T>::Scalar,typename traits<U>::Scalar>::ReturnType ResScalar;
-  static inline ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
+  typedef scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> conj_prod;
+  typedef typename conj_prod::result_type ResScalar;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE
+  static ResScalar run(const MatrixBase<T>& a, const MatrixBase<U>& b)
   {
-    return a.transpose().template binaryExpr<scalar_conj_product_op<typename traits<T>::Scalar,typename traits<U>::Scalar> >(b).sum();
+    return a.transpose().template binaryExpr<conj_prod>(b).sum();
   }
 };
 
 } // end namespace internal
 
-/** \returns the dot product of *this with other.
+/** \fn MatrixBase::dot
+  * \returns the dot product of *this with other.
   *
   * \only_for_vectors
   *
@@ -59,55 +66,31 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE
+typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
 MatrixBase<Derived>::dot(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
   EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
+#if !(defined(EIGEN_NO_STATIC_ASSERT) && defined(EIGEN_NO_DEBUG))
   typedef internal::scalar_conj_product_op<Scalar,typename OtherDerived::Scalar> func;
   EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar);
-
+#endif
+  
   eigen_assert(size() == other.size());
 
   return internal::dot_nocheck<Derived,OtherDerived>::run(*this, other);
 }
 
-#ifdef EIGEN2_SUPPORT
-/** \returns the dot product of *this with other, with the Eigen2 convention that the dot product is linear in the first variable
-  * (conjugating the second variable). Of course this only makes a difference in the complex case.
-  *
-  * This method is only available in EIGEN2_SUPPORT mode.
-  *
-  * \only_for_vectors
-  *
-  * \sa dot()
-  */
-template<typename Derived>
-template<typename OtherDerived>
-typename internal::traits<Derived>::Scalar
-MatrixBase<Derived>::eigen2_dot(const MatrixBase<OtherDerived>& other) const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
-  EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived)
-  EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
-    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-  eigen_assert(size() == other.size());
-
-  return internal::dot_nocheck<OtherDerived,Derived>::run(other,*this);
-}
-#endif
-
-
 //---------- implementation of L2 norm and related functions ----------
 
 /** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the Frobenius norm.
   * In both cases, it consists in the sum of the square of all the matrix entries.
   * For vectors, this is also equals to the dot product of \c *this with itself.
   *
-  * \sa dot(), norm()
+  * \sa dot(), norm(), lpNorm()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::squaredNorm() const
@@ -119,41 +102,98 @@
   * In both cases, it consists in the square root of the sum of the square of all the matrix entries.
   * For vectors, this is also equals to the square root of the dot product of \c *this with itself.
   *
-  * \sa dot(), squaredNorm()
+  * \sa lpNorm(), dot(), squaredNorm()
   */
 template<typename Derived>
-inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
+EIGEN_STRONG_INLINE typename NumTraits<typename internal::traits<Derived>::Scalar>::Real MatrixBase<Derived>::norm() const
 {
-  using std::sqrt;
-  return sqrt(squaredNorm());
+  return numext::sqrt(squaredNorm());
 }
 
-/** \returns an expression of the quotient of *this by its own norm.
+/** \returns an expression of the quotient of \c *this by its own norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0),
+  *          then this function returns a copy of the input.
   *
   * \only_for_vectors
   *
   * \sa norm(), normalize()
   */
 template<typename Derived>
-inline const typename MatrixBase<Derived>::PlainObject
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::normalized() const
 {
-  typedef typename internal::nested<Derived>::type Nested;
-  typedef typename internal::remove_reference<Nested>::type _Nested;
+  typedef typename internal::nested_eval<Derived,2>::type _Nested;
   _Nested n(derived());
-  return n / n.norm();
+  RealScalar z = n.squaredNorm();
+  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
+  if(z>RealScalar(0))
+    return n / numext::sqrt(z);
+  else
+    return n;
 }
 
 /** Normalizes the vector, i.e. divides it by its own norm.
   *
   * \only_for_vectors
   *
+  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
+  *
   * \sa norm(), normalized()
   */
 template<typename Derived>
-inline void MatrixBase<Derived>::normalize()
+EIGEN_STRONG_INLINE void MatrixBase<Derived>::normalize()
 {
-  *this /= norm();
+  RealScalar z = squaredNorm();
+  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
+  if(z>RealScalar(0))
+    derived() /= numext::sqrt(z);
+}
+
+/** \returns an expression of the quotient of \c *this by its own norm while avoiding underflow and overflow.
+  *
+  * \only_for_vectors
+  *
+  * This method is analogue to the normalized() method, but it reduces the risk of
+  * underflow and overflow when computing the norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0),
+  *          then this function returns a copy of the input.
+  *
+  * \sa stableNorm(), stableNormalize(), normalized()
+  */
+template<typename Derived>
+EIGEN_STRONG_INLINE const typename MatrixBase<Derived>::PlainObject
+MatrixBase<Derived>::stableNormalized() const
+{
+  typedef typename internal::nested_eval<Derived,3>::type _Nested;
+  _Nested n(derived());
+  RealScalar w = n.cwiseAbs().maxCoeff();
+  RealScalar z = (n/w).squaredNorm();
+  if(z>RealScalar(0))
+    return n / (numext::sqrt(z)*w);
+  else
+    return n;
+}
+
+/** Normalizes the vector while avoid underflow and overflow
+  *
+  * \only_for_vectors
+  *
+  * This method is analogue to the normalize() method, but it reduces the risk of
+  * underflow and overflow when computing the norm.
+  *
+  * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged.
+  *
+  * \sa stableNorm(), stableNormalized(), normalize()
+  */
+template<typename Derived>
+EIGEN_STRONG_INLINE void MatrixBase<Derived>::stableNormalize()
+{
+  RealScalar w = cwiseAbs().maxCoeff();
+  RealScalar z = (derived()/w).squaredNorm();
+  if(z>RealScalar(0))
+    derived() /= numext::sqrt(z)*w;
 }
 
 //---------- implementation of other norms ----------
@@ -164,9 +204,10 @@
 struct lpNorm_selector
 {
   typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const MatrixBase<Derived>& m)
   {
-    using std::pow;
+    EIGEN_USING_STD_MATH(pow)
     return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p);
   }
 };
@@ -174,6 +215,7 @@
 template<typename Derived>
 struct lpNorm_selector<Derived, 1>
 {
+  EIGEN_DEVICE_FUNC
   static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.cwiseAbs().sum();
@@ -183,6 +225,7 @@
 template<typename Derived>
 struct lpNorm_selector<Derived, 2>
 {
+  EIGEN_DEVICE_FUNC
   static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
   {
     return m.norm();
@@ -192,23 +235,35 @@
 template<typename Derived>
 struct lpNorm_selector<Derived, Infinity>
 {
-  static inline typename NumTraits<typename traits<Derived>::Scalar>::Real run(const MatrixBase<Derived>& m)
+  typedef typename NumTraits<typename traits<Derived>::Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const MatrixBase<Derived>& m)
   {
+    if(Derived::SizeAtCompileTime==0 || (Derived::SizeAtCompileTime==Dynamic && m.size()==0))
+      return RealScalar(0);
     return m.cwiseAbs().maxCoeff();
   }
 };
 
 } // end namespace internal
 
-/** \returns the \f$ \ell^p \f$ norm of *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
-  *          of the coefficients of *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$
-  *          norm, that is the maximum of the absolute values of the coefficients of *this.
+/** \returns the \b coefficient-wise \f$ \ell^p \f$ norm of \c *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values
+  *          of the coefficients of \c *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$
+  *          norm, that is the maximum of the absolute values of the coefficients of \c *this.
+  *
+  * In all cases, if \c *this is empty, then the value 0 is returned.
+  *
+  * \note For matrices, this function does not compute the <a href="https://en.wikipedia.org/wiki/Operator_norm">operator-norm</a>. That is, if \c *this is a matrix, then its coefficients are interpreted as a 1D vector. Nonetheless, you can easily compute the 1-norm and \f$\infty\f$-norm matrix operator norms using \link TutorialReductionsVisitorsBroadcastingReductionsNorm partial reductions \endlink.
   *
   * \sa norm()
   */
 template<typename Derived>
 template<int p>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
+#else
+MatrixBase<Derived>::RealScalar
+#endif
 MatrixBase<Derived>::lpNorm() const
 {
   return internal::lpNorm_selector<Derived, p>::run(*this);
@@ -227,8 +282,8 @@
 bool MatrixBase<Derived>::isOrthogonal
 (const MatrixBase<OtherDerived>& other, const RealScalar& prec) const
 {
-  typename internal::nested<Derived,2>::type nested(derived());
-  typename internal::nested<OtherDerived,2>::type otherNested(other.derived());
+  typename internal::nested_eval<Derived,2>::type nested(derived());
+  typename internal::nested_eval<OtherDerived,2>::type otherNested(other.derived());
   return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm();
 }
 
@@ -246,13 +301,13 @@
 template<typename Derived>
 bool MatrixBase<Derived>::isUnitary(const RealScalar& prec) const
 {
-  typename Derived::Nested nested(derived());
+  typename internal::nested_eval<Derived,1>::type self(derived());
   for(Index i = 0; i < cols(); ++i)
   {
-    if(!internal::isApprox(nested.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
+    if(!internal::isApprox(self.col(i).squaredNorm(), static_cast<RealScalar>(1), prec))
       return false;
     for(Index j = 0; j < i; ++j)
-      if(!internal::isMuchSmallerThan(nested.col(i).dot(nested.col(j)), static_cast<Scalar>(1), prec))
+      if(!internal::isMuchSmallerThan(self.col(i).dot(self.col(j)), static_cast<Scalar>(1), prec))
         return false;
   }
   return true;
diff --git a/Eigen/src/Core/EigenBase.h b/Eigen/src/Core/EigenBase.h
index fadb458..b195506 100644
--- a/Eigen/src/Core/EigenBase.h
+++ b/Eigen/src/Core/EigenBase.h
@@ -13,7 +13,10 @@
 
 namespace Eigen {
 
-/** Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
+/** \class EigenBase
+  * \ingroup Core_Module
+  * 
+  * Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T).
   *
   * In other words, an EigenBase object is an object that can be copied into a MatrixBase.
   *
@@ -21,39 +24,57 @@
   *
   * Notice that this class is trivial, it is only used to disambiguate overloaded functions.
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived> struct EigenBase
 {
 //   typedef typename internal::plain_matrix_type<Derived>::type PlainObject;
+  
+  /** \brief The interface type of indices
+    * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
+    * \deprecated Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead.
+    * \sa StorageIndex, \ref TopicPreprocessorDirectives.
+    */
+  typedef Eigen::Index Index;
 
+  // FIXME is it needed?
   typedef typename internal::traits<Derived>::StorageKind StorageKind;
-  typedef typename internal::traits<Derived>::Index Index;
 
   /** \returns a reference to the derived object */
+  EIGEN_DEVICE_FUNC
   Derived& derived() { return *static_cast<Derived*>(this); }
   /** \returns a const reference to the derived object */
+  EIGEN_DEVICE_FUNC
   const Derived& derived() const { return *static_cast<const Derived*>(this); }
 
+  EIGEN_DEVICE_FUNC
   inline Derived& const_cast_derived() const
   { return *static_cast<Derived*>(const_cast<EigenBase*>(this)); }
+  EIGEN_DEVICE_FUNC
   inline const Derived& const_derived() const
   { return *static_cast<const Derived*>(this); }
 
   /** \returns the number of rows. \sa cols(), RowsAtCompileTime */
+  EIGEN_DEVICE_FUNC
   inline Index rows() const { return derived().rows(); }
   /** \returns the number of columns. \sa rows(), ColsAtCompileTime*/
+  EIGEN_DEVICE_FUNC
   inline Index cols() const { return derived().cols(); }
   /** \returns the number of coefficients, which is rows()*cols().
     * \sa rows(), cols(), SizeAtCompileTime. */
+  EIGEN_DEVICE_FUNC
   inline Index size() const { return rows() * cols(); }
 
   /** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */
-  template<typename Dest> inline void evalTo(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void evalTo(Dest& dst) const
   { derived().evalTo(dst); }
 
   /** \internal Don't use it, but do the equivalent: \code dst += *this; \endcode */
-  template<typename Dest> inline void addTo(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void addTo(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -63,7 +84,9 @@
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst -= *this; \endcode */
-  template<typename Dest> inline void subTo(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC
+  inline void subTo(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -73,7 +96,8 @@
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheRight(*this); \endcode */
-  template<typename Dest> inline void applyThisOnTheRight(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC inline void applyThisOnTheRight(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -81,7 +105,8 @@
   }
 
   /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheLeft(*this); \endcode */
-  template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC inline void applyThisOnTheLeft(Dest& dst) const
   {
     // This is the default implementation,
     // derived class can reimplement it in a more optimized way.
@@ -104,25 +129,28 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().evalTo(derived());
+  call_assignment(derived(), other.derived());
   return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().addTo(derived());
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
 template<typename Derived>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 Derived& DenseBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
 {
-  other.derived().subTo(derived());
+  call_assignment(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
   return derived();
 }
 
diff --git a/Eigen/src/Core/Flagged.h b/Eigen/src/Core/Flagged.h
deleted file mode 100644
index 1f2955f..0000000
--- a/Eigen/src/Core/Flagged.h
+++ /dev/null
@@ -1,140 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_FLAGGED_H
-#define EIGEN_FLAGGED_H
-
-namespace Eigen { 
-
-/** \class Flagged
-  * \ingroup Core_Module
-  *
-  * \brief Expression with modified flags
-  *
-  * \param ExpressionType the type of the object of which we are modifying the flags
-  * \param Added the flags added to the expression
-  * \param Removed the flags removed from the expression (has priority over Added).
-  *
-  * This class represents an expression whose flags have been modified.
-  * It is the return type of MatrixBase::flagged()
-  * and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::flagged()
-  */
-
-namespace internal {
-template<typename ExpressionType, unsigned int Added, unsigned int Removed>
-struct traits<Flagged<ExpressionType, Added, Removed> > : traits<ExpressionType>
-{
-  enum { Flags = (ExpressionType::Flags | Added) & ~Removed };
-};
-}
-
-template<typename ExpressionType, unsigned int Added, unsigned int Removed> class Flagged
-  : public MatrixBase<Flagged<ExpressionType, Added, Removed> >
-{
-  public:
-
-    typedef MatrixBase<Flagged> Base;
-    
-    EIGEN_DENSE_PUBLIC_INTERFACE(Flagged)
-    typedef typename internal::conditional<internal::must_nest_by_value<ExpressionType>::ret,
-        ExpressionType, const ExpressionType&>::type ExpressionTypeNested;
-    typedef typename ExpressionType::InnerIterator InnerIterator;
-
-    inline Flagged(const ExpressionType& matrix) : m_matrix(matrix) {}
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-    inline Index outerStride() const { return m_matrix.outerStride(); }
-    inline Index innerStride() const { return m_matrix.innerStride(); }
-
-    inline CoeffReturnType coeff(Index row, Index col) const
-    {
-      return m_matrix.coeff(row, col);
-    }
-
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return m_matrix.coeff(index);
-    }
-    
-    inline const Scalar& coeffRef(Index row, Index col) const
-    {
-      return m_matrix.const_cast_derived().coeffRef(row, col);
-    }
-
-    inline const Scalar& coeffRef(Index index) const
-    {
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      return m_matrix.const_cast_derived().coeffRef(row, col);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index row, Index col) const
-    {
-      return m_matrix.template packet<LoadMode>(row, col);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(row, col, x);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return m_matrix.template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(index, x);
-    }
-
-    const ExpressionType& _expression() const { return m_matrix; }
-
-    template<typename OtherDerived>
-    typename ExpressionType::PlainObject solveTriangular(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived>
-    void solveTriangularInPlace(const MatrixBase<OtherDerived>& other) const;
-
-  protected:
-    ExpressionTypeNested m_matrix;
-};
-
-/** \returns an expression of *this with added and removed flags
-  *
-  * This is mostly for internal use.
-  *
-  * \sa class Flagged
-  */
-template<typename Derived>
-template<unsigned int Added,unsigned int Removed>
-inline const Flagged<Derived, Added, Removed>
-DenseBase<Derived>::flagged() const
-{
-  return derived();
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_FLAGGED_H
diff --git a/Eigen/src/Core/ForceAlignedAccess.h b/Eigen/src/Core/ForceAlignedAccess.h
index 807c7a2..7b08b45 100644
--- a/Eigen/src/Core/ForceAlignedAccess.h
+++ b/Eigen/src/Core/ForceAlignedAccess.h
@@ -39,29 +39,29 @@
     typedef typename internal::dense_xpr_base<ForceAlignedAccess>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(ForceAlignedAccess)
 
-    inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
+    EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
     {
       return m_expression.coeff(row, col);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
     {
       return m_expression.const_cast_derived().coeffRef(row, col);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
     {
       return m_expression.coeff(index);
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
     {
       return m_expression.const_cast_derived().coeffRef(index);
     }
@@ -90,7 +90,7 @@
       m_expression.const_cast_derived().template writePacket<Aligned>(index, x);
     }
 
-    operator const ExpressionType&() const { return m_expression; }
+    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType& m_expression;
@@ -127,7 +127,7 @@
 inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type
 MatrixBase<Derived>::forceAlignedAccessIf() const
 {
-  return derived();
+  return derived();  // FIXME This should not work but apparently is never used
 }
 
 /** \returns an expression of *this with forced aligned access if \a Enable is true.
@@ -138,7 +138,7 @@
 inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type
 MatrixBase<Derived>::forceAlignedAccessIf()
 {
-  return derived();
+  return derived();  // FIXME This should not work but apparently is never used
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/Functors.h b/Eigen/src/Core/Functors.h
deleted file mode 100644
index 5f14c65..0000000
--- a/Eigen/src/Core/Functors.h
+++ /dev/null
@@ -1,1026 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_FUNCTORS_H
-#define EIGEN_FUNCTORS_H
-
-namespace Eigen {
-
-namespace internal {
-
-// associative functors:
-
-/** \internal
-  * \brief Template functor to compute the sum of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::operator+, class VectorwiseOp, MatrixBase::sum()
-  */
-template<typename Scalar> struct scalar_sum_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_sum_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a + b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return internal::padd(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
-  { return internal::predux(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_sum_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = packet_traits<Scalar>::HasAdd
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the product of two scalars
-  *
-  * \sa class CwiseBinaryOp, Cwise::operator*(), class VectorwiseOp, MatrixBase::redux()
-  */
-template<typename LhsScalar,typename RhsScalar> struct scalar_product_op {
-  enum {
-    // TODO vectorize mixed product
-    Vectorizable = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasMul && packet_traits<RhsScalar>::HasMul
-  };
-  typedef typename scalar_product_traits<LhsScalar,RhsScalar>::ReturnType result_type;
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_product_op)
-  EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a * b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return internal::pmul(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
-  { return internal::predux_mul(a); }
-};
-template<typename LhsScalar,typename RhsScalar>
-struct functor_traits<scalar_product_op<LhsScalar,RhsScalar> > {
-  enum {
-    Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost)/2, // rough estimate!
-    PacketAccess = scalar_product_op<LhsScalar,RhsScalar>::Vectorizable
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the conjugate product of two scalars
-  *
-  * This is a short cut for conj(x) * y which is needed for optimization purpose; in Eigen2 support mode, this becomes x * conj(y)
-  */
-template<typename LhsScalar,typename RhsScalar> struct scalar_conj_product_op {
-
-  enum {
-    Conj = NumTraits<LhsScalar>::IsComplex
-  };
-  
-  typedef typename scalar_product_traits<LhsScalar,RhsScalar>::ReturnType result_type;
-  
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_conj_product_op)
-  EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const
-  { return conj_helper<LhsScalar,RhsScalar,Conj,false>().pmul(a,b); }
-  
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return conj_helper<Packet,Packet,Conj,false>().pmul(a,b); }
-};
-template<typename LhsScalar,typename RhsScalar>
-struct functor_traits<scalar_conj_product_op<LhsScalar,RhsScalar> > {
-  enum {
-    Cost = NumTraits<LhsScalar>::MulCost,
-    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMul
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the min of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::cwiseMin, class VectorwiseOp, MatrixBase::minCoeff()
-  */
-template<typename Scalar> struct scalar_min_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_min_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { using std::min; return (min)(a, b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return internal::pmin(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
-  { return internal::predux_min(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_min_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = packet_traits<Scalar>::HasMin
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the max of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::cwiseMax, class VectorwiseOp, MatrixBase::maxCoeff()
-  */
-template<typename Scalar> struct scalar_max_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_max_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { using std::max; return (max)(a, b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return internal::pmax(a,b); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const
-  { return internal::predux_max(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_max_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = packet_traits<Scalar>::HasMax
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the hypot of two scalars
-  *
-  * \sa MatrixBase::stableNorm(), class Redux
-  */
-template<typename Scalar> struct scalar_hypot_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_hypot_op)
-//   typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& _x, const Scalar& _y) const
-  {
-    using std::max;
-    using std::min;
-    using std::sqrt;
-    Scalar p = (max)(_x, _y);
-    Scalar q = (min)(_x, _y);
-    Scalar qp = q/p;
-    return p * sqrt(Scalar(1) + qp*qp);
-  }
-};
-template<typename Scalar>
-struct functor_traits<scalar_hypot_op<Scalar> > {
-  enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess=0 };
-};
-
-/** \internal
-  * \brief Template functor to compute the pow of two scalars
-  */
-template<typename Scalar, typename OtherScalar> struct scalar_binary_pow_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_binary_pow_op)
-  inline Scalar operator() (const Scalar& a, const OtherScalar& b) const { return numext::pow(a, b); }
-};
-template<typename Scalar, typename OtherScalar>
-struct functor_traits<scalar_binary_pow_op<Scalar,OtherScalar> > {
-  enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false };
-};
-
-// other binary functors:
-
-/** \internal
-  * \brief Template functor to compute the difference of two scalars
-  *
-  * \sa class CwiseBinaryOp, MatrixBase::operator-
-  */
-template<typename Scalar> struct scalar_difference_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_difference_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a - b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return internal::psub(a,b); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_difference_op<Scalar> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = packet_traits<Scalar>::HasSub
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the quotient of two scalars
-  *
-  * \sa class CwiseBinaryOp, Cwise::operator/()
-  */
-template<typename LhsScalar,typename RhsScalar> struct scalar_quotient_op {
-  enum {
-    // TODO vectorize mixed product
-    Vectorizable = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasDiv && packet_traits<RhsScalar>::HasDiv
-  };
-  typedef typename scalar_product_traits<LhsScalar,RhsScalar>::ReturnType result_type;
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_op)
-  EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a / b; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
-  { return internal::pdiv(a,b); }
-};
-template<typename LhsScalar,typename RhsScalar>
-struct functor_traits<scalar_quotient_op<LhsScalar,RhsScalar> > {
-  enum {
-    Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost), // rough estimate!
-    PacketAccess = scalar_quotient_op<LhsScalar,RhsScalar>::Vectorizable
-  };
-};
-
-
-
-/** \internal
-  * \brief Template functor to compute the and of two booleans
-  *
-  * \sa class CwiseBinaryOp, ArrayBase::operator&&
-  */
-struct scalar_boolean_and_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_and_op)
-  EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a && b; }
-};
-template<> struct functor_traits<scalar_boolean_and_op> {
-  enum {
-    Cost = NumTraits<bool>::AddCost,
-    PacketAccess = false
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the or of two booleans
-  *
-  * \sa class CwiseBinaryOp, ArrayBase::operator||
-  */
-struct scalar_boolean_or_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_or_op)
-  EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a || b; }
-};
-template<> struct functor_traits<scalar_boolean_or_op> {
-  enum {
-    Cost = NumTraits<bool>::AddCost,
-    PacketAccess = false
-  };
-};
-
-/** \internal
-  * \brief Template functors for comparison of two scalars
-  * \todo Implement packet-comparisons
-  */
-template<typename Scalar, ComparisonName cmp> struct scalar_cmp_op;
-
-template<typename Scalar, ComparisonName cmp>
-struct functor_traits<scalar_cmp_op<Scalar, cmp> > {
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = false
-  };
-};
-
-template<ComparisonName Cmp, typename Scalar>
-struct result_of<scalar_cmp_op<Scalar, Cmp>(Scalar,Scalar)> {
-  typedef bool type;
-};
-
-
-template<typename Scalar> struct scalar_cmp_op<Scalar, cmp_EQ> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
-  EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a==b;}
-};
-template<typename Scalar> struct scalar_cmp_op<Scalar, cmp_LT> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
-  EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a<b;}
-};
-template<typename Scalar> struct scalar_cmp_op<Scalar, cmp_LE> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
-  EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a<=b;}
-};
-template<typename Scalar> struct scalar_cmp_op<Scalar, cmp_UNORD> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
-  EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return !(a<=b || b<=a);}
-};
-template<typename Scalar> struct scalar_cmp_op<Scalar, cmp_NEQ> {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
-  EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a!=b;}
-};
-
-// unary functors:
-
-/** \internal
-  * \brief Template functor to compute the opposite of a scalar
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::operator-
-  */
-template<typename Scalar> struct scalar_opposite_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_opposite_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return -a; }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
-  { return internal::pnegate(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_opposite_op<Scalar> >
-{ enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = packet_traits<Scalar>::HasNegate };
-};
-
-/** \internal
-  * \brief Template functor to compute the absolute value of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::abs
-  */
-template<typename Scalar> struct scalar_abs_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { using std::abs; return abs(a); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
-  { return internal::pabs(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_abs_op<Scalar> >
-{
-  enum {
-    Cost = NumTraits<Scalar>::AddCost,
-    PacketAccess = packet_traits<Scalar>::HasAbs
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the squared absolute value of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::abs2
-  */
-template<typename Scalar> struct scalar_abs2_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs2_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return numext::abs2(a); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
-  { return internal::pmul(a,a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_abs2_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasAbs2 }; };
-
-/** \internal
-  * \brief Template functor to compute the conjugate of a complex value
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::conjugate()
-  */
-template<typename Scalar> struct scalar_conjugate_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_conjugate_op)
-  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { using numext::conj; return conj(a); }
-  template<typename Packet>
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const { return internal::pconj(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_conjugate_op<Scalar> >
-{
-  enum {
-    Cost = NumTraits<Scalar>::IsComplex ? NumTraits<Scalar>::AddCost : 0,
-    PacketAccess = packet_traits<Scalar>::HasConj
-  };
-};
-
-/** \internal
-  * \brief Template functor to cast a scalar to another type
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::cast()
-  */
-template<typename Scalar, typename NewType>
-struct scalar_cast_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
-  typedef NewType result_type;
-  EIGEN_STRONG_INLINE const NewType operator() (const Scalar& a) const { return cast<Scalar, NewType>(a); }
-};
-template<typename Scalar, typename NewType>
-struct functor_traits<scalar_cast_op<Scalar,NewType> >
-{ enum { Cost = is_same<Scalar, NewType>::value ? 0 : NumTraits<NewType>::AddCost, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to extract the real part of a complex
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::real()
-  */
-template<typename Scalar>
-struct scalar_real_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::real(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_real_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to extract the imaginary part of a complex
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::imag()
-  */
-template<typename Scalar>
-struct scalar_imag_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::imag(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_imag_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to extract the real part of a complex as a reference
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::real()
-  */
-template<typename Scalar>
-struct scalar_real_ref_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_ref_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::real_ref(*const_cast<Scalar*>(&a)); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_real_ref_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to extract the imaginary part of a complex as a reference
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::imag()
-  */
-template<typename Scalar>
-struct scalar_imag_ref_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_ref_op)
-  typedef typename NumTraits<Scalar>::Real result_type;
-  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::imag_ref(*const_cast<Scalar*>(&a)); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_imag_ref_op<Scalar> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-/** \internal
-  *
-  * \brief Template functor to compute the exponential of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::exp()
-  */
-template<typename Scalar> struct scalar_exp_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_exp_op)
-  inline const Scalar operator() (const Scalar& a) const { using std::exp; return exp(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::pexp(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_exp_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasExp }; };
-
-/** \internal
-  *
-  * \brief Template functor to compute the logarithm of a scalar
-  *
-  * \sa class CwiseUnaryOp, Cwise::log()
-  */
-template<typename Scalar> struct scalar_log_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_log_op)
-  inline const Scalar operator() (const Scalar& a) const { using std::log; return log(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::plog(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_log_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasLog }; };
-
-/** \internal
-  * \brief Template functor to multiply a scalar by a fixed other one
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::operator*, MatrixBase::operator/
-  */
-/* NOTE why doing the pset1() in packetOp *is* an optimization ?
- * indeed it seems better to declare m_other as a Packet and do the pset1() once
- * in the constructor. However, in practice:
- *  - GCC does not like m_other as a Packet and generate a load every time it needs it
- *  - on the other hand GCC is able to moves the pset1() outside the loop :)
- *  - simpler code ;)
- * (ICC and gcc 4.4 seems to perform well in both cases, the issue is visible with y = a*x + b*y)
- */
-template<typename Scalar>
-struct scalar_multiple_op {
-  typedef typename packet_traits<Scalar>::type Packet;
-  // FIXME default copy constructors seems bugged with std::complex<>
-  EIGEN_STRONG_INLINE scalar_multiple_op(const scalar_multiple_op& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE scalar_multiple_op(const Scalar& other) : m_other(other) { }
-  EIGEN_STRONG_INLINE Scalar operator() (const Scalar& a) const { return a * m_other; }
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
-  { return internal::pmul(a, pset1<Packet>(m_other)); }
-  typename add_const_on_value_type<typename NumTraits<Scalar>::Nested>::type m_other;
-};
-template<typename Scalar>
-struct functor_traits<scalar_multiple_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
-
-template<typename Scalar1, typename Scalar2>
-struct scalar_multiple2_op {
-  typedef typename scalar_product_traits<Scalar1,Scalar2>::ReturnType result_type;
-  EIGEN_STRONG_INLINE scalar_multiple2_op(const scalar_multiple2_op& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE scalar_multiple2_op(const Scalar2& other) : m_other(other) { }
-  EIGEN_STRONG_INLINE result_type operator() (const Scalar1& a) const { return a * m_other; }
-  typename add_const_on_value_type<typename NumTraits<Scalar2>::Nested>::type m_other;
-};
-template<typename Scalar1,typename Scalar2>
-struct functor_traits<scalar_multiple2_op<Scalar1,Scalar2> >
-{ enum { Cost = NumTraits<Scalar1>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to divide a scalar by a fixed other one
-  *
-  * This functor is used to implement the quotient of a matrix by
-  * a scalar where the scalar type is not necessarily a floating point type.
-  *
-  * \sa class CwiseUnaryOp, MatrixBase::operator/
-  */
-template<typename Scalar>
-struct scalar_quotient1_op {
-  typedef typename packet_traits<Scalar>::type Packet;
-  // FIXME default copy constructors seems bugged with std::complex<>
-  EIGEN_STRONG_INLINE scalar_quotient1_op(const scalar_quotient1_op& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE scalar_quotient1_op(const Scalar& other) : m_other(other) {}
-  EIGEN_STRONG_INLINE Scalar operator() (const Scalar& a) const { return a / m_other; }
-  EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
-  { return internal::pdiv(a, pset1<Packet>(m_other)); }
-  typename add_const_on_value_type<typename NumTraits<Scalar>::Nested>::type m_other;
-};
-template<typename Scalar>
-struct functor_traits<scalar_quotient1_op<Scalar> >
-{ enum { Cost = 2 * NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasDiv }; };
-
-// nullary functors
-
-template<typename Scalar>
-struct scalar_constant_op {
-  typedef typename packet_traits<Scalar>::type Packet;
-  EIGEN_STRONG_INLINE scalar_constant_op(const scalar_constant_op& other) : m_other(other.m_other) { }
-  EIGEN_STRONG_INLINE scalar_constant_op(const Scalar& other) : m_other(other) { }
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index, Index = 0) const { return m_other; }
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index, Index = 0) const { return internal::pset1<Packet>(m_other); }
-  const Scalar m_other;
-};
-template<typename Scalar>
-struct functor_traits<scalar_constant_op<Scalar> >
-// FIXME replace this packet test by a safe one
-{ enum { Cost = 1, PacketAccess = packet_traits<Scalar>::Vectorizable, IsRepeatable = true }; };
-
-template<typename Scalar> struct scalar_identity_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_identity_op)
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index row, Index col) const { return row==col ? Scalar(1) : Scalar(0); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_identity_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = false, IsRepeatable = true }; };
-
-template <typename Scalar, bool RandomAccess> struct linspaced_op_impl;
-
-// linear access for packet ops:
-// 1) initialization
-//   base = [low, ..., low] + ([step, ..., step] * [-size, ..., 0])
-// 2) each step (where size is 1 for coeff access or PacketSize for packet access)
-//   base += [size*step, ..., size*step]
-//
-// TODO: Perhaps it's better to initialize lazily (so not in the constructor but in packetOp)
-//       in order to avoid the padd() in operator() ?
-template <typename Scalar>
-struct linspaced_op_impl<Scalar,false>
-{
-  typedef typename packet_traits<Scalar>::type Packet;
-
-  linspaced_op_impl(const Scalar& low, const Scalar& step) :
-  m_low(low), m_step(step),
-  m_packetStep(pset1<Packet>(packet_traits<Scalar>::size*step)),
-  m_base(padd(pset1<Packet>(low), pmul(pset1<Packet>(step),plset<Scalar>(-packet_traits<Scalar>::size)))) {}
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index i) const 
-  { 
-    m_base = padd(m_base, pset1<Packet>(m_step));
-    return m_low+Scalar(i)*m_step; 
-  }
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index) const { return m_base = padd(m_base,m_packetStep); }
-
-  const Scalar m_low;
-  const Scalar m_step;
-  const Packet m_packetStep;
-  mutable Packet m_base;
-};
-
-// random access for packet ops:
-// 1) each step
-//   [low, ..., low] + ( [step, ..., step] * ( [i, ..., i] + [0, ..., size] ) )
-template <typename Scalar>
-struct linspaced_op_impl<Scalar,true>
-{
-  typedef typename packet_traits<Scalar>::type Packet;
-
-  linspaced_op_impl(const Scalar& low, const Scalar& step) :
-  m_low(low), m_step(step),
-  m_lowPacket(pset1<Packet>(m_low)), m_stepPacket(pset1<Packet>(m_step)), m_interPacket(plset<Scalar>(0)) {}
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index i) const { return m_low+i*m_step; }
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index i) const
-  { return internal::padd(m_lowPacket, pmul(m_stepPacket, padd(pset1<Packet>(Scalar(i)),m_interPacket))); }
-
-  const Scalar m_low;
-  const Scalar m_step;
-  const Packet m_lowPacket;
-  const Packet m_stepPacket;
-  const Packet m_interPacket;
-};
-
-// ----- Linspace functor ----------------------------------------------------------------
-
-// Forward declaration (we default to random access which does not really give
-// us a speed gain when using packet access but it allows to use the functor in
-// nested expressions).
-template <typename Scalar, bool RandomAccess = true> struct linspaced_op;
-template <typename Scalar, bool RandomAccess> struct functor_traits< linspaced_op<Scalar,RandomAccess> >
-{ enum { Cost = 1, PacketAccess = packet_traits<Scalar>::HasSetLinear, IsRepeatable = true }; };
-template <typename Scalar, bool RandomAccess> struct linspaced_op
-{
-  typedef typename packet_traits<Scalar>::type Packet;
-  linspaced_op(const Scalar& low, const Scalar& high, DenseIndex num_steps) : impl((num_steps==1 ? high : low), (num_steps==1 ? Scalar() : (high-low)/Scalar(num_steps-1))) {}
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index i) const { return impl(i); }
-
-  // We need this function when assigning e.g. a RowVectorXd to a MatrixXd since
-  // there row==0 and col is used for the actual iteration.
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Scalar operator() (Index row, Index col) const 
-  {
-    eigen_assert(col==0 || row==0);
-    return impl(col + row);
-  }
-
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index i) const { return impl.packetOp(i); }
-
-  // We need this function when assigning e.g. a RowVectorXd to a MatrixXd since
-  // there row==0 and col is used for the actual iteration.
-  template<typename Index>
-  EIGEN_STRONG_INLINE const Packet packetOp(Index row, Index col) const
-  {
-    eigen_assert(col==0 || row==0);
-    return impl.packetOp(col + row);
-  }
-
-  // This proxy object handles the actual required temporaries, the different
-  // implementations (random vs. sequential access) as well as the
-  // correct piping to size 2/4 packet operations.
-  const linspaced_op_impl<Scalar,RandomAccess> impl;
-};
-
-// all functors allow linear access, except scalar_identity_op. So we fix here a quick meta
-// to indicate whether a functor allows linear access, just always answering 'yes' except for
-// scalar_identity_op.
-// FIXME move this to functor_traits adding a functor_default
-template<typename Functor> struct functor_has_linear_access { enum { ret = 1 }; };
-template<typename Scalar> struct functor_has_linear_access<scalar_identity_op<Scalar> > { enum { ret = 0 }; };
-
-// In Eigen, any binary op (Product, CwiseBinaryOp) require the Lhs and Rhs to have the same scalar type, except for multiplication
-// where the mixing of different types is handled by scalar_product_traits
-// In particular, real * complex<real> is allowed.
-// FIXME move this to functor_traits adding a functor_default
-template<typename Functor> struct functor_is_product_like { enum { ret = 0 }; };
-template<typename LhsScalar,typename RhsScalar> struct functor_is_product_like<scalar_product_op<LhsScalar,RhsScalar> > { enum { ret = 1 }; };
-template<typename LhsScalar,typename RhsScalar> struct functor_is_product_like<scalar_conj_product_op<LhsScalar,RhsScalar> > { enum { ret = 1 }; };
-template<typename LhsScalar,typename RhsScalar> struct functor_is_product_like<scalar_quotient_op<LhsScalar,RhsScalar> > { enum { ret = 1 }; };
-
-
-/** \internal
-  * \brief Template functor to add a scalar to a fixed other one
-  * \sa class CwiseUnaryOp, Array::operator+
-  */
-/* If you wonder why doing the pset1() in packetOp() is an optimization check scalar_multiple_op */
-template<typename Scalar>
-struct scalar_add_op {
-  typedef typename packet_traits<Scalar>::type Packet;
-  // FIXME default copy constructors seems bugged with std::complex<>
-  inline scalar_add_op(const scalar_add_op& other) : m_other(other.m_other) { }
-  inline scalar_add_op(const Scalar& other) : m_other(other) { }
-  inline Scalar operator() (const Scalar& a) const { return a + m_other; }
-  inline const Packet packetOp(const Packet& a) const
-  { return internal::padd(a, pset1<Packet>(m_other)); }
-  const Scalar m_other;
-};
-template<typename Scalar>
-struct functor_traits<scalar_add_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = packet_traits<Scalar>::HasAdd }; };
-
-/** \internal
-  * \brief Template functor to compute the square root of a scalar
-  * \sa class CwiseUnaryOp, Cwise::sqrt()
-  */
-template<typename Scalar> struct scalar_sqrt_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_sqrt_op)
-  inline const Scalar operator() (const Scalar& a) const { using std::sqrt; return sqrt(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::psqrt(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_sqrt_op<Scalar> >
-{ enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = packet_traits<Scalar>::HasSqrt
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the cosine of a scalar
-  * \sa class CwiseUnaryOp, ArrayBase::cos()
-  */
-template<typename Scalar> struct scalar_cos_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cos_op)
-  inline Scalar operator() (const Scalar& a) const { using std::cos; return cos(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::pcos(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_cos_op<Scalar> >
-{
-  enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = packet_traits<Scalar>::HasCos
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the sine of a scalar
-  * \sa class CwiseUnaryOp, ArrayBase::sin()
-  */
-template<typename Scalar> struct scalar_sin_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_sin_op)
-  inline const Scalar operator() (const Scalar& a) const { using std::sin; return sin(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::psin(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_sin_op<Scalar> >
-{
-  enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = packet_traits<Scalar>::HasSin
-  };
-};
-
-
-/** \internal
-  * \brief Template functor to compute the tan of a scalar
-  * \sa class CwiseUnaryOp, ArrayBase::tan()
-  */
-template<typename Scalar> struct scalar_tan_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_tan_op)
-  inline const Scalar operator() (const Scalar& a) const { using std::tan; return tan(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::ptan(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_tan_op<Scalar> >
-{
-  enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = packet_traits<Scalar>::HasTan
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the arc cosine of a scalar
-  * \sa class CwiseUnaryOp, ArrayBase::acos()
-  */
-template<typename Scalar> struct scalar_acos_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_acos_op)
-  inline const Scalar operator() (const Scalar& a) const { using std::acos; return acos(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::pacos(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_acos_op<Scalar> >
-{
-  enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = packet_traits<Scalar>::HasACos
-  };
-};
-
-/** \internal
-  * \brief Template functor to compute the arc sine of a scalar
-  * \sa class CwiseUnaryOp, ArrayBase::asin()
-  */
-template<typename Scalar> struct scalar_asin_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_asin_op)
-  inline const Scalar operator() (const Scalar& a) const { using std::asin; return asin(a); }
-  typedef typename packet_traits<Scalar>::type Packet;
-  inline Packet packetOp(const Packet& a) const { return internal::pasin(a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_asin_op<Scalar> >
-{
-  enum {
-    Cost = 5 * NumTraits<Scalar>::MulCost,
-    PacketAccess = packet_traits<Scalar>::HasASin
-  };
-};
-
-/** \internal
-  * \brief Template functor to raise a scalar to a power
-  * \sa class CwiseUnaryOp, Cwise::pow
-  */
-template<typename Scalar>
-struct scalar_pow_op {
-  // FIXME default copy constructors seems bugged with std::complex<>
-  inline scalar_pow_op(const scalar_pow_op& other) : m_exponent(other.m_exponent) { }
-  inline scalar_pow_op(const Scalar& exponent) : m_exponent(exponent) {}
-  inline Scalar operator() (const Scalar& a) const { return numext::pow(a, m_exponent); }
-  const Scalar m_exponent;
-};
-template<typename Scalar>
-struct functor_traits<scalar_pow_op<Scalar> >
-{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false }; };
-
-/** \internal
-  * \brief Template functor to compute the quotient between a scalar and array entries.
-  * \sa class CwiseUnaryOp, Cwise::inverse()
-  */
-template<typename Scalar>
-struct scalar_inverse_mult_op {
-  scalar_inverse_mult_op(const Scalar& other) : m_other(other) {}
-  inline Scalar operator() (const Scalar& a) const { return m_other / a; }
-  template<typename Packet>
-  inline const Packet packetOp(const Packet& a) const
-  { return internal::pdiv(pset1<Packet>(m_other),a); }
-  Scalar m_other;
-};
-
-/** \internal
-  * \brief Template functor to compute the inverse of a scalar
-  * \sa class CwiseUnaryOp, Cwise::inverse()
-  */
-template<typename Scalar>
-struct scalar_inverse_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_inverse_op)
-  inline Scalar operator() (const Scalar& a) const { return Scalar(1)/a; }
-  template<typename Packet>
-  inline const Packet packetOp(const Packet& a) const
-  { return internal::pdiv(pset1<Packet>(Scalar(1)),a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_inverse_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasDiv }; };
-
-/** \internal
-  * \brief Template functor to compute the square of a scalar
-  * \sa class CwiseUnaryOp, Cwise::square()
-  */
-template<typename Scalar>
-struct scalar_square_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_square_op)
-  inline Scalar operator() (const Scalar& a) const { return a*a; }
-  template<typename Packet>
-  inline const Packet packetOp(const Packet& a) const
-  { return internal::pmul(a,a); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_square_op<Scalar> >
-{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
-
-/** \internal
-  * \brief Template functor to compute the cube of a scalar
-  * \sa class CwiseUnaryOp, Cwise::cube()
-  */
-template<typename Scalar>
-struct scalar_cube_op {
-  EIGEN_EMPTY_STRUCT_CTOR(scalar_cube_op)
-  inline Scalar operator() (const Scalar& a) const { return a*a*a; }
-  template<typename Packet>
-  inline const Packet packetOp(const Packet& a) const
-  { return internal::pmul(a,pmul(a,a)); }
-};
-template<typename Scalar>
-struct functor_traits<scalar_cube_op<Scalar> >
-{ enum { Cost = 2*NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
-
-// default functor traits for STL functors:
-
-template<typename T>
-struct functor_traits<std::multiplies<T> >
-{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::divides<T> >
-{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::plus<T> >
-{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::minus<T> >
-{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::negate<T> >
-{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::logical_or<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::logical_and<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::logical_not<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::greater<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::less<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::greater_equal<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::less_equal<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::equal_to<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::not_equal_to<T> >
-{ enum { Cost = 1, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::binder2nd<T> >
-{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::binder1st<T> >
-{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::unary_negate<T> >
-{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
-
-template<typename T>
-struct functor_traits<std::binary_negate<T> >
-{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
-
-#ifdef EIGEN_STDEXT_SUPPORT
-
-template<typename T0,typename T1>
-struct functor_traits<std::project1st<T0,T1> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct functor_traits<std::project2nd<T0,T1> >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct functor_traits<std::select2nd<std::pair<T0,T1> > >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct functor_traits<std::select1st<std::pair<T0,T1> > >
-{ enum { Cost = 0, PacketAccess = false }; };
-
-template<typename T0,typename T1>
-struct functor_traits<std::unary_compose<T0,T1> >
-{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost, PacketAccess = false }; };
-
-template<typename T0,typename T1,typename T2>
-struct functor_traits<std::binary_compose<T0,T1,T2> >
-{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost + functor_traits<T2>::Cost, PacketAccess = false }; };
-
-#endif // EIGEN_STDEXT_SUPPORT
-
-// allow to add new functors and specializations of functor_traits from outside Eigen.
-// this macro is really needed because functor_traits must be specialized after it is declared but before it is used...
-#ifdef EIGEN_FUNCTORS_PLUGIN
-#include EIGEN_FUNCTORS_PLUGIN
-#endif
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_FUNCTORS_H
diff --git a/Eigen/src/Core/Fuzzy.h b/Eigen/src/Core/Fuzzy.h
index fe63bd2..3e403a0 100644
--- a/Eigen/src/Core/Fuzzy.h
+++ b/Eigen/src/Core/Fuzzy.h
@@ -19,18 +19,19 @@
 template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
 struct isApprox_selector
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
   {
-    using std::min;
-    typename internal::nested<Derived,2>::type nested(x);
-    typename internal::nested<OtherDerived,2>::type otherNested(y);
-    return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * (min)(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum());
+    typename internal::nested_eval<Derived,2>::type nested(x);
+    typename internal::nested_eval<OtherDerived,2>::type otherNested(y);
+    return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * numext::mini(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum());
   }
 };
 
 template<typename Derived, typename OtherDerived>
 struct isApprox_selector<Derived, OtherDerived, true>
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar&)
   {
     return x.matrix() == y.matrix();
@@ -40,6 +41,7 @@
 template<typename Derived, typename OtherDerived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
 struct isMuchSmallerThan_object_selector
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec)
   {
     return x.cwiseAbs2().sum() <= numext::abs2(prec) * y.cwiseAbs2().sum();
@@ -49,6 +51,7 @@
 template<typename Derived, typename OtherDerived>
 struct isMuchSmallerThan_object_selector<Derived, OtherDerived, true>
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const OtherDerived&, const typename Derived::RealScalar&)
   {
     return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
@@ -58,6 +61,7 @@
 template<typename Derived, bool is_integer = NumTraits<typename Derived::Scalar>::IsInteger>
 struct isMuchSmallerThan_scalar_selector
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const typename Derived::RealScalar& y, const typename Derived::RealScalar& prec)
   {
     return x.cwiseAbs2().sum() <= numext::abs2(prec * y);
@@ -67,6 +71,7 @@
 template<typename Derived>
 struct isMuchSmallerThan_scalar_selector<Derived, true>
 {
+  EIGEN_DEVICE_FUNC
   static bool run(const Derived& x, const typename Derived::RealScalar&, const typename Derived::RealScalar&)
   {
     return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix();
diff --git a/Eigen/src/Core/GeneralProduct.h b/Eigen/src/Core/GeneralProduct.h
index 9e805a8..6f0cc80 100644
--- a/Eigen/src/Core/GeneralProduct.h
+++ b/Eigen/src/Core/GeneralProduct.h
@@ -11,29 +11,7 @@
 #ifndef EIGEN_GENERAL_PRODUCT_H
 #define EIGEN_GENERAL_PRODUCT_H
 
-namespace Eigen { 
-
-/** \class GeneralProduct
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the product of two general matrices or vectors
-  *
-  * \param LhsNested the type used to store the left-hand side
-  * \param RhsNested the type used to store the right-hand side
-  * \param ProductMode the type of the product
-  *
-  * This class represents an expression of the product of two general matrices.
-  * We call a general matrix, a dense matrix with full storage. For instance,
-  * This excludes triangular, selfadjoint, and sparse matrices.
-  * It is the return type of the operator* between general matrices. Its template
-  * arguments are determined automatically by ProductReturnType. Therefore,
-  * GeneralProduct should never be used direclty. To determine the result type of a
-  * function which involves a matrix product, use ProductReturnType::Type.
-  *
-  * \sa ProductReturnType, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
-  */
-template<typename Lhs, typename Rhs, int ProductType = internal::product_type<Lhs,Rhs>::value>
-class GeneralProduct;
+namespace Eigen {
 
 enum {
   Large = 2,
@@ -46,11 +24,17 @@
 
 template<int Size, int MaxSize> struct product_size_category
 {
-  enum { is_large = MaxSize == Dynamic ||
-                    Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD,
-         value = is_large  ? Large
-               : Size == 1 ? 1
-                           : Small
+  enum {
+    #ifndef EIGEN_CUDA_ARCH
+    is_large = MaxSize == Dynamic ||
+               Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD ||
+               (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD),
+    #else
+    is_large = 0,
+    #endif
+    value = is_large  ? Large
+          : Size == 1 ? 1
+                      : Small
   };
 };
 
@@ -59,15 +43,14 @@
   typedef typename remove_all<Lhs>::type _Lhs;
   typedef typename remove_all<Rhs>::type _Rhs;
   enum {
-    MaxRows  = _Lhs::MaxRowsAtCompileTime,
-    Rows  = _Lhs::RowsAtCompileTime,
-    MaxCols  = _Rhs::MaxColsAtCompileTime,
-    Cols  = _Rhs::ColsAtCompileTime,
-    MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime,
-                                           _Rhs::MaxRowsAtCompileTime),
-    Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime,
-                                        _Rhs::RowsAtCompileTime),
-    LargeThreshold = EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+    MaxRows = traits<_Lhs>::MaxRowsAtCompileTime,
+    Rows    = traits<_Lhs>::RowsAtCompileTime,
+    MaxCols = traits<_Rhs>::MaxColsAtCompileTime,
+    Cols    = traits<_Rhs>::ColsAtCompileTime,
+    MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::MaxColsAtCompileTime,
+                                           traits<_Rhs>::MaxRowsAtCompileTime),
+    Depth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::ColsAtCompileTime,
+                                        traits<_Rhs>::RowsAtCompileTime)
   };
 
   // the splitting into different lines of code here, introducing the _select enums and the typedef below,
@@ -82,7 +65,8 @@
 
 public:
   enum {
-    value = selector::ret
+    value = selector::ret,
+    ret = selector::ret
   };
 #ifdef EIGEN_DEBUG_PRODUCT
   static void debug()
@@ -98,12 +82,13 @@
 #endif
 };
 
-
 /* The following allows to select the kind of product at compile time
  * based on the three dimensions of the product.
  * This is a compile time mapping from {1,Small,Large}^3 -> {product types} */
 // FIXME I'm not sure the current mapping is the ideal one.
 template<int M, int N>  struct product_type_selector<M,N,1>              { enum { ret = OuterProduct }; };
+template<int M>         struct product_type_selector<M, 1, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
+template<int N>         struct product_type_selector<1, N, 1>            { enum { ret = LazyCoeffBasedProductMode }; };
 template<int Depth>     struct product_type_selector<1,    1,    Depth>  { enum { ret = InnerProduct }; };
 template<>              struct product_type_selector<1,    1,    1>      { enum { ret = InnerProduct }; };
 template<>              struct product_type_selector<Small,1,    Small>  { enum { ret = CoeffBasedProductMode }; };
@@ -122,60 +107,12 @@
 template<>              struct product_type_selector<Large,Small,Large>  { enum { ret = GemmProduct }; };
 template<>              struct product_type_selector<Small,Large,Large>  { enum { ret = GemmProduct }; };
 template<>              struct product_type_selector<Large,Large,Large>  { enum { ret = GemmProduct }; };
-template<>              struct product_type_selector<Large,Small,Small>  { enum { ret = GemmProduct }; };
-template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = GemmProduct }; };
+template<>              struct product_type_selector<Large,Small,Small>  { enum { ret = CoeffBasedProductMode }; };
+template<>              struct product_type_selector<Small,Large,Small>  { enum { ret = CoeffBasedProductMode }; };
 template<>              struct product_type_selector<Large,Large,Small>  { enum { ret = GemmProduct }; };
 
 } // end namespace internal
 
-/** \class ProductReturnType
-  * \ingroup Core_Module
-  *
-  * \brief Helper class to get the correct and optimized returned type of operator*
-  *
-  * \param Lhs the type of the left-hand side
-  * \param Rhs the type of the right-hand side
-  * \param ProductMode the type of the product (determined automatically by internal::product_mode)
-  *
-  * This class defines the typename Type representing the optimized product expression
-  * between two matrix expressions. In practice, using ProductReturnType<Lhs,Rhs>::Type
-  * is the recommended way to define the result type of a function returning an expression
-  * which involve a matrix product. The class Product should never be
-  * used directly.
-  *
-  * \sa class Product, MatrixBase::operator*(const MatrixBase<OtherDerived>&)
-  */
-template<typename Lhs, typename Rhs, int ProductType>
-struct ProductReturnType
-{
-  // TODO use the nested type to reduce instanciations ????
-//   typedef typename internal::nested<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
-//   typedef typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
-
-  typedef GeneralProduct<Lhs/*Nested*/, Rhs/*Nested*/, ProductType> Type;
-};
-
-template<typename Lhs, typename Rhs>
-struct ProductReturnType<Lhs,Rhs,CoeffBasedProductMode>
-{
-  typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
-  typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
-  typedef CoeffBasedProduct<LhsNested, RhsNested, EvalBeforeAssigningBit | EvalBeforeNestingBit> Type;
-};
-
-template<typename Lhs, typename Rhs>
-struct ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
-{
-  typedef typename internal::nested<Lhs, Rhs::ColsAtCompileTime, typename internal::plain_matrix_type<Lhs>::type >::type LhsNested;
-  typedef typename internal::nested<Rhs, Lhs::RowsAtCompileTime, typename internal::plain_matrix_type<Rhs>::type >::type RhsNested;
-  typedef CoeffBasedProduct<LhsNested, RhsNested, NestByRefBit> Type;
-};
-
-// this is a workaround for sun CC
-template<typename Lhs, typename Rhs>
-struct LazyProductReturnType : public ProductReturnType<Lhs,Rhs,LazyCoeffBasedProductMode>
-{};
-
 /***********************************************************************
 *  Implementation of Inner Vector Vector Product
 ***********************************************************************/
@@ -187,119 +124,10 @@
 // product ends up to a row-vector times col-vector product... To tackle this use
 // case, we could have a specialization for Block<MatrixType,1,1> with: operator=(Scalar x);
 
-namespace internal {
-
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,InnerProduct> >
- : traits<Matrix<typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> >
-{};
-
-}
-
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, InnerProduct>
-  : internal::no_assignment_operator,
-    public Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1>
-{
-    typedef Matrix<typename internal::scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType,1,1> Base;
-  public:
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs)
-    {
-      EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-      Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
-    }
-
-    /** Convertion to scalar */
-    operator const typename Base::Scalar() const {
-      return Base::coeff(0,0);
-    }
-};
-
 /***********************************************************************
 *  Implementation of Outer Vector Vector Product
 ***********************************************************************/
 
-namespace internal {
-
-// Column major
-template<typename ProductType, typename Dest, typename Func>
-EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const false_type&)
-{
-  typedef typename Dest::Index Index;
-  // FIXME make sure lhs is sequentially stored
-  // FIXME not very good if rhs is real and lhs complex while alpha is real too
-  const Index cols = dest.cols();
-  for (Index j=0; j<cols; ++j)
-    func(dest.col(j), prod.rhs().coeff(0,j) * prod.lhs());
-}
-
-// Row major
-template<typename ProductType, typename Dest, typename Func>
-EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const true_type&) {
-  typedef typename Dest::Index Index;
-  // FIXME make sure rhs is sequentially stored
-  // FIXME not very good if lhs is real and rhs complex while alpha is real too
-  const Index rows = dest.rows();
-  for (Index i=0; i<rows; ++i)
-    func(dest.row(i), prod.lhs().coeff(i,0) * prod.rhs());
-}
-
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,OuterProduct> >
- : traits<ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs> >
-{};
-
-}
-
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, OuterProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,OuterProduct>, Lhs, Rhs>
-{
-    template<typename T> struct IsRowMajor : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {};
-    
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {
-      EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::RealScalar, typename Rhs::RealScalar>::value),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-    }
-    
-    struct set  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived()  = src; } };
-    struct add  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
-    struct sub  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
-    struct adds {
-      Scalar m_scale;
-      adds(const Scalar& s) : m_scale(s) {}
-      template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const {
-        dst.const_cast_derived() += m_scale * src;
-      }
-    };
-    
-    template<typename Dest>
-    inline void evalTo(Dest& dest) const {
-      internal::outer_product_selector_run(*this, dest, set(), IsRowMajor<Dest>());
-    }
-    
-    template<typename Dest>
-    inline void addTo(Dest& dest) const {
-      internal::outer_product_selector_run(*this, dest, add(), IsRowMajor<Dest>());
-    }
-
-    template<typename Dest>
-    inline void subTo(Dest& dest) const {
-      internal::outer_product_selector_run(*this, dest, sub(), IsRowMajor<Dest>());
-    }
-
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
-    {
-      internal::outer_product_selector_run(*this, dest, adds(alpha), IsRowMajor<Dest>());
-    }
-};
-
 /***********************************************************************
 *  Implementation of General Matrix Vector Product
 ***********************************************************************/
@@ -313,60 +141,13 @@
  */
 namespace internal {
 
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,GemvProduct> >
- : traits<ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs> >
-{};
-
 template<int Side, int StorageOrder, bool BlasCompatible>
-struct gemv_selector;
+struct gemv_dense_selector;
 
 } // end namespace internal
 
-template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, GemvProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,GemvProduct>, Lhs, Rhs>
-{
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-
-    typedef typename Lhs::Scalar LhsScalar;
-    typedef typename Rhs::Scalar RhsScalar;
-
-    GeneralProduct(const Lhs& a_lhs, const Rhs& a_rhs) : Base(a_lhs,a_rhs)
-    {
-//       EIGEN_STATIC_ASSERT((internal::is_same<typename Lhs::Scalar, typename Rhs::Scalar>::value),
-//         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-    }
-
-    enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
-    typedef typename internal::conditional<int(Side)==OnTheRight,_LhsNested,_RhsNested>::type MatrixType;
-
-    template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
-    {
-      eigen_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols());
-      internal::gemv_selector<Side,(int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
-                       bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)>::run(*this, dst, alpha);
-    }
-};
-
 namespace internal {
 
-// The vector is on the left => transposition
-template<int StorageOrder, bool BlasCompatible>
-struct gemv_selector<OnTheLeft,StorageOrder,BlasCompatible>
-{
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
-  {
-    Transpose<Dest> destT(dest);
-    enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
-    gemv_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
-      ::run(GeneralProduct<Transpose<const typename ProductType::_RhsNested>,Transpose<const typename ProductType::_LhsNested>, GemvProduct>
-        (prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha);
-  }
-};
-
 template<typename Scalar,int Size,int MaxSize,bool Cond> struct gemv_static_vector_if;
 
 template<typename Scalar,int Size,int MaxSize>
@@ -384,126 +165,161 @@
 template<typename Scalar,int Size,int MaxSize>
 struct gemv_static_vector_if<Scalar,Size,MaxSize,true>
 {
-  #if EIGEN_ALIGN_STATICALLY
-  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0> m_data;
-  EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
-  #else
-  // Some architectures cannot align on the stack,
-  // => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
   enum {
     ForceAlignment  = internal::packet_traits<Scalar>::Vectorizable,
     PacketSize      = internal::packet_traits<Scalar>::size
   };
-  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?PacketSize:0),0> m_data;
+  #if EIGEN_MAX_STATIC_ALIGN_BYTES!=0
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize),0,EIGEN_PLAIN_ENUM_MIN(AlignedMax,PacketSize)> m_data;
+  EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; }
+  #else
+  // Some architectures cannot align on the stack,
+  // => let's manually enforce alignment by allocating more data and return the address of the first aligned element.
+  internal::plain_array<Scalar,EIGEN_SIZE_MIN_PREFER_FIXED(Size,MaxSize)+(ForceAlignment?EIGEN_MAX_ALIGN_BYTES:0),0> m_data;
   EIGEN_STRONG_INLINE Scalar* data() {
     return ForceAlignment
-            ? reinterpret_cast<Scalar*>((reinterpret_cast<size_t>(m_data.array) & ~(size_t(15))) + 16)
+            ? reinterpret_cast<Scalar*>((internal::UIntPtr(m_data.array) & ~(std::size_t(EIGEN_MAX_ALIGN_BYTES-1))) + EIGEN_MAX_ALIGN_BYTES)
             : m_data.array;
   }
   #endif
 };
 
-template<> struct gemv_selector<OnTheRight,ColMajor,true>
+// The vector is on the left => transposition
+template<int StorageOrder, bool BlasCompatible>
+struct gemv_dense_selector<OnTheLeft,StorageOrder,BlasCompatible>
 {
-  template<typename ProductType, typename Dest>
-  static inline void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::LhsScalar   LhsScalar;
-    typedef typename ProductType::RhsScalar   RhsScalar;
-    typedef typename ProductType::Scalar      ResScalar;
-    typedef typename ProductType::RealScalar  RealScalar;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-    typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
+    Transpose<Dest> destT(dest);
+    enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor };
+    gemv_dense_selector<OnTheRight,OtherStorageOrder,BlasCompatible>
+      ::run(rhs.transpose(), lhs.transpose(), destT, alpha);
+  }
+};
 
-    ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs());
-    ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs());
+template<> struct gemv_dense_selector<OnTheRight,ColMajor,true>
+{
+  template<typename Lhs, typename Rhs, typename Dest>
+  static inline void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
+  {
+    typedef typename Lhs::Scalar   LhsScalar;
+    typedef typename Rhs::Scalar   RhsScalar;
+    typedef typename Dest::Scalar  ResScalar;
+    typedef typename Dest::RealScalar  RealScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  
+    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
 
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    ActualLhsType actualLhs = LhsBlasTraits::extract(lhs);
+    ActualRhsType actualRhs = RhsBlasTraits::extract(rhs);
+
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(rhs);
+
+    // make sure Dest is a compile-time vector type (bug 1166)
+    typedef typename conditional<Dest::IsVectorAtCompileTime, Dest, typename Dest::ColXpr>::type ActualDest;
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
       // on, the other hand it is good for the cache to pack the vector anyways...
-      EvalToDestAtCompileTime = Dest::InnerStrideAtCompileTime==1,
+      EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1),
       ComplexByReal = (NumTraits<LhsScalar>::IsComplex) && (!NumTraits<RhsScalar>::IsComplex),
-      MightCannotUseDest = (Dest::InnerStrideAtCompileTime!=1) || ComplexByReal
+      MightCannotUseDest = (!EvalToDestAtCompileTime) || ComplexByReal
     };
 
-    gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
-
-    bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
-    bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
-    
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
     RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
-    ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
-                                                  evalToDest ? dest.data() : static_dest.data());
-    
-    if(!evalToDest)
+    if(!MightCannotUseDest)
     {
-      #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = dest.size();
-      EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      #endif
-      if(!alphaIsCompatible)
-      {
-        MappedDest(actualDestPtr, dest.size()).setZero();
-        compatibleAlpha = RhsScalar(1);
-      }
-      else
-        MappedDest(actualDestPtr, dest.size()) = dest;
+      // shortcut if we are sure to be able to use dest directly,
+      // this ease the compiler to generate cleaner and more optimzized code for most common cases
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          dest.data(), 1,
+          compatibleAlpha);
     }
-
-    general_matrix_vector_product
-      <Index,LhsScalar,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
-        actualLhs.rows(), actualLhs.cols(),
-        actualLhs.data(), actualLhs.outerStride(),
-        actualRhs.data(), actualRhs.innerStride(),
-        actualDestPtr, 1,
-        compatibleAlpha);
-
-    if (!evalToDest)
+    else
     {
-      if(!alphaIsCompatible)
-        dest += actualAlpha * MappedDest(actualDestPtr, dest.size());
-      else
-        dest = MappedDest(actualDestPtr, dest.size());
+      gemv_static_vector_if<ResScalar,ActualDest::SizeAtCompileTime,ActualDest::MaxSizeAtCompileTime,MightCannotUseDest> static_dest;
+
+      const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
+      const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
+
+      ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
+                                                    evalToDest ? dest.data() : static_dest.data());
+
+      if(!evalToDest)
+      {
+        #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        Index size = dest.size();
+        EIGEN_DENSE_STORAGE_CTOR_PLUGIN
+        #endif
+        if(!alphaIsCompatible)
+        {
+          MappedDest(actualDestPtr, dest.size()).setZero();
+          compatibleAlpha = RhsScalar(1);
+        }
+        else
+          MappedDest(actualDestPtr, dest.size()) = dest;
+      }
+
+      general_matrix_vector_product
+          <Index,LhsScalar,LhsMapper,ColMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
+          actualLhs.rows(), actualLhs.cols(),
+          LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+          RhsMapper(actualRhs.data(), actualRhs.innerStride()),
+          actualDestPtr, 1,
+          compatibleAlpha);
+
+      if (!evalToDest)
+      {
+        if(!alphaIsCompatible)
+          dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size());
+        else
+          dest = MappedDest(actualDestPtr, dest.size());
+      }
     }
   }
 };
 
-template<> struct gemv_selector<OnTheRight,RowMajor,true>
+template<> struct gemv_dense_selector<OnTheRight,RowMajor,true>
 {
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename ProductType::LhsScalar LhsScalar;
-    typedef typename ProductType::RhsScalar RhsScalar;
-    typedef typename ProductType::Scalar    ResScalar;
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::_ActualRhsType _ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
+    typedef typename Lhs::Scalar   LhsScalar;
+    typedef typename Rhs::Scalar   RhsScalar;
+    typedef typename Dest::Scalar  ResScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
 
-    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
-    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
+    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
+    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
 
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(lhs)
+                                  * RhsBlasTraits::extractScalarFactor(rhs);
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
       // on, the other hand it is good for the cache to pack the vector anyways...
-      DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
+      DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
     };
 
-    gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
+    gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
 
     ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
         DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
@@ -511,45 +327,48 @@
     if(!DirectlyUseRhs)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = actualRhs.size();
+      Index size = actualRhs.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
-      Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
+      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
     }
 
+    typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
     general_matrix_vector_product
-      <Index,LhsScalar,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsBlasTraits::NeedToConjugate>::run(
+        <Index,LhsScalar,LhsMapper,RowMajor,LhsBlasTraits::NeedToConjugate,RhsScalar,RhsMapper,RhsBlasTraits::NeedToConjugate>::run(
         actualLhs.rows(), actualLhs.cols(),
-        actualLhs.data(), actualLhs.outerStride(),
-        actualRhsPtr, 1,
-        dest.data(), dest.innerStride(),
+        LhsMapper(actualLhs.data(), actualLhs.outerStride()),
+        RhsMapper(actualRhsPtr, 1),
+        dest.data(), dest.col(0).innerStride(), //NOTE  if dest is not a vector at compile-time, then dest.innerStride() might be wrong. (bug 1166)
         actualAlpha);
   }
 };
 
-template<> struct gemv_selector<OnTheRight,ColMajor,false>
+template<> struct gemv_dense_selector<OnTheRight,ColMajor,false>
 {
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename Dest::Index Index;
-    // TODO makes sure dest is sequentially stored in memory, otherwise use a temp
-    const Index size = prod.rhs().rows();
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
+    // TODO if rhs is large enough it might be beneficial to make sure that dest is sequentially stored in memory, otherwise use a temp
+    typename nested_eval<Rhs,1>::type actual_rhs(rhs);
+    const Index size = rhs.rows();
     for(Index k=0; k<size; ++k)
-      dest += (alpha*prod.rhs().coeff(k)) * prod.lhs().col(k);
+      dest += (alpha*actual_rhs.coeff(k)) * lhs.col(k);
   }
 };
 
-template<> struct gemv_selector<OnTheRight,RowMajor,false>
+template<> struct gemv_dense_selector<OnTheRight,RowMajor,false>
 {
-  template<typename ProductType, typename Dest>
-  static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef typename Dest::Index Index;
-    // TODO makes sure rhs is sequentially stored in memory, otherwise use a temp
-    const Index rows = prod.rows();
+    EIGEN_STATIC_ASSERT((!nested_eval<Lhs,1>::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE);
+    typename nested_eval<Rhs,Lhs::RowsAtCompileTime>::type actual_rhs(rhs);
+    const Index rows = dest.rows();
     for(Index i=0; i<rows; ++i)
-      dest.coeffRef(i) += alpha * (prod.lhs().row(i).cwiseProduct(prod.rhs().transpose())).sum();
+      dest.coeffRef(i) += alpha * (lhs.row(i).cwiseProduct(actual_rhs.transpose())).sum();
   }
 };
 
@@ -567,7 +386,7 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-inline const typename ProductReturnType<Derived, OtherDerived>::Type
+inline const Product<Derived, OtherDerived>
 MatrixBase<Derived>::operator*(const MatrixBase<OtherDerived> &other) const
 {
   // A note regarding the function declaration: In MSVC, this function will sometimes
@@ -592,7 +411,8 @@
 #ifdef EIGEN_DEBUG_PRODUCT
   internal::product_type<Derived,OtherDerived>::debug();
 #endif
-  return typename ProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
+
+  return Product<Derived, OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation.
@@ -608,7 +428,7 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-const typename LazyProductReturnType<Derived,OtherDerived>::Type
+const Product<Derived,OtherDerived,LazyProduct>
 MatrixBase<Derived>::lazyProduct(const MatrixBase<OtherDerived> &other) const
 {
   enum {
@@ -627,7 +447,7 @@
     INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
   EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
 
-  return typename LazyProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
+  return Product<Derived,OtherDerived,LazyProduct>(derived(), other.derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/GenericPacketMath.h b/Eigen/src/Core/GenericPacketMath.h
index 5f783eb..029f8ac 100644
--- a/Eigen/src/Core/GenericPacketMath.h
+++ b/Eigen/src/Core/GenericPacketMath.h
@@ -42,21 +42,28 @@
 struct default_packet_traits
 {
   enum {
+    HasHalfPacket = 0,
+
     HasAdd    = 1,
     HasSub    = 1,
     HasMul    = 1,
     HasNegate = 1,
     HasAbs    = 1,
+    HasArg    = 0,
     HasAbs2   = 1,
     HasMin    = 1,
     HasMax    = 1,
     HasConj   = 1,
     HasSetLinear = 1,
+    HasBlend  = 0,
 
     HasDiv    = 0,
     HasSqrt   = 0,
+    HasRsqrt  = 0,
     HasExp    = 0,
     HasLog    = 0,
+    HasLog1p  = 0,
+    HasLog10  = 0,
     HasPow    = 0,
 
     HasSin    = 0,
@@ -64,17 +71,37 @@
     HasTan    = 0,
     HasASin   = 0,
     HasACos   = 0,
-    HasATan   = 0
+    HasATan   = 0,
+    HasSinh   = 0,
+    HasCosh   = 0,
+    HasTanh   = 0,
+    HasLGamma = 0,
+    HasDiGamma = 0,
+    HasZeta = 0,
+    HasPolygamma = 0,
+    HasErf = 0,
+    HasErfc = 0,
+    HasIGamma = 0,
+    HasIGammac = 0,
+    HasBetaInc = 0,
+
+    HasRound  = 0,
+    HasFloor  = 0,
+    HasCeil   = 0,
+
+    HasSign   = 0
   };
 };
 
 template<typename T> struct packet_traits : default_packet_traits
 {
   typedef T type;
+  typedef T half;
   enum {
     Vectorizable = 0,
     size = 1,
-    AlignedOnScalar = 0
+    AlignedOnScalar = 0,
+    HasHalfPacket = 0
   };
   enum {
     HasAdd    = 0,
@@ -90,135 +117,239 @@
   };
 };
 
+template<typename T> struct packet_traits<const T> : packet_traits<T> { };
+
+template <typename Src, typename Tgt> struct type_casting_traits {
+  enum {
+    VectorizedCast = 0,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+
+/** \internal \returns static_cast<TgtType>(a) (coeff-wise) */
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a) {
+  return static_cast<TgtPacket>(a);
+}
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a, const SrcPacket& /*b*/) {
+  return static_cast<TgtPacket>(a);
+}
+
+template <typename SrcPacket, typename TgtPacket>
+EIGEN_DEVICE_FUNC inline TgtPacket
+pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) {
+  return static_cast<TgtPacket>(a);
+}
+
 /** \internal \returns a + b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 padd(const Packet& a,
         const Packet& b) { return a+b; }
 
 /** \internal \returns a - b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 psub(const Packet& a,
         const Packet& b) { return a-b; }
 
 /** \internal \returns -a (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pnegate(const Packet& a) { return -a; }
 
 /** \internal \returns conj(a) (coeff-wise) */
-template<typename Packet> inline Packet
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pconj(const Packet& a) { return numext::conj(a); }
 
 /** \internal \returns a * b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmul(const Packet& a,
         const Packet& b) { return a*b; }
 
 /** \internal \returns a / b (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pdiv(const Packet& a,
         const Packet& b) { return a/b; }
 
 /** \internal \returns the min of \a a and \a b  (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmin(const Packet& a,
-        const Packet& b) { using std::min; return (min)(a, b); }
+        const Packet& b) { return numext::mini(a, b); }
 
 /** \internal \returns the max of \a a and \a b  (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmax(const Packet& a,
-        const Packet& b) { using std::max; return (max)(a, b); }
+        const Packet& b) { return numext::maxi(a, b); }
 
 /** \internal \returns the absolute value of \a a */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pabs(const Packet& a) { using std::abs; return abs(a); }
 
+/** \internal \returns the phase angle of \a a */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+parg(const Packet& a) { using numext::arg; return arg(a); }
+
 /** \internal \returns the bitwise and of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pand(const Packet& a, const Packet& b) { return a & b; }
 
 /** \internal \returns the bitwise or of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 por(const Packet& a, const Packet& b) { return a | b; }
 
 /** \internal \returns the bitwise xor of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pxor(const Packet& a, const Packet& b) { return a ^ b; }
 
 /** \internal \returns the bitwise andnot of \a a and \a b */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pandnot(const Packet& a, const Packet& b) { return a & (!b); }
 
 /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pload(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
 /** \internal \returns a packet version of \a *from, (un-aligned load) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 ploadu(const typename unpacket_traits<Packet>::type* from) { return *from; }
 
-/** \internal \returns a packet with elements of \a *from duplicated.
-  * For instance, for a packet of 8 elements, 4 scalar will be read from \a *from and
-  * duplicated to form: {from[0],from[0],from[1],from[1],,from[2],from[2],,from[3],from[3]}
-  * Currently, this function is only used for scalar * complex products.
- */
-template<typename Packet> inline Packet
-ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
-
 /** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pset1(const typename unpacket_traits<Packet>::type& a) { return a; }
 
+/** \internal \returns a packet with constant coefficients \a a[0], e.g.: (a[0],a[0],a[0],a[0]) */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pload1(const typename unpacket_traits<Packet>::type  *a) { return pset1<Packet>(*a); }
+
+/** \internal \returns a packet with elements of \a *from duplicated.
+  * For instance, for a packet of 8 elements, 4 scalars will be read from \a *from and
+  * duplicated to form: {from[0],from[0],from[1],from[1],from[2],from[2],from[3],from[3]}
+  * Currently, this function is only used for scalar * complex products.
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
+ploaddup(const typename unpacket_traits<Packet>::type* from) { return *from; }
+
+/** \internal \returns a packet with elements of \a *from quadrupled.
+  * For instance, for a packet of 8 elements, 2 scalars will be read from \a *from and
+  * replicated to form: {from[0],from[0],from[0],from[0],from[1],from[1],from[1],from[1]}
+  * Currently, this function is only used in matrix products.
+  * For packet-size smaller or equal to 4, this function is equivalent to pload1 
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+ploadquad(const typename unpacket_traits<Packet>::type* from)
+{ return pload1<Packet>(from); }
+
+/** \internal equivalent to
+  * \code
+  * a0 = pload1(a+0);
+  * a1 = pload1(a+1);
+  * a2 = pload1(a+2);
+  * a3 = pload1(a+3);
+  * \endcode
+  * \sa pset1, pload1, ploaddup, pbroadcast2
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC
+inline void pbroadcast4(const typename unpacket_traits<Packet>::type *a,
+                        Packet& a0, Packet& a1, Packet& a2, Packet& a3)
+{
+  a0 = pload1<Packet>(a+0);
+  a1 = pload1<Packet>(a+1);
+  a2 = pload1<Packet>(a+2);
+  a3 = pload1<Packet>(a+3);
+}
+
+/** \internal equivalent to
+  * \code
+  * a0 = pload1(a+0);
+  * a1 = pload1(a+1);
+  * \endcode
+  * \sa pset1, pload1, ploaddup, pbroadcast4
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC
+inline void pbroadcast2(const typename unpacket_traits<Packet>::type *a,
+                        Packet& a0, Packet& a1)
+{
+  a0 = pload1<Packet>(a+0);
+  a1 = pload1<Packet>(a+1);
+}
+
 /** \internal \brief Returns a packet with coefficients (a,a+1,...,a+packet_size-1). */
-template<typename Scalar> inline typename packet_traits<Scalar>::type
-plset(const Scalar& a) { return a; }
+template<typename Packet> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet
+plset(const typename unpacket_traits<Packet>::type& a) { return a; }
 
 /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */
-template<typename Scalar, typename Packet> inline void pstore(Scalar* to, const Packet& from)
+template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from)
 { (*to) = from; }
 
 /** \internal copy the packet \a from to \a *to, (un-aligned store) */
-template<typename Scalar, typename Packet> inline void pstoreu(Scalar* to, const Packet& from)
-{ (*to) = from; }
+template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from)
+{  (*to) = from; }
+
+ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, Index /*stride*/)
+ { return ploadu<Packet>(from); }
+
+ template<typename Scalar, typename Packet> EIGEN_DEVICE_FUNC inline void pscatter(Scalar* to, const Packet& from, Index /*stride*/)
+ { pstore(to, from); }
 
 /** \internal tries to do cache prefetching of \a addr */
-template<typename Scalar> inline void prefetch(const Scalar* addr)
+template<typename Scalar> EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* addr)
 {
-#if !defined(_MSC_VER)
-__builtin_prefetch(addr);
+#ifdef __CUDA_ARCH__
+#if defined(__LP64__)
+  // 64-bit pointer operand constraint for inlined asm
+  asm(" prefetch.L1 [ %1 ];" : "=l"(addr) : "l"(addr));
+#else
+  // 32-bit pointer operand constraint for inlined asm
+  asm(" prefetch.L1 [ %1 ];" : "=r"(addr) : "r"(addr));
+#endif
+#elif (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC)
+  __builtin_prefetch(addr);
 #endif
 }
 
 /** \internal \returns the first element of a packet */
-template<typename Packet> inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type pfirst(const Packet& a)
 { return a; }
 
 /** \internal \returns a packet where the element i contains the sum of the packet of \a vec[i] */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 preduxp(const Packet* vecs) { return vecs[0]; }
 
 /** \internal \returns the sum of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux(const Packet& a)
+{ return a; }
+
+/** \internal \returns the sum of the elements of \a a by block of 4 elements.
+  * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7}
+  * For packet-size smaller or equal to 4, this boils down to a noop.
+  */
+template<typename Packet> EIGEN_DEVICE_FUNC inline
+typename conditional<(unpacket_traits<Packet>::size%8)==0,typename unpacket_traits<Packet>::half,Packet>::type
+predux_downto4(const Packet& a)
 { return a; }
 
 /** \internal \returns the product of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(const Packet& a)
 { return a; }
 
 /** \internal \returns the min of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(const Packet& a)
 { return a; }
 
 /** \internal \returns the max of the elements of \a a*/
-template<typename Packet> inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(const Packet& a)
 { return a; }
 
 /** \internal \returns the reversed elements of \a a*/
-template<typename Packet> inline Packet preverse(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a)
 { return a; }
 
-
 /** \internal \returns \a a with real and imaginary part flipped (for complex type only) */
-template<typename Packet> inline Packet pcplxflip(const Packet& a)
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a)
 {
   // FIXME: uncomment the following in case we drop the internal imag and real functions.
 //   using std::imag;
@@ -250,6 +381,22 @@
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pacos(const Packet& a) { using std::acos; return acos(a); }
 
+/** \internal \returns the arc tangent of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet patan(const Packet& a) { using std::atan; return atan(a); }
+
+/** \internal \returns the hyperbolic sine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet psinh(const Packet& a) { using std::sinh; return sinh(a); }
+
+/** \internal \returns the hyperbolic cosine of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pcosh(const Packet& a) { using std::cosh; return cosh(a); }
+
+/** \internal \returns the hyperbolic tan of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet ptanh(const Packet& a) { using std::tanh; return tanh(a); }
+
 /** \internal \returns the exp of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet pexp(const Packet& a) { using std::exp; return exp(a); }
@@ -258,10 +405,36 @@
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet plog(const Packet& a) { using std::log; return log(a); }
 
+/** \internal \returns the log1p of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plog1p(const Packet& a) { return numext::log1p(a); }
+
+/** \internal \returns the log10 of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet plog10(const Packet& a) { using std::log10; return log10(a); }
+
 /** \internal \returns the square-root of \a a (coeff-wise) */
 template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
 Packet psqrt(const Packet& a) { using std::sqrt; return sqrt(a); }
 
+/** \internal \returns the reciprocal square-root of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet prsqrt(const Packet& a) {
+  return pdiv(pset1<Packet>(1), psqrt(a));
+}
+
+/** \internal \returns the rounded value of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pround(const Packet& a) { using numext::round; return round(a); }
+
+/** \internal \returns the floor of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pfloor(const Packet& a) { using numext::floor; return floor(a); }
+
+/** \internal \returns the ceil of \a a (coeff-wise) */
+template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
+Packet pceil(const Packet& a) { using numext::ceil; return ceil(a); }
+
 /***************************************************************************
 * The following functions might not have to be overwritten for vectorized types
 ***************************************************************************/
@@ -275,34 +448,45 @@
 }
 
 /** \internal \returns a * b + c (coeff-wise) */
-template<typename Packet> inline Packet
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
 pmadd(const Packet&  a,
          const Packet&  b,
          const Packet&  c)
 { return padd(pmul(a, b),c); }
 
 /** \internal \returns a packet version of \a *from.
-  * If LoadMode equals #Aligned, \a from must be 16 bytes aligned */
-template<typename Packet, int LoadMode>
-inline Packet ploadt(const typename unpacket_traits<Packet>::type* from)
+  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
+template<typename Packet, int Alignment>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt(const typename unpacket_traits<Packet>::type* from)
 {
-  if(LoadMode == Aligned)
+  if(Alignment >= unpacket_traits<Packet>::alignment)
     return pload<Packet>(from);
   else
     return ploadu<Packet>(from);
 }
 
 /** \internal copy the packet \a from to \a *to.
-  * If StoreMode equals #Aligned, \a to must be 16 bytes aligned */
-template<typename Scalar, typename Packet, int LoadMode>
-inline void pstoret(Scalar* to, const Packet& from)
+  * The pointer \a from must be aligned on a \a Alignment bytes boundary. */
+template<typename Scalar, typename Packet, int Alignment>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& from)
 {
-  if(LoadMode == Aligned)
+  if(Alignment >= unpacket_traits<Packet>::alignment)
     pstore(to, from);
   else
     pstoreu(to, from);
 }
 
+/** \internal \returns a packet version of \a *from.
+  * Unlike ploadt, ploadt_ro takes advantage of the read-only memory path on the
+  * hardware if available to speedup the loading of data that won't be modified
+  * by the current computation.
+  */
+template<typename Packet, int LoadMode>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_ro(const typename unpacket_traits<Packet>::type* from)
+{
+  return ploadt<Packet, LoadMode>(from);
+}
+
 /** \internal default implementation of palign() allowing partial specialization */
 template<int Offset,typename PacketType>
 struct palign_impl
@@ -336,15 +520,74 @@
 * Fast complex products (GCC generates a function call which is very slow)
 ***************************************************************************/
 
+// Eigen+CUDA does not support complexes.
+#ifndef __CUDACC__
+
 template<> inline std::complex<float> pmul(const std::complex<float>& a, const std::complex<float>& b)
 { return std::complex<float>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
 
 template<> inline std::complex<double> pmul(const std::complex<double>& a, const std::complex<double>& b)
 { return std::complex<double>(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); }
 
+#endif
+
+
+/***************************************************************************
+ * PacketBlock, that is a collection of N packets where the number of words
+ * in the packet is a multiple of N.
+***************************************************************************/
+template <typename Packet,int N=unpacket_traits<Packet>::size> struct PacketBlock {
+  Packet packet[N];
+};
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet,1>& /*kernel*/) {
+  // Nothing to do in the scalar case, i.e. a 1x1 matrix.
+}
+
+/***************************************************************************
+ * Selector, i.e. vector of N boolean values used to select (i.e. blend)
+ * words from 2 packets.
+***************************************************************************/
+template <size_t N> struct Selector {
+  bool select[N];
+};
+
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pblend(const Selector<unpacket_traits<Packet>::size>& ifPacket, const Packet& thenPacket, const Packet& elsePacket) {
+  return ifPacket.select[0] ? thenPacket : elsePacket;
+}
+
+/** \internal \returns \a a with the first coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertfirst(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  mask.select[0] = true;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=1; i<unpacket_traits<Packet>::size; ++i)
+    mask.select[i] = false;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
+/** \internal \returns \a a with the last coefficient replaced by the scalar b */
+template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
+pinsertlast(const Packet& a, typename unpacket_traits<Packet>::type b)
+{
+  // Default implementation based on pblend.
+  // It must be specialized for higher performance.
+  Selector<unpacket_traits<Packet>::size> mask;
+  // This for loop should be optimized away by the compiler.
+  for(Index i=0; i<unpacket_traits<Packet>::size-1; ++i)
+    mask.select[i] = false;
+  mask.select[unpacket_traits<Packet>::size-1] = true;
+  return pblend(mask, pset1<Packet>(b), a);
+}
+
 } // end namespace internal
 
 } // end namespace Eigen
 
 #endif // EIGEN_GENERIC_PACKET_MATH_H
-
diff --git a/Eigen/src/Core/GlobalFunctions.h b/Eigen/src/Core/GlobalFunctions.h
index 2acf977..769dc25 100644
--- a/Eigen/src/Core/GlobalFunctions.h
+++ b/Eigen/src/Core/GlobalFunctions.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2010-2012 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -11,13 +11,30 @@
 #ifndef EIGEN_GLOBAL_FUNCTIONS_H
 #define EIGEN_GLOBAL_FUNCTIONS_H
 
-#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR) \
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
+  /** \returns an expression of the coefficient-wise DOC_OP of \a x
+
+    DOC_DETAILS
+
+    \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_##NAME">Math functions</a>, class CwiseUnaryOp
+    */ \
   template<typename Derived> \
   inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
-  NAME(const Eigen::ArrayBase<Derived>& x) { \
-    return x.derived(); \
+  NAME(const Eigen::ArrayBase<Derived>& x);
+
+#else
+
+#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \
+  template<typename Derived> \
+  inline const Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived> \
+  (NAME)(const Eigen::ArrayBase<Derived>& x) { \
+    return Eigen::CwiseUnaryOp<Eigen::internal::FUNCTOR<typename Derived::Scalar>, const Derived>(x.derived()); \
   }
 
+#endif // EIGEN_PARSED_BY_DOXYGEN
+
 #define EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(NAME,FUNCTOR) \
   \
   template<typename Derived> \
@@ -30,55 +47,133 @@
   { \
     static inline typename NAME##_retval<ArrayBase<Derived> >::type run(const Eigen::ArrayBase<Derived>& x) \
     { \
-      return x.derived(); \
+      return typename NAME##_retval<ArrayBase<Derived> >::type(x.derived()); \
     } \
   };
 
-
 namespace Eigen
 {
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op)
-  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op,real part,\sa ArrayBase::real)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op,imaginary part,\sa ArrayBase::imag)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op,complex conjugate,\sa ArrayBase::conjugate)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(inverse,scalar_inverse_op,inverse,\sa ArrayBase::inverse)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op,sine,\sa ArrayBase::sin)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op,cosine,\sa ArrayBase::cos)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op,tangent,\sa ArrayBase::tan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atan,scalar_atan_op,arc-tangent,\sa ArrayBase::atan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op,arc-sine,\sa ArrayBase::asin)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op,arc-consine,\sa ArrayBase::acos)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op,hyperbolic sine,\sa ArrayBase::sinh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op,hyperbolic cosine,\sa ArrayBase::cosh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op,hyperbolic tangent,\sa ArrayBase::tanh)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op,not-a-number test,\sa Eigen::isinf DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isnan)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op,infinite value test,\sa Eigen::isnan DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isinf)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op,finite value test,\sa Eigen::isinf DOXCOMMA Eigen::isnan DOXCOMMA ArrayBase::isfinite)
+  EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op,sign (or 0),\sa ArrayBase::sign)
   
-  template<typename Derived>
-  inline const Eigen::CwiseUnaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar>, const Derived>
-  pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) {
+  /** \returns an expression of the coefficient-wise power of \a x to the given constant \a exponent.
+    *
+    * \tparam ScalarExponent is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression (\c Derived::Scalar).
+    *
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+  template<typename Derived,typename ScalarExponent>
+  inline const CwiseBinaryOp<internal::scalar_pow_op<Derived::Scalar,ScalarExponent>,Derived,Constant<ScalarExponent> >
+  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent);
+#else
+  template<typename Derived,typename ScalarExponent>
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,ScalarExponent>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,typename Derived::Scalar,ScalarExponent),
+          const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,ScalarExponent,pow) >::type
+  pow(const Eigen::ArrayBase<Derived>& x, const ScalarExponent& exponent) {
     return x.derived().pow(exponent);
   }
 
   template<typename Derived>
-  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const Derived, const Derived>
-  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<Derived>& exponents) 
+  inline const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename Derived::Scalar,pow)
+  pow(const Eigen::ArrayBase<Derived>& x, const typename Derived::Scalar& exponent) {
+    return x.derived().pow(exponent);
+  }
+#endif
+
+  /** \returns an expression of the coefficient-wise power of \a x to the given array of \a exponents.
+    *
+    * This function computes the coefficient-wise power.
+    *
+    * Example: \include Cwise_array_power_array.cpp
+    * Output: \verbinclude Cwise_array_power_array.out
+    * 
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+  template<typename Derived,typename ExponentDerived>
+  inline const Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>
+  pow(const Eigen::ArrayBase<Derived>& x, const Eigen::ArrayBase<ExponentDerived>& exponents) 
   {
-    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_binary_pow_op<typename Derived::Scalar, typename Derived::Scalar>, const Derived, const Derived>(
+    return Eigen::CwiseBinaryOp<Eigen::internal::scalar_pow_op<typename Derived::Scalar, typename ExponentDerived::Scalar>, const Derived, const ExponentDerived>(
       x.derived(),
       exponents.derived()
     );
   }
   
-  /**
-  * \brief Component-wise division of a scalar by array elements.
-  **/
-  template <typename Derived>
-  inline const Eigen::CwiseUnaryOp<Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>, const Derived>
-    operator/(const typename Derived::Scalar& s, const Eigen::ArrayBase<Derived>& a)
+  /** \returns an expression of the coefficient-wise power of the scalar \a x to the given array of \a exponents.
+    *
+    * This function computes the coefficient-wise power between a scalar and an array of exponents.
+    *
+    * \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar).
+    *
+    * Example: \include Cwise_scalar_power_array.cpp
+    * Output: \verbinclude Cwise_scalar_power_array.out
+    * 
+    * \sa ArrayBase::pow()
+    *
+    * \relates ArrayBase
+    */
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+  template<typename Scalar,typename Derived>
+  inline const CwiseBinaryOp<internal::scalar_pow_op<Scalar,Derived::Scalar>,Constant<Scalar>,Derived>
+  pow(const Scalar& x,const Eigen::ArrayBase<Derived>& x);
+#else
+  template<typename Scalar, typename Derived>
+  inline typename internal::enable_if<   !(internal::is_same<typename Derived::Scalar,Scalar>::value) && EIGEN_SCALAR_BINARY_SUPPORTED(pow,Scalar,typename Derived::Scalar),
+          const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow) >::type
+  pow(const Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
   {
-    return Eigen::CwiseUnaryOp<Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>, const Derived>(
-      a.derived(),
-      Eigen::internal::scalar_inverse_mult_op<typename Derived::Scalar>(s)  
-    );
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,Derived,pow)(
+            typename internal::plain_constant_type<Derived,Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
   }
 
+  template<typename Derived>
+  inline const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)
+  pow(const typename Derived::Scalar& x, const Eigen::ArrayBase<Derived>& exponents)
+  {
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename Derived::Scalar,Derived,pow)(
+      typename internal::plain_constant_type<Derived,typename Derived::Scalar>::type(exponents.rows(), exponents.cols(), x), exponents.derived() );
+  }
+#endif
+
+
   namespace internal
   {
     EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(real,scalar_real_op)
diff --git a/Eigen/src/Core/IO.h b/Eigen/src/Core/IO.h
index 8d4bc59..da7fd6c 100644
--- a/Eigen/src/Core/IO.h
+++ b/Eigen/src/Core/IO.h
@@ -49,7 +49,7 @@
   */
 struct IOFormat
 {
-  /** Default contructor, see class IOFormat for the meaning of the parameters */
+  /** Default constructor, see class IOFormat for the meaning of the parameters */
   IOFormat(int _precision = StreamPrecision, int _flags = 0,
     const std::string& _coeffSeparator = " ",
     const std::string& _rowSeparator = "\n", const std::string& _rowPrefix="", const std::string& _rowSuffix="",
@@ -57,6 +57,10 @@
   : matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator),
     rowSpacer(""), coeffSeparator(_coeffSeparator), precision(_precision), flags(_flags)
   {
+    // TODO check if rowPrefix, rowSuffix or rowSeparator contains a newline
+    // don't add rowSpacer if columns are not to be aligned
+    if((flags & DontAlignCols))
+      return;
     int i = int(matSuffix.length())-1;
     while (i>=0 && matSuffix[i]!='\n')
     {
@@ -76,7 +80,7 @@
   *
   * \brief Pseudo expression providing matrix output with given format
   *
-  * \param ExpressionType the type of the object on which IO stream operations are performed
+  * \tparam ExpressionType the type of the object on which IO stream operations are performed
   *
   * This class represents an expression with stream operators controlled by a given IOFormat.
   * It is the return type of DenseBase::format()
@@ -101,51 +105,23 @@
     }
 
   protected:
-    const typename ExpressionType::Nested m_matrix;
+    typename ExpressionType::Nested m_matrix;
     IOFormat m_format;
 };
 
-/** \returns a WithFormat proxy object allowing to print a matrix the with given
-  * format \a fmt.
-  *
-  * See class IOFormat for some examples.
-  *
-  * \sa class IOFormat, class WithFormat
-  */
-template<typename Derived>
-inline const WithFormat<Derived>
-DenseBase<Derived>::format(const IOFormat& fmt) const
-{
-  return WithFormat<Derived>(derived(), fmt);
-}
-
 namespace internal {
 
-template<typename Scalar, bool IsInteger>
-struct significant_decimals_default_impl
-{
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  static inline int run()
-  {
-    using std::ceil;
-    using std::log;
-    return cast<RealScalar,int>(ceil(-log(NumTraits<RealScalar>::epsilon())/log(RealScalar(10))));
-  }
-};
-
-template<typename Scalar>
-struct significant_decimals_default_impl<Scalar, true>
-{
-  static inline int run()
-  {
-    return 0;
-  }
-};
-
+// NOTE: This helper is kept for backward compatibility with previous code specializing
+//       this internal::significant_decimals_impl structure. In the future we should directly
+//       call digits10() which has been introduced in July 2016 in 3.3.
 template<typename Scalar>
 struct significant_decimals_impl
-  : significant_decimals_default_impl<Scalar, NumTraits<Scalar>::IsInteger>
-{};
+{
+  static inline int run()
+  {
+    return NumTraits<Scalar>::digits10();
+  }
+};
 
 /** \internal
   * print the matrix \a _m to the output stream \a s using the output format \a fmt */
@@ -160,7 +136,6 @@
   
   typename Derived::Nested m = _m;
   typedef typename Derived::Scalar Scalar;
-  typedef typename Derived::Index Index;
 
   Index width = 0;
 
diff --git a/Eigen/src/Core/Inverse.h b/Eigen/src/Core/Inverse.h
new file mode 100644
index 0000000..b76f043
--- /dev/null
+++ b/Eigen/src/Core/Inverse.h
@@ -0,0 +1,118 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_INVERSE_H
+#define EIGEN_INVERSE_H
+
+namespace Eigen { 
+
+template<typename XprType,typename StorageKind> class InverseImpl;
+
+namespace internal {
+
+template<typename XprType>
+struct traits<Inverse<XprType> >
+  : traits<typename XprType::PlainObject>
+{
+  typedef typename XprType::PlainObject PlainObject;
+  typedef traits<PlainObject> BaseTraits;
+  enum {
+    Flags = BaseTraits::Flags & RowMajorBit
+  };
+};
+
+} // end namespace internal
+
+/** \class Inverse
+  *
+  * \brief Expression of the inverse of another expression
+  *
+  * \tparam XprType the type of the expression we are taking the inverse
+  *
+  * This class represents an abstract expression of A.inverse()
+  * and most of the time this is the only way it is used.
+  *
+  */
+template<typename XprType>
+class Inverse : public InverseImpl<XprType,typename internal::traits<XprType>::StorageKind>
+{
+public:
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename XprType::PlainObject                       PlainObject;
+  typedef typename XprType::Scalar                            Scalar;
+  typedef typename internal::ref_selector<XprType>::type      XprTypeNested;
+  typedef typename internal::remove_all<XprTypeNested>::type  XprTypeNestedCleaned;
+  typedef typename internal::ref_selector<Inverse>::type Nested;
+  typedef typename internal::remove_all<XprType>::type NestedExpression;
+  
+  explicit EIGEN_DEVICE_FUNC Inverse(const XprType &xpr)
+    : m_xpr(xpr)
+  {}
+
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+
+  EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; }
+
+protected:
+  XprTypeNested m_xpr;
+};
+
+// Generic API dispatcher
+template<typename XprType, typename StorageKind>
+class InverseImpl
+  : public internal::generic_xpr_base<Inverse<XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<Inverse<XprType> >::type Base;
+  typedef typename XprType::Scalar Scalar;
+private:
+
+  Scalar coeff(Index row, Index col) const;
+  Scalar coeff(Index i) const;
+};
+
+namespace internal {
+
+/** \internal
+  * \brief Default evaluator for Inverse expression.
+  * 
+  * This default evaluator for Inverse expression simply evaluate the inverse into a temporary
+  * by a call to internal::call_assignment_no_alias.
+  * Therefore, inverse implementers only have to specialize Assignment<Dst,Inverse<...>, ...> for
+  * there own nested expression.
+  *
+  * \sa class Inverse
+  */
+template<typename ArgType>
+struct unary_evaluator<Inverse<ArgType> >
+  : public evaluator<typename Inverse<ArgType>::PlainObject>
+{
+  typedef Inverse<ArgType> InverseType;
+  typedef typename InverseType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+  
+  enum { Flags = Base::Flags | EvalBeforeNestingBit };
+
+  unary_evaluator(const InverseType& inv_xpr)
+    : m_result(inv_xpr.rows(), inv_xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    internal::call_assignment_no_alias(m_result, inv_xpr);
+  }
+  
+protected:
+  PlainObject m_result;
+};
+  
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_INVERSE_H
diff --git a/Eigen/src/Core/Map.h b/Eigen/src/Core/Map.h
index f804c89..548bf9a 100644
--- a/Eigen/src/Core/Map.h
+++ b/Eigen/src/Core/Map.h
@@ -13,13 +13,41 @@
 
 namespace Eigen { 
 
+namespace internal {
+template<typename PlainObjectType, int MapOptions, typename StrideType>
+struct traits<Map<PlainObjectType, MapOptions, StrideType> >
+  : public traits<PlainObjectType>
+{
+  typedef traits<PlainObjectType> TraitsBase;
+  enum {
+    PlainObjectTypeInnerSize = ((traits<PlainObjectType>::Flags&RowMajorBit)==RowMajorBit)
+                             ? PlainObjectType::ColsAtCompileTime
+                             : PlainObjectType::RowsAtCompileTime,
+
+    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
+                             ? int(PlainObjectType::InnerStrideAtCompileTime)
+                             : int(StrideType::InnerStrideAtCompileTime),
+    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
+                             ? (InnerStrideAtCompileTime==Dynamic || PlainObjectTypeInnerSize==Dynamic
+                                ? Dynamic
+                                : int(InnerStrideAtCompileTime) * int(PlainObjectTypeInnerSize))
+                             : int(StrideType::OuterStrideAtCompileTime),
+    Alignment = int(MapOptions)&int(AlignedMask),
+    Flags0 = TraitsBase::Flags & (~NestByRefBit),
+    Flags = is_lvalue<PlainObjectType>::value ? int(Flags0) : (int(Flags0) & ~LvalueBit)
+  };
+private:
+  enum { Options }; // Expressions don't have Options
+};
+}
+
 /** \class Map
   * \ingroup Core_Module
   *
   * \brief A matrix or vector expression mapping an existing array of data.
   *
   * \tparam PlainObjectType the equivalent matrix type of the mapped data
-  * \tparam MapOptions specifies whether the pointer is \c #Aligned, or \c #Unaligned.
+  * \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
   *                The default is \c #Unaligned.
   * \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout
   *                   of an ordinary, contiguous array. This can be overridden by specifying strides.
@@ -63,44 +91,6 @@
   *
   * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
   */
-
-namespace internal {
-template<typename PlainObjectType, int MapOptions, typename StrideType>
-struct traits<Map<PlainObjectType, MapOptions, StrideType> >
-  : public traits<PlainObjectType>
-{
-  typedef traits<PlainObjectType> TraitsBase;
-  typedef typename PlainObjectType::Index Index;
-  typedef typename PlainObjectType::Scalar Scalar;
-  enum {
-    InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0
-                             ? int(PlainObjectType::InnerStrideAtCompileTime)
-                             : int(StrideType::InnerStrideAtCompileTime),
-    OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0
-                             ? int(PlainObjectType::OuterStrideAtCompileTime)
-                             : int(StrideType::OuterStrideAtCompileTime),
-    HasNoInnerStride = InnerStrideAtCompileTime == 1,
-    HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0,
-    HasNoStride = HasNoInnerStride && HasNoOuterStride,
-    IsAligned = bool(EIGEN_ALIGN) && ((int(MapOptions)&Aligned)==Aligned),
-    IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic,
-    KeepsPacketAccess = bool(HasNoInnerStride)
-                        && ( bool(IsDynamicSize)
-                           || HasNoOuterStride
-                           || ( OuterStrideAtCompileTime!=Dynamic
-                           && ((static_cast<int>(sizeof(Scalar))*OuterStrideAtCompileTime)%16)==0 ) ),
-    Flags0 = TraitsBase::Flags & (~NestByRefBit),
-    Flags1 = IsAligned ? (int(Flags0) | AlignedBit) : (int(Flags0) & ~AlignedBit),
-    Flags2 = (bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime))
-           ? int(Flags1) : int(Flags1 & ~LinearAccessBit),
-    Flags3 = is_lvalue<PlainObjectType>::value ? int(Flags2) : (int(Flags2) & ~LvalueBit),
-    Flags = KeepsPacketAccess ? int(Flags3) : (int(Flags3) & ~PacketAccessBit)
-  };
-private:
-  enum { Options }; // Expressions don't have Options
-};
-}
-
 template<typename PlainObjectType, int MapOptions, typename StrideType> class Map
   : public MapBase<Map<PlainObjectType, MapOptions, StrideType> >
 {
@@ -110,34 +100,34 @@
     EIGEN_DENSE_PUBLIC_INTERFACE(Map)
 
     typedef typename Base::PointerType PointerType;
-#if EIGEN2_SUPPORT_STAGE <= STAGE30_FULL_EIGEN3_API
-    typedef const Scalar* PointerArgType;
-    inline PointerType cast_to_pointer_type(PointerArgType ptr) { return const_cast<PointerType>(ptr); }
-#else
     typedef PointerType PointerArgType;
+    EIGEN_DEVICE_FUNC
     inline PointerType cast_to_pointer_type(PointerArgType ptr) { return ptr; }
-#endif
 
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const
     {
       return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const
     {
-      return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
-           : IsVectorAtCompileTime ? this->size()
-           : int(Flags)&RowMajorBit ? this->cols()
-           : this->rows();
+      return int(StrideType::OuterStrideAtCompileTime) != 0 ? m_stride.outer()
+           : int(internal::traits<Map>::OuterStrideAtCompileTime) != Dynamic ? Index(internal::traits<Map>::OuterStrideAtCompileTime)
+           : IsVectorAtCompileTime ? (this->size() * innerStride())
+           : (int(Flags)&RowMajorBit) ? (this->cols() * innerStride())
+           : (this->rows() * innerStride());
     }
 
     /** Constructor in the fixed-size case.
       *
       * \param dataPtr pointer to the array to map
-      * \param a_stride optional Stride object, passing the strides.
+      * \param stride optional Stride object, passing the strides.
       */
-    inline Map(PointerArgType dataPtr, const StrideType& a_stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr)), m_stride(a_stride)
+    EIGEN_DEVICE_FUNC
+    explicit inline Map(PointerArgType dataPtr, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr)), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -145,11 +135,12 @@
     /** Constructor in the dynamic-size vector case.
       *
       * \param dataPtr pointer to the array to map
-      * \param a_size the size of the vector expression
-      * \param a_stride optional Stride object, passing the strides.
+      * \param size the size of the vector expression
+      * \param stride optional Stride object, passing the strides.
       */
-    inline Map(PointerArgType dataPtr, Index a_size, const StrideType& a_stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr), a_size), m_stride(a_stride)
+    EIGEN_DEVICE_FUNC
+    inline Map(PointerArgType dataPtr, Index size, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), size), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -157,12 +148,13 @@
     /** Constructor in the dynamic-size matrix case.
       *
       * \param dataPtr pointer to the array to map
-      * \param nbRows the number of rows of the matrix expression
-      * \param nbCols the number of columns of the matrix expression
-      * \param a_stride optional Stride object, passing the strides.
+      * \param rows the number of rows of the matrix expression
+      * \param cols the number of columns of the matrix expression
+      * \param stride optional Stride object, passing the strides.
       */
-    inline Map(PointerArgType dataPtr, Index nbRows, Index nbCols, const StrideType& a_stride = StrideType())
-      : Base(cast_to_pointer_type(dataPtr), nbRows, nbCols), m_stride(a_stride)
+    EIGEN_DEVICE_FUNC
+    inline Map(PointerArgType dataPtr, Index rows, Index cols, const StrideType& stride = StrideType())
+      : Base(cast_to_pointer_type(dataPtr), rows, cols), m_stride(stride)
     {
       PlainObjectType::Base::_check_template_params();
     }
@@ -173,19 +165,6 @@
     StrideType m_stride;
 };
 
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-inline Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>
-  ::Array(const Scalar *data)
-{
-  this->_set_noalias(Eigen::Map<const Array>(data));
-}
-
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-inline Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>
-  ::Matrix(const Scalar *data)
-{
-  this->_set_noalias(Eigen::Map<const Matrix>(data));
-}
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/Core/MapBase.h b/Eigen/src/Core/MapBase.h
index 92e1141..668922f 100644
--- a/Eigen/src/Core/MapBase.h
+++ b/Eigen/src/Core/MapBase.h
@@ -12,15 +12,25 @@
 #define EIGEN_MAPBASE_H
 
 #define EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) \
-      EIGEN_STATIC_ASSERT((int(internal::traits<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
+      EIGEN_STATIC_ASSERT((int(internal::evaluator<Derived>::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \
                           YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT)
 
 namespace Eigen { 
 
-/** \class MapBase
-  * \ingroup Core_Module
+/** \ingroup Core_Module
   *
-  * \brief Base class for Map and Block expression with direct access
+  * \brief Base class for dense Map and Block expression with direct access
+  *
+  * This base class provides the const low-level accessors (e.g. coeff, coeffRef) of dense
+  * Map and Block objects with direct access.
+  * Typical users do not have to directly deal with this class.
+  *
+  * This class can be extended by through the macro plugin \c EIGEN_MAPBASE_PLUGIN.
+  * See \link TopicCustomizing_Plugins customizing Eigen \endlink for details.
+  *
+  * The \c Derived class has to provide the following two methods describing the memory layout:
+  *  \code Index innerStride() const; \endcode
+  *  \code Index outerStride() const; \endcode
   *
   * \sa class Map, class Block
   */
@@ -33,11 +43,11 @@
     enum {
       RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
       ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
+      InnerStrideAtCompileTime = internal::traits<Derived>::InnerStrideAtCompileTime,
       SizeAtCompileTime = Base::SizeAtCompileTime
     };
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -76,8 +86,10 @@
 
     typedef typename Base::CoeffReturnType CoeffReturnType;
 
-    inline Index rows() const { return m_rows.value(); }
-    inline Index cols() const { return m_cols.value(); }
+    /** \copydoc DenseBase::rows() */
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_rows.value(); }
+    /** \copydoc DenseBase::cols() */
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_cols.value(); }
 
     /** Returns a pointer to the first coefficient of the matrix or vector.
       *
@@ -85,30 +97,39 @@
       *
       * \sa innerStride(), outerStride()
       */
-    inline const Scalar* data() const { return m_data; }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return m_data; }
 
+    /** \copydoc PlainObjectBase::coeff(Index,Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeff(Index rowId, Index colId) const
     {
       return m_data[colId * colStride() + rowId * rowStride()];
     }
 
+    /** \copydoc PlainObjectBase::coeff(Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeff(Index index) const
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
       return m_data[index * innerStride()];
     }
 
+    /** \copydoc PlainObjectBase::coeffRef(Index,Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return this->m_data[colId * colStride() + rowId * rowStride()];
     }
 
+    /** \copydoc PlainObjectBase::coeffRef(Index) const */
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
       return this->m_data[index * innerStride()];
     }
 
+    /** \internal */
     template<int LoadMode>
     inline PacketScalar packet(Index rowId, Index colId) const
     {
@@ -116,6 +137,7 @@
                (m_data + (colId * colStride() + rowId * rowStride()));
     }
 
+    /** \internal */
     template<int LoadMode>
     inline PacketScalar packet(Index index) const
     {
@@ -123,12 +145,16 @@
       return internal::ploadt<PacketScalar, LoadMode>(m_data + index * innerStride());
     }
 
-    inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
+    /** \internal Constructor for fixed size matrices or vectors */
+    EIGEN_DEVICE_FUNC
+    explicit inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime)
     {
       EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived)
-      checkSanity();
+      checkSanity<Derived>();
     }
 
+    /** \internal Constructor for dynamically sized vectors */
+    EIGEN_DEVICE_FUNC
     inline MapBase(PointerType dataPtr, Index vecSize)
             : m_data(dataPtr),
               m_rows(RowsAtCompileTime == Dynamic ? vecSize : Index(RowsAtCompileTime)),
@@ -137,34 +163,59 @@
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
       eigen_assert(vecSize >= 0);
       eigen_assert(dataPtr == 0 || SizeAtCompileTime == Dynamic || SizeAtCompileTime == vecSize);
-      checkSanity();
+      checkSanity<Derived>();
     }
 
-    inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols)
-            : m_data(dataPtr), m_rows(nbRows), m_cols(nbCols)
+    /** \internal Constructor for dynamically sized matrices */
+    EIGEN_DEVICE_FUNC
+    inline MapBase(PointerType dataPtr, Index rows, Index cols)
+            : m_data(dataPtr), m_rows(rows), m_cols(cols)
     {
       eigen_assert( (dataPtr == 0)
-              || (   nbRows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows)
-                  && nbCols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols)));
-      checkSanity();
+              || (   rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows)
+                  && cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols)));
+      checkSanity<Derived>();
     }
 
+    #ifdef EIGEN_MAPBASE_PLUGIN
+    #include EIGEN_MAPBASE_PLUGIN
+    #endif
+
   protected:
 
-    void checkSanity() const
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    void checkSanity(typename internal::enable_if<(internal::traits<T>::Alignment>0),void*>::type = 0) const
     {
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(internal::traits<Derived>::Flags&PacketAccessBit,
-                                        internal::inner_stride_at_compile_time<Derived>::ret==1),
-                          PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1);
-      eigen_assert(EIGEN_IMPLIES(internal::traits<Derived>::Flags&AlignedBit, (size_t(m_data) % 16) == 0)
-                   && "input pointer is not aligned on a 16 byte boundary");
+#if EIGEN_MAX_ALIGN_BYTES>0
+      // innerStride() is not set yet when this function is called, so we optimistically assume the lowest plausible value:
+      const Index minInnerStride = InnerStrideAtCompileTime == Dynamic ? 1 : Index(InnerStrideAtCompileTime);
+      EIGEN_ONLY_USED_FOR_DEBUG(minInnerStride);
+      eigen_assert((   ((internal::UIntPtr(m_data) % internal::traits<Derived>::Alignment) == 0)
+                    || (cols() * rows() * minInnerStride * sizeof(Scalar)) < internal::traits<Derived>::Alignment ) && "data is not aligned");
+#endif
     }
 
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    void checkSanity(typename internal::enable_if<internal::traits<T>::Alignment==0,void*>::type = 0) const
+    {}
+
     PointerType m_data;
     const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_rows;
     const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_cols;
 };
 
+/** \ingroup Core_Module
+  *
+  * \brief Base class for non-const dense Map and Block expression with direct access
+  *
+  * This base class provides the non-const low-level accessors (e.g. coeff and coeffRef) of
+  * dense Map and Block objects with direct access.
+  * It inherits MapBase<Derived, ReadOnlyAccessors> which defines the const variant for reading specific entries.
+  *
+  * \sa class Map, class Block
+  */
 template<typename Derived> class MapBase<Derived, WriteAccessors>
   : public MapBase<Derived, ReadOnlyAccessors>
 {
@@ -175,7 +226,7 @@
 
     typedef typename Base::Scalar Scalar;
     typedef typename Base::PacketScalar PacketScalar;
-    typedef typename Base::Index Index;
+    typedef typename Base::StorageIndex StorageIndex;
     typedef typename Base::PointerType PointerType;
 
     using Base::derived;
@@ -196,14 +247,18 @@
                     const Scalar
                   >::type ScalarWithConstIfNotLvalue;
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar* data() const { return this->m_data; }
+    EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue* data() { return this->m_data; } // no const-cast here so non-const-correct code will give a compile error
 
+    EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue& coeffRef(Index row, Index col)
     {
       return this->m_data[col * colStride() + row * rowStride()];
     }
 
+    EIGEN_DEVICE_FUNC
     inline ScalarWithConstIfNotLvalue& coeffRef(Index index)
     {
       EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived)
@@ -225,10 +280,11 @@
                 (this->m_data + index * innerStride(), val);
     }
 
-    explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {}
-    inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {}
-    inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols) : Base(dataPtr, nbRows, nbCols) {}
+    EIGEN_DEVICE_FUNC explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {}
+    EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {}
+    EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index rows, Index cols) : Base(dataPtr, rows, cols) {}
 
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const MapBase& other)
     {
       ReadOnlyMapBase::Base::operator=(other);
diff --git a/Eigen/src/Core/MathFunctions.h b/Eigen/src/Core/MathFunctions.h
index 2bfc5eb..b249ce0 100644
--- a/Eigen/src/Core/MathFunctions.h
+++ b/Eigen/src/Core/MathFunctions.h
@@ -10,11 +10,25 @@
 #ifndef EIGEN_MATHFUNCTIONS_H
 #define EIGEN_MATHFUNCTIONS_H
 
+// source: http://www.geom.uiuc.edu/~huberty/math5337/groupe/digits.html
+// TODO this should better be moved to NumTraits
+#define EIGEN_PI 3.141592653589793238462643383279502884197169399375105820974944592307816406L
+
+
 namespace Eigen {
 
+// On WINCE, std::abs is defined for int only, so let's defined our own overloads:
+// This issue has been confirmed with MSVC 2008 only, but the issue might exist for more recent versions too.
+#if EIGEN_OS_WINCE && EIGEN_COMP_MSVC && EIGEN_COMP_MSVC<=1500
+long        abs(long        x) { return (labs(x));  }
+double      abs(double      x) { return (fabs(x));  }
+float       abs(float       x) { return (fabsf(x)); }
+long double abs(long double x) { return (fabsl(x)); }
+#endif
+
 namespace internal {
 
-/** \internal \struct global_math_functions_filtering_base
+/** \internal \class global_math_functions_filtering_base
   *
   * What it does:
   * Defines a typedef 'type' as follows:
@@ -62,6 +76,7 @@
 struct real_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     return x;
@@ -72,6 +87,7 @@
 struct real_default_impl<Scalar,true>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     using std::real;
@@ -81,13 +97,25 @@
 
 template<typename Scalar> struct real_impl : real_default_impl<Scalar> {};
 
+#ifdef __CUDA_ARCH__
+template<typename T>
+struct real_impl<std::complex<T> >
+{
+  typedef T RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline T run(const std::complex<T>& x)
+  {
+    return x.real();
+  }
+};
+#endif
+
 template<typename Scalar>
 struct real_retval
 {
   typedef typename NumTraits<Scalar>::Real type;
 };
 
-
 /****************************************************************************
 * Implementation of imag                                                 *
 ****************************************************************************/
@@ -96,6 +124,7 @@
 struct imag_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar&)
   {
     return RealScalar(0);
@@ -106,6 +135,7 @@
 struct imag_default_impl<Scalar,true>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     using std::imag;
@@ -115,6 +145,19 @@
 
 template<typename Scalar> struct imag_impl : imag_default_impl<Scalar> {};
 
+#ifdef __CUDA_ARCH__
+template<typename T>
+struct imag_impl<std::complex<T> >
+{
+  typedef T RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline T run(const std::complex<T>& x)
+  {
+    return x.imag();
+  }
+};
+#endif
+
 template<typename Scalar>
 struct imag_retval
 {
@@ -129,10 +172,12 @@
 struct real_ref_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar& run(Scalar& x)
   {
     return reinterpret_cast<RealScalar*>(&x)[0];
   }
+  EIGEN_DEVICE_FUNC
   static inline const RealScalar& run(const Scalar& x)
   {
     return reinterpret_cast<const RealScalar*>(&x)[0];
@@ -153,10 +198,12 @@
 struct imag_ref_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar& run(Scalar& x)
   {
     return reinterpret_cast<RealScalar*>(&x)[1];
   }
+  EIGEN_DEVICE_FUNC
   static inline const RealScalar& run(const Scalar& x)
   {
     return reinterpret_cast<RealScalar*>(&x)[1];
@@ -166,10 +213,12 @@
 template<typename Scalar>
 struct imag_ref_default_impl<Scalar, false>
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(Scalar&)
   {
     return Scalar(0);
   }
+  EIGEN_DEVICE_FUNC
   static inline const Scalar run(const Scalar&)
   {
     return Scalar(0);
@@ -192,6 +241,7 @@
 template<typename Scalar, bool IsComplex = NumTraits<Scalar>::IsComplex>
 struct conj_impl
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(const Scalar& x)
   {
     return x;
@@ -201,6 +251,7 @@
 template<typename Scalar>
 struct conj_impl<Scalar,true>
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(const Scalar& x)
   {
     using std::conj;
@@ -218,26 +269,40 @@
 * Implementation of abs2                                                 *
 ****************************************************************************/
 
-template<typename Scalar>
-struct abs2_impl
+template<typename Scalar,bool IsComplex>
+struct abs2_impl_default
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
     return x*x;
   }
 };
 
-template<typename RealScalar>
-struct abs2_impl<std::complex<RealScalar> >
+template<typename Scalar>
+struct abs2_impl_default<Scalar, true> // IsComplex
 {
-  static inline RealScalar run(const std::complex<RealScalar>& x)
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const Scalar& x)
   {
     return real(x)*real(x) + imag(x)*imag(x);
   }
 };
 
 template<typename Scalar>
+struct abs2_impl
+{
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
+  static inline RealScalar run(const Scalar& x)
+  {
+    return abs2_impl_default<Scalar,NumTraits<Scalar>::IsComplex>::run(x);
+  }
+};
+
+template<typename Scalar>
 struct abs2_retval
 {
   typedef typename NumTraits<Scalar>::Real type;
@@ -251,9 +316,10 @@
 struct norm1_default_impl
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline RealScalar run(const Scalar& x)
   {
-    using std::abs;
+    EIGEN_USING_STD_MATH(abs);
     return abs(real(x)) + abs(imag(x));
   }
 };
@@ -261,9 +327,10 @@
 template<typename Scalar>
 struct norm1_default_impl<Scalar, false>
 {
+  EIGEN_DEVICE_FUNC
   static inline Scalar run(const Scalar& x)
   {
-    using std::abs;
+    EIGEN_USING_STD_MATH(abs);
     return abs(x);
   }
 };
@@ -281,25 +348,7 @@
 * Implementation of hypot                                                *
 ****************************************************************************/
 
-template<typename Scalar>
-struct hypot_impl
-{
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  static inline RealScalar run(const Scalar& x, const Scalar& y)
-  {
-    using std::max;
-    using std::min;
-    using std::abs;
-    using std::sqrt;
-    RealScalar _x = abs(x);
-    RealScalar _y = abs(y);
-    RealScalar p = (max)(_x, _y);
-    if(p==RealScalar(0)) return 0;
-    RealScalar q = (min)(_x, _y);
-    RealScalar qp = q/p;
-    return p * sqrt(RealScalar(1) + qp*qp);
-  }
-};
+template<typename Scalar> struct hypot_impl;
 
 template<typename Scalar>
 struct hypot_retval
@@ -314,6 +363,7 @@
 template<typename OldType, typename NewType>
 struct cast_impl
 {
+  EIGEN_DEVICE_FUNC
   static inline NewType run(const OldType& x)
   {
     return static_cast<NewType>(x);
@@ -323,48 +373,124 @@
 // here, for once, we're plainly returning NewType: we don't want cast to do weird things.
 
 template<typename OldType, typename NewType>
+EIGEN_DEVICE_FUNC
 inline NewType cast(const OldType& x)
 {
   return cast_impl<OldType, NewType>::run(x);
 }
 
 /****************************************************************************
-* Implementation of atanh2                                                *
+* Implementation of round                                                   *
 ****************************************************************************/
 
-template<typename Scalar, bool IsInteger>
-struct atanh2_default_impl
-{
-  typedef Scalar retval;
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  static inline Scalar run(const Scalar& x, const Scalar& y)
+#if EIGEN_HAS_CXX11_MATH
+  template<typename Scalar>
+  struct round_impl {
+    static inline Scalar run(const Scalar& x)
+    {
+      EIGEN_STATIC_ASSERT((!NumTraits<Scalar>::IsComplex), NUMERIC_TYPE_MUST_BE_REAL)
+      using std::round;
+      return round(x);
+    }
+  };
+#else
+  template<typename Scalar>
+  struct round_impl
   {
-    using std::abs;
-    using std::log;
-    using std::sqrt;
-    Scalar z = x / y;
-    if (y == Scalar(0) || abs(z) > sqrt(NumTraits<RealScalar>::epsilon()))
-      return RealScalar(0.5) * log((y + x) / (y - x));
-    else
-      return z + z*z*z / RealScalar(3);
-  }
-};
+    static inline Scalar run(const Scalar& x)
+    {
+      EIGEN_STATIC_ASSERT((!NumTraits<Scalar>::IsComplex), NUMERIC_TYPE_MUST_BE_REAL)
+      EIGEN_USING_STD_MATH(floor);
+      EIGEN_USING_STD_MATH(ceil);
+      return (x > Scalar(0)) ? floor(x + Scalar(0.5)) : ceil(x - Scalar(0.5));
+    }
+  };
+#endif
 
 template<typename Scalar>
-struct atanh2_default_impl<Scalar, true>
+struct round_retval
 {
-  static inline Scalar run(const Scalar&, const Scalar&)
+  typedef Scalar type;
+};
+
+/****************************************************************************
+* Implementation of arg                                                     *
+****************************************************************************/
+
+#if EIGEN_HAS_CXX11_MATH
+  template<typename Scalar>
+  struct arg_impl {
+    static inline Scalar run(const Scalar& x)
+    {
+      EIGEN_USING_STD_MATH(arg);
+      return arg(x);
+    }
+  };
+#else
+  template<typename Scalar, bool IsComplex = NumTraits<Scalar>::IsComplex>
+  struct arg_default_impl
+  {
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    EIGEN_DEVICE_FUNC
+    static inline RealScalar run(const Scalar& x)
+    {
+      return (x < Scalar(0)) ? Scalar(EIGEN_PI) : Scalar(0); }
+  };
+
+  template<typename Scalar>
+  struct arg_default_impl<Scalar,true>
+  {
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    EIGEN_DEVICE_FUNC
+    static inline RealScalar run(const Scalar& x)
+    {
+      EIGEN_USING_STD_MATH(arg);
+      return arg(x);
+    }
+  };
+
+  template<typename Scalar> struct arg_impl : arg_default_impl<Scalar> {};
+#endif
+
+template<typename Scalar>
+struct arg_retval
+{
+  typedef typename NumTraits<Scalar>::Real type;
+};
+
+/****************************************************************************
+* Implementation of log1p                                                   *
+****************************************************************************/
+
+namespace std_fallback {
+  // fallback log1p implementation in case there is no log1p(Scalar) function in namespace of Scalar,
+  // or that there is no suitable std::log1p function available
+  template<typename Scalar>
+  EIGEN_DEVICE_FUNC inline Scalar log1p(const Scalar& x) {
+    EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar)
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    EIGEN_USING_STD_MATH(log);
+    Scalar x1p = RealScalar(1) + x;
+    return numext::equal_strict(x1p, Scalar(1)) ? x : x * ( log(x1p) / (x1p - RealScalar(1)) );
+  }
+}
+
+template<typename Scalar>
+struct log1p_impl {
+  static inline Scalar run(const Scalar& x)
   {
     EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar)
-    return Scalar(0);
+    #if EIGEN_HAS_CXX11_MATH
+    using std::log1p;
+    #endif
+    using std_fallback::log1p;
+    return log1p(x);
   }
 };
 
-template<typename Scalar>
-struct atanh2_impl : atanh2_default_impl<Scalar, NumTraits<Scalar>::IsInteger> {};
 
 template<typename Scalar>
-struct atanh2_retval
+struct log1p_retval
 {
   typedef Scalar type;
 };
@@ -373,24 +499,26 @@
 * Implementation of pow                                                  *
 ****************************************************************************/
 
-template<typename Scalar, bool IsInteger>
-struct pow_default_impl
+template<typename ScalarX,typename ScalarY, bool IsInteger = NumTraits<ScalarX>::IsInteger&&NumTraits<ScalarY>::IsInteger>
+struct pow_impl
 {
-  typedef Scalar retval;
-  static inline Scalar run(const Scalar& x, const Scalar& y)
+  //typedef Scalar retval;
+  typedef typename ScalarBinaryOpTraits<ScalarX,ScalarY,internal::scalar_pow_op<ScalarX,ScalarY> >::ReturnType result_type;
+  static EIGEN_DEVICE_FUNC inline result_type run(const ScalarX& x, const ScalarY& y)
   {
-    using std::pow;
+    EIGEN_USING_STD_MATH(pow);
     return pow(x, y);
   }
 };
 
-template<typename Scalar>
-struct pow_default_impl<Scalar, true>
+template<typename ScalarX,typename ScalarY>
+struct pow_impl<ScalarX,ScalarY, true>
 {
-  static inline Scalar run(Scalar x, Scalar y)
+  typedef ScalarX result_type;
+  static EIGEN_DEVICE_FUNC inline ScalarX run(ScalarX x, ScalarY y)
   {
-    Scalar res(1);
-    eigen_assert(!NumTraits<Scalar>::IsSigned || y >= 0);
+    ScalarX res(1);
+    eigen_assert(!NumTraits<ScalarY>::IsSigned || y >= 0);
     if(y & 1) res *= x;
     y >>= 1;
     while(y)
@@ -403,15 +531,6 @@
   }
 };
 
-template<typename Scalar>
-struct pow_impl : pow_default_impl<Scalar, NumTraits<Scalar>::IsInteger> {};
-
-template<typename Scalar>
-struct pow_retval
-{
-  typedef Scalar type;
-};
-
 /****************************************************************************
 * Implementation of random                                               *
 ****************************************************************************/
@@ -447,48 +566,48 @@
 };
 
 enum {
-  floor_log2_terminate,
-  floor_log2_move_up,
-  floor_log2_move_down,
-  floor_log2_bogus
+  meta_floor_log2_terminate,
+  meta_floor_log2_move_up,
+  meta_floor_log2_move_down,
+  meta_floor_log2_bogus
 };
 
-template<unsigned int n, int lower, int upper> struct floor_log2_selector
+template<unsigned int n, int lower, int upper> struct meta_floor_log2_selector
 {
   enum { middle = (lower + upper) / 2,
-         value = (upper <= lower + 1) ? int(floor_log2_terminate)
-               : (n < (1 << middle)) ? int(floor_log2_move_down)
-               : (n==0) ? int(floor_log2_bogus)
-               : int(floor_log2_move_up)
+         value = (upper <= lower + 1) ? int(meta_floor_log2_terminate)
+               : (n < (1 << middle)) ? int(meta_floor_log2_move_down)
+               : (n==0) ? int(meta_floor_log2_bogus)
+               : int(meta_floor_log2_move_up)
   };
 };
 
 template<unsigned int n,
          int lower = 0,
          int upper = sizeof(unsigned int) * CHAR_BIT - 1,
-         int selector = floor_log2_selector<n, lower, upper>::value>
-struct floor_log2 {};
+         int selector = meta_floor_log2_selector<n, lower, upper>::value>
+struct meta_floor_log2 {};
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_move_down>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_move_down>
 {
-  enum { value = floor_log2<n, lower, floor_log2_selector<n, lower, upper>::middle>::value };
+  enum { value = meta_floor_log2<n, lower, meta_floor_log2_selector<n, lower, upper>::middle>::value };
 };
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_move_up>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_move_up>
 {
-  enum { value = floor_log2<n, floor_log2_selector<n, lower, upper>::middle, upper>::value };
+  enum { value = meta_floor_log2<n, meta_floor_log2_selector<n, lower, upper>::middle, upper>::value };
 };
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_terminate>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_terminate>
 {
   enum { value = (n >= ((unsigned int)(1) << (lower+1))) ? lower+1 : lower };
 };
 
 template<unsigned int n, int lower, int upper>
-struct floor_log2<n, lower, upper, floor_log2_bogus>
+struct meta_floor_log2<n, lower, upper, meta_floor_log2_bogus>
 {
   // no value, error at compile time
 };
@@ -496,11 +615,31 @@
 template<typename Scalar>
 struct random_default_impl<Scalar, false, true>
 {
-  typedef typename NumTraits<Scalar>::NonInteger NonInteger;
-
   static inline Scalar run(const Scalar& x, const Scalar& y)
   {
-    return x + Scalar((NonInteger(y)-x+1) * std::rand() / (RAND_MAX + NonInteger(1)));
+    if (y <= x)
+      return x;
+    // ScalarU is the unsigned counterpart of Scalar, possibly Scalar itself.
+    typedef typename make_unsigned<Scalar>::type ScalarU;
+    // ScalarX is the widest of ScalarU and unsigned int.
+    // We'll deal only with ScalarX and unsigned int below thus avoiding signed
+    // types and arithmetic and signed overflows (which are undefined behavior).
+    typedef typename conditional<(ScalarU(-1) > unsigned(-1)), ScalarU, unsigned>::type ScalarX;
+    // The following difference doesn't overflow, provided our integer types are two's
+    // complement and have the same number of padding bits in signed and unsigned variants.
+    // This is the case in most modern implementations of C++.
+    ScalarX range = ScalarX(y) - ScalarX(x);
+    ScalarX offset = 0;
+    ScalarX divisor = 1;
+    ScalarX multiplier = 1;
+    const unsigned rand_max = RAND_MAX;
+    if (range <= rand_max) divisor = (rand_max + 1) / (range + 1);
+    else                   multiplier = 1 + range / (rand_max + 1);
+    // Rejection sampling.
+    do {
+      offset = (unsigned(std::rand()) * multiplier) / divisor;
+    } while (offset > range);
+    return Scalar(ScalarX(x) + offset);
   }
 
   static inline Scalar run()
@@ -508,7 +647,7 @@
 #ifdef EIGEN_MAKING_DOCS
     return run(Scalar(NumTraits<Scalar>::IsSigned ? -10 : 0), Scalar(10));
 #else
-    enum { rand_bits = floor_log2<(unsigned int)(RAND_MAX)+1>::value,
+    enum { rand_bits = meta_floor_log2<(unsigned int)(RAND_MAX)+1>::value,
            scalar_bits = sizeof(Scalar) * CHAR_BIT,
            shift = EIGEN_PLAIN_ENUM_MAX(0, int(rand_bits) - int(scalar_bits)),
            offset = NumTraits<Scalar>::IsSigned ? (1 << (EIGEN_PLAIN_ENUM_MIN(rand_bits,scalar_bits)-1)) : 0
@@ -545,97 +684,602 @@
   return EIGEN_MATHFUNC_IMPL(random, Scalar)::run();
 }
 
+// Implementatin of is* functions
+
+// std::is* do not work with fast-math and gcc, std::is* are available on MSVC 2013 and newer, as well as in clang.
+#if (EIGEN_HAS_CXX11_MATH && !(EIGEN_COMP_GNUC_STRICT && __FINITE_MATH_ONLY__)) || (EIGEN_COMP_MSVC>=1800) || (EIGEN_COMP_CLANG)
+#define EIGEN_USE_STD_FPCLASSIFY 1
+#else
+#define EIGEN_USE_STD_FPCLASSIFY 0
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<internal::is_integral<T>::value,bool>::type
+isnan_impl(const T&) { return false; }
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<internal::is_integral<T>::value,bool>::type
+isinf_impl(const T&) { return false; }
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<internal::is_integral<T>::value,bool>::type
+isfinite_impl(const T&) { return true; }
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type
+isfinite_impl(const T& x)
+{
+  #ifdef __CUDA_ARCH__
+    return (::isfinite)(x);
+  #elif EIGEN_USE_STD_FPCLASSIFY
+    using std::isfinite;
+    return isfinite EIGEN_NOT_A_MACRO (x);
+  #else
+    return x<=NumTraits<T>::highest() && x>=NumTraits<T>::lowest();
+  #endif
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type
+isinf_impl(const T& x)
+{
+  #ifdef __CUDA_ARCH__
+    return (::isinf)(x);
+  #elif EIGEN_USE_STD_FPCLASSIFY
+    using std::isinf;
+    return isinf EIGEN_NOT_A_MACRO (x);
+  #else
+    return x>NumTraits<T>::highest() || x<NumTraits<T>::lowest();
+  #endif
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+typename internal::enable_if<(!internal::is_integral<T>::value)&&(!NumTraits<T>::IsComplex),bool>::type
+isnan_impl(const T& x)
+{
+  #ifdef __CUDA_ARCH__
+    return (::isnan)(x);
+  #elif EIGEN_USE_STD_FPCLASSIFY
+    using std::isnan;
+    return isnan EIGEN_NOT_A_MACRO (x);
+  #else
+    return x != x;
+  #endif
+}
+
+#if (!EIGEN_USE_STD_FPCLASSIFY)
+
+#if EIGEN_COMP_MSVC
+
+template<typename T> EIGEN_DEVICE_FUNC bool isinf_msvc_helper(T x)
+{
+  return _fpclass(x)==_FPCLASS_NINF || _fpclass(x)==_FPCLASS_PINF;
+}
+
+//MSVC defines a _isnan builtin function, but for double only
+EIGEN_DEVICE_FUNC inline bool isnan_impl(const long double& x) { return _isnan(x)!=0; }
+EIGEN_DEVICE_FUNC inline bool isnan_impl(const double& x)      { return _isnan(x)!=0; }
+EIGEN_DEVICE_FUNC inline bool isnan_impl(const float& x)       { return _isnan(x)!=0; }
+
+EIGEN_DEVICE_FUNC inline bool isinf_impl(const long double& x) { return isinf_msvc_helper(x); }
+EIGEN_DEVICE_FUNC inline bool isinf_impl(const double& x)      { return isinf_msvc_helper(x); }
+EIGEN_DEVICE_FUNC inline bool isinf_impl(const float& x)       { return isinf_msvc_helper(x); }
+
+#elif (defined __FINITE_MATH_ONLY__ && __FINITE_MATH_ONLY__ && EIGEN_COMP_GNUC)
+
+#if EIGEN_GNUC_AT_LEAST(5,0)
+  #define EIGEN_TMP_NOOPT_ATTRIB EIGEN_DEVICE_FUNC inline __attribute__((optimize("no-finite-math-only")))
+#else
+  // NOTE the inline qualifier and noinline attribute are both needed: the former is to avoid linking issue (duplicate symbol),
+  //      while the second prevent too aggressive optimizations in fast-math mode:
+  #define EIGEN_TMP_NOOPT_ATTRIB EIGEN_DEVICE_FUNC inline __attribute__((noinline,optimize("no-finite-math-only")))
+#endif
+
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const long double& x) { return __builtin_isnan(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const double& x)      { return __builtin_isnan(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const float& x)       { return __builtin_isnan(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const double& x)      { return __builtin_isinf(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const float& x)       { return __builtin_isinf(x); }
+template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const long double& x) { return __builtin_isinf(x); }
+
+#undef EIGEN_TMP_NOOPT_ATTRIB
+
+#endif
+
+#endif
+
+// The following overload are defined at the end of this file
+template<typename T> EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex<T>& x);
+template<typename T> EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex<T>& x);
+template<typename T> EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex<T>& x);
+
+template<typename T> T generic_fast_tanh_float(const T& a_x);
+
 } // end namespace internal
 
 /****************************************************************************
-* Generic math function                                                    *
+* Generic math functions                                                    *
 ****************************************************************************/
 
 namespace numext {
 
+#ifndef __CUDA_ARCH__
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T mini(const T& x, const T& y)
+{
+  EIGEN_USING_STD_MATH(min);
+  return min EIGEN_NOT_A_MACRO (x,y);
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T maxi(const T& x, const T& y)
+{
+  EIGEN_USING_STD_MATH(max);
+  return max EIGEN_NOT_A_MACRO (x,y);
+}
+#else
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T mini(const T& x, const T& y)
+{
+  return y < x ? y : x;
+}
+template<>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE float mini(const float& x, const float& y)
+{
+  return fminf(x, y);
+}
+template<typename T>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE T maxi(const T& x, const T& y)
+{
+  return x < y ? y : x;
+}
+template<>
+EIGEN_DEVICE_FUNC
+EIGEN_ALWAYS_INLINE float maxi(const float& x, const float& y)
+{
+  return fmaxf(x, y);
+}
+#endif
+
+
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(real, Scalar) real(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(real, Scalar)::run(x);
-}  
+}
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline typename internal::add_const_on_value_type< EIGEN_MATHFUNC_RETVAL(real_ref, Scalar) >::type real_ref(const Scalar& x)
 {
   return internal::real_ref_impl<Scalar>::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(real_ref, Scalar) real_ref(Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(real_ref, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(imag, Scalar) imag(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(imag, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
+inline EIGEN_MATHFUNC_RETVAL(arg, Scalar) arg(const Scalar& x)
+{
+  return EIGEN_MATHFUNC_IMPL(arg, Scalar)::run(x);
+}
+
+template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline typename internal::add_const_on_value_type< EIGEN_MATHFUNC_RETVAL(imag_ref, Scalar) >::type imag_ref(const Scalar& x)
 {
   return internal::imag_ref_impl<Scalar>::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(imag_ref, Scalar) imag_ref(Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(imag_ref, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(conj, Scalar) conj(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(conj, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(abs2, Scalar) abs2(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(abs2, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(norm1, Scalar) norm1(const Scalar& x)
 {
   return EIGEN_MATHFUNC_IMPL(norm1, Scalar)::run(x);
 }
 
 template<typename Scalar>
+EIGEN_DEVICE_FUNC
 inline EIGEN_MATHFUNC_RETVAL(hypot, Scalar) hypot(const Scalar& x, const Scalar& y)
 {
   return EIGEN_MATHFUNC_IMPL(hypot, Scalar)::run(x, y);
 }
 
 template<typename Scalar>
-inline EIGEN_MATHFUNC_RETVAL(atanh2, Scalar) atanh2(const Scalar& x, const Scalar& y)
+EIGEN_DEVICE_FUNC
+inline EIGEN_MATHFUNC_RETVAL(log1p, Scalar) log1p(const Scalar& x)
 {
-  return EIGEN_MATHFUNC_IMPL(atanh2, Scalar)::run(x, y);
+  return EIGEN_MATHFUNC_IMPL(log1p, Scalar)::run(x);
 }
 
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float log1p(const float &x) { return ::log1pf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double log1p(const double &x) { return ::log1p(x); }
+#endif
+
+template<typename ScalarX,typename ScalarY>
+EIGEN_DEVICE_FUNC
+inline typename internal::pow_impl<ScalarX,ScalarY>::result_type pow(const ScalarX& x, const ScalarY& y)
+{
+  return internal::pow_impl<ScalarX,ScalarY>::run(x, y);
+}
+
+template<typename T> EIGEN_DEVICE_FUNC bool (isnan)   (const T &x) { return internal::isnan_impl(x); }
+template<typename T> EIGEN_DEVICE_FUNC bool (isinf)   (const T &x) { return internal::isinf_impl(x); }
+template<typename T> EIGEN_DEVICE_FUNC bool (isfinite)(const T &x) { return internal::isfinite_impl(x); }
+
 template<typename Scalar>
-inline EIGEN_MATHFUNC_RETVAL(pow, Scalar) pow(const Scalar& x, const Scalar& y)
+EIGEN_DEVICE_FUNC
+inline EIGEN_MATHFUNC_RETVAL(round, Scalar) round(const Scalar& x)
 {
-  return EIGEN_MATHFUNC_IMPL(pow, Scalar)::run(x, y);
+  return EIGEN_MATHFUNC_IMPL(round, Scalar)::run(x);
 }
 
-// std::isfinite is non standard, so let's define our own version,
-// even though it is not very efficient.
-template<typename T> bool (isfinite)(const T& x)
+template<typename T>
+EIGEN_DEVICE_FUNC
+T (floor)(const T& x)
 {
-  return x<NumTraits<T>::highest() && x>NumTraits<T>::lowest();
+  EIGEN_USING_STD_MATH(floor);
+  return floor(x);
 }
 
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float floor(const float &x) { return ::floorf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double floor(const double &x) { return ::floor(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC
+T (ceil)(const T& x)
+{
+  EIGEN_USING_STD_MATH(ceil);
+  return ceil(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float ceil(const float &x) { return ::ceilf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double ceil(const double &x) { return ::ceil(x); }
+#endif
+
+
+/** Log base 2 for 32 bits positive integers.
+  * Conveniently returns 0 for x==0. */
+inline int log2(int x)
+{
+  eigen_assert(x>=0);
+  unsigned int v(x);
+  static const int table[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+  v |= v >> 1;
+  v |= v >> 2;
+  v |= v >> 4;
+  v |= v >> 8;
+  v |= v >> 16;
+  return table[(v * 0x07C4ACDDU) >> 27];
+}
+
+/** \returns the square root of \a x.
+  *
+  * It is essentially equivalent to
+  * \code using std::sqrt; return sqrt(x); \endcode
+  * but slightly faster for float/double and some compilers (e.g., gcc), thanks to
+  * specializations when SSE is enabled.
+  *
+  * It's usage is justified in performance critical functions, like norm/normalize.
+  */
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T sqrt(const T &x)
+{
+  EIGEN_USING_STD_MATH(sqrt);
+  return sqrt(x);
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T log(const T &x) {
+  EIGEN_USING_STD_MATH(log);
+  return log(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float log(const float &x) { return ::logf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double log(const double &x) { return ::log(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+typename internal::enable_if<NumTraits<T>::IsSigned || NumTraits<T>::IsComplex,typename NumTraits<T>::Real>::type
+abs(const T &x) {
+  EIGEN_USING_STD_MATH(abs);
+  return abs(x);
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+typename internal::enable_if<!(NumTraits<T>::IsSigned || NumTraits<T>::IsComplex),typename NumTraits<T>::Real>::type
+abs(const T &x) {
+  return x;
+}
+
+#if defined(__SYCL_DEVICE_ONLY__)
+EIGEN_ALWAYS_INLINE float   abs(float x) { return cl::sycl::fabs(x); }
+EIGEN_ALWAYS_INLINE double  abs(double x) { return cl::sycl::fabs(x); }
+#endif // defined(__SYCL_DEVICE_ONLY__)
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float abs(const float &x) { return ::fabsf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double abs(const double &x) { return ::fabs(x); }
+
+template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float abs(const std::complex<float>& x) {
+  return ::hypotf(x.real(), x.imag());
+}
+
+template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double abs(const std::complex<double>& x) {
+  return ::hypot(x.real(), x.imag());
+}
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T exp(const T &x) {
+  EIGEN_USING_STD_MATH(exp);
+  return exp(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float exp(const float &x) { return ::expf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double exp(const double &x) { return ::exp(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T cos(const T &x) {
+  EIGEN_USING_STD_MATH(cos);
+  return cos(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float cos(const float &x) { return ::cosf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double cos(const double &x) { return ::cos(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T sin(const T &x) {
+  EIGEN_USING_STD_MATH(sin);
+  return sin(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float sin(const float &x) { return ::sinf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double sin(const double &x) { return ::sin(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T tan(const T &x) {
+  EIGEN_USING_STD_MATH(tan);
+  return tan(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float tan(const float &x) { return ::tanf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double tan(const double &x) { return ::tan(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T acos(const T &x) {
+  EIGEN_USING_STD_MATH(acos);
+  return acos(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float acos(const float &x) { return ::acosf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double acos(const double &x) { return ::acos(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T asin(const T &x) {
+  EIGEN_USING_STD_MATH(asin);
+  return asin(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float asin(const float &x) { return ::asinf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double asin(const double &x) { return ::asin(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T atan(const T &x) {
+  EIGEN_USING_STD_MATH(atan);
+  return atan(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float atan(const float &x) { return ::atanf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double atan(const double &x) { return ::atan(x); }
+#endif
+
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T cosh(const T &x) {
+  EIGEN_USING_STD_MATH(cosh);
+  return cosh(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float cosh(const float &x) { return ::coshf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double cosh(const double &x) { return ::cosh(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T sinh(const T &x) {
+  EIGEN_USING_STD_MATH(sinh);
+  return sinh(x);
+}
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float sinh(const float &x) { return ::sinhf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double sinh(const double &x) { return ::sinh(x); }
+#endif
+
+template<typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T tanh(const T &x) {
+  EIGEN_USING_STD_MATH(tanh);
+  return tanh(x);
+}
+
+#if (!defined(__CUDACC__)) && EIGEN_FAST_MATH
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float tanh(float x) { return internal::generic_fast_tanh_float(x); }
+#endif
+
+#ifdef __CUDACC__
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float tanh(const float &x) { return ::tanhf(x); }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double tanh(const double &x) { return ::tanh(x); }
+#endif
+
+template <typename T>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+T fmod(const T& a, const T& b) {
+  EIGEN_USING_STD_MATH(fmod);
+  return fmod(a, b);
+}
+
+#ifdef __CUDACC__
+template <>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float fmod(const float& a, const float& b) {
+  return ::fmodf(a, b);
+}
+
+template <>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double fmod(const double& a, const double& b) {
+  return ::fmod(a, b);
+}
+#endif
+
 } // end namespace numext
 
 namespace internal {
 
+template<typename T>
+EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex<T>& x)
+{
+  return (numext::isfinite)(numext::real(x)) && (numext::isfinite)(numext::imag(x));
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex<T>& x)
+{
+  return (numext::isnan)(numext::real(x)) || (numext::isnan)(numext::imag(x));
+}
+
+template<typename T>
+EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex<T>& x)
+{
+  return ((numext::isinf)(numext::real(x)) || (numext::isinf)(numext::imag(x))) && (!(numext::isnan)(x));
+}
+
 /****************************************************************************
 * Implementation of fuzzy comparisons                                       *
 ****************************************************************************/
@@ -649,18 +1293,17 @@
 struct scalar_fuzzy_default_impl<Scalar, false, false>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const RealScalar& prec)
   {
-    using std::abs;
-    return abs(x) <= abs(y) * prec;
+    return numext::abs(x) <= numext::abs(y) * prec;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar& prec)
   {
-    using std::min;
-    using std::abs;
-    return abs(x - y) <= (min)(abs(x), abs(y)) * prec;
+    return numext::abs(x - y) <= numext::mini(numext::abs(x), numext::abs(y)) * prec;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const RealScalar& prec)
   {
     return x <= y || isApprox(x, y, prec);
@@ -671,15 +1314,17 @@
 struct scalar_fuzzy_default_impl<Scalar, false, true>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const Scalar& x, const Scalar&, const RealScalar&)
   {
     return x == Scalar(0);
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar&)
   {
     return x == y;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const RealScalar&)
   {
     return x <= y;
@@ -690,38 +1335,38 @@
 struct scalar_fuzzy_default_impl<Scalar, true, false>
 {
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const RealScalar& prec)
   {
     return numext::abs2(x) <= numext::abs2(y) * prec * prec;
   }
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar& prec)
   {
-    using std::min;
-    return numext::abs2(x - y) <= (min)(numext::abs2(x), numext::abs2(y)) * prec * prec;
+    return numext::abs2(x - y) <= numext::mini(numext::abs2(x), numext::abs2(y)) * prec * prec;
   }
 };
 
 template<typename Scalar>
 struct scalar_fuzzy_impl : scalar_fuzzy_default_impl<Scalar, NumTraits<Scalar>::IsComplex, NumTraits<Scalar>::IsInteger> {};
 
-template<typename Scalar, typename OtherScalar>
+template<typename Scalar, typename OtherScalar> EIGEN_DEVICE_FUNC
 inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y,
-                                   typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
+                              const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision())
 {
   return scalar_fuzzy_impl<Scalar>::template isMuchSmallerThan<OtherScalar>(x, y, precision);
 }
 
-template<typename Scalar>
+template<typename Scalar> EIGEN_DEVICE_FUNC
 inline bool isApprox(const Scalar& x, const Scalar& y,
-                          typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
+                     const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision())
 {
   return scalar_fuzzy_impl<Scalar>::isApprox(x, y, precision);
 }
 
-template<typename Scalar>
+template<typename Scalar> EIGEN_DEVICE_FUNC
 inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y,
-                                    typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
+                               const typename NumTraits<Scalar>::Real &precision = NumTraits<Scalar>::dummy_precision())
 {
   return scalar_fuzzy_impl<Scalar>::isApproxOrLessThan(x, y, precision);
 }
@@ -742,17 +1387,19 @@
 {
   typedef bool RealScalar;
   
-  template<typename OtherScalar>
+  template<typename OtherScalar> EIGEN_DEVICE_FUNC
   static inline bool isMuchSmallerThan(const bool& x, const bool&, const bool&)
   {
     return !x;
   }
   
+  EIGEN_DEVICE_FUNC
   static inline bool isApprox(bool x, bool y, bool)
   {
     return x == y;
   }
 
+  EIGEN_DEVICE_FUNC
   static inline bool isApproxOrLessThan(const bool& x, const bool& y, const bool&)
   {
     return (!x) || y;
diff --git a/Eigen/src/Core/MathFunctionsImpl.h b/Eigen/src/Core/MathFunctionsImpl.h
new file mode 100644
index 0000000..9c1ceb0
--- /dev/null
+++ b/Eigen/src/Core/MathFunctionsImpl.h
@@ -0,0 +1,101 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com)
+// Copyright (C) 2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_MATHFUNCTIONSIMPL_H
+#define EIGEN_MATHFUNCTIONSIMPL_H
+
+namespace Eigen {
+
+namespace internal {
+
+/** \internal \returns the hyperbolic tan of \a a (coeff-wise)
+    Doesn't do anything fancy, just a 13/6-degree rational interpolant which
+    is accurate up to a couple of ulp in the range [-9, 9], outside of which
+    the tanh(x) = +/-1.
+
+    This implementation works on both scalars and packets.
+*/
+template<typename T>
+T generic_fast_tanh_float(const T& a_x)
+{
+  // Clamp the inputs to the range [-9, 9] since anything outside
+  // this range is +/-1.0f in single-precision.
+  const T plus_9 = pset1<T>(9.f);
+  const T minus_9 = pset1<T>(-9.f);
+  // NOTE GCC prior to 6.3 might improperly optimize this max/min
+  //      step such that if a_x is nan, x will be either 9 or -9,
+  //      and tanh will return 1 or -1 instead of nan.
+  //      This is supposed to be fixed in gcc6.3,
+  //      see: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=72867
+  const T x = pmax(minus_9,pmin(plus_9,a_x));
+  // The monomial coefficients of the numerator polynomial (odd).
+  const T alpha_1 = pset1<T>(4.89352455891786e-03f);
+  const T alpha_3 = pset1<T>(6.37261928875436e-04f);
+  const T alpha_5 = pset1<T>(1.48572235717979e-05f);
+  const T alpha_7 = pset1<T>(5.12229709037114e-08f);
+  const T alpha_9 = pset1<T>(-8.60467152213735e-11f);
+  const T alpha_11 = pset1<T>(2.00018790482477e-13f);
+  const T alpha_13 = pset1<T>(-2.76076847742355e-16f);
+
+  // The monomial coefficients of the denominator polynomial (even).
+  const T beta_0 = pset1<T>(4.89352518554385e-03f);
+  const T beta_2 = pset1<T>(2.26843463243900e-03f);
+  const T beta_4 = pset1<T>(1.18534705686654e-04f);
+  const T beta_6 = pset1<T>(1.19825839466702e-06f);
+
+  // Since the polynomials are odd/even, we need x^2.
+  const T x2 = pmul(x, x);
+
+  // Evaluate the numerator polynomial p.
+  T p = pmadd(x2, alpha_13, alpha_11);
+  p = pmadd(x2, p, alpha_9);
+  p = pmadd(x2, p, alpha_7);
+  p = pmadd(x2, p, alpha_5);
+  p = pmadd(x2, p, alpha_3);
+  p = pmadd(x2, p, alpha_1);
+  p = pmul(x, p);
+
+  // Evaluate the denominator polynomial p.
+  T q = pmadd(x2, beta_6, beta_4);
+  q = pmadd(x2, q, beta_2);
+  q = pmadd(x2, q, beta_0);
+
+  // Divide the numerator by the denominator.
+  return pdiv(p, q);
+}
+
+template<typename RealScalar>
+EIGEN_STRONG_INLINE
+RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y)
+{
+  EIGEN_USING_STD_MATH(sqrt);
+  RealScalar p, qp;
+  p = numext::maxi(x,y);
+  if(p==RealScalar(0)) return RealScalar(0);
+  qp = numext::mini(y,x) / p;    
+  return p * sqrt(RealScalar(1) + qp*qp);
+}
+
+template<typename Scalar>
+struct hypot_impl
+{
+  typedef typename NumTraits<Scalar>::Real RealScalar;
+  static inline RealScalar run(const Scalar& x, const Scalar& y)
+  {
+    EIGEN_USING_STD_MATH(abs);
+    return positive_real_hypot<RealScalar>(abs(x), abs(y));
+  }
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_MATHFUNCTIONSIMPL_H
diff --git a/Eigen/src/Core/Matrix.h b/Eigen/src/Core/Matrix.h
index d7d0b5b..7f4a7af 100644
--- a/Eigen/src/Core/Matrix.h
+++ b/Eigen/src/Core/Matrix.h
@@ -13,6 +13,45 @@
 
 namespace Eigen {
 
+namespace internal {
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+{
+private:
+  enum { size = internal::size_at_compile_time<_Rows,_Cols>::ret };
+  typedef typename find_best_packet<_Scalar,size>::type PacketScalar;
+  enum {
+      row_major_bit = _Options&RowMajor ? RowMajorBit : 0,
+      is_dynamic_size_storage = _MaxRows==Dynamic || _MaxCols==Dynamic,
+      max_size = is_dynamic_size_storage ? Dynamic : _MaxRows*_MaxCols,
+      default_alignment = compute_default_alignment<_Scalar,max_size>::value,
+      actual_alignment = ((_Options&DontAlign)==0) ? default_alignment : 0,
+      required_alignment = unpacket_traits<PacketScalar>::alignment,
+      packet_access_bit = (packet_traits<_Scalar>::Vectorizable && (EIGEN_UNALIGNED_VECTORIZE || (actual_alignment>=required_alignment))) ? PacketAccessBit : 0
+    };
+    
+public:
+  typedef _Scalar Scalar;
+  typedef Dense StorageKind;
+  typedef Eigen::Index StorageIndex;
+  typedef MatrixXpr XprKind;
+  enum {
+    RowsAtCompileTime = _Rows,
+    ColsAtCompileTime = _Cols,
+    MaxRowsAtCompileTime = _MaxRows,
+    MaxColsAtCompileTime = _MaxCols,
+    Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
+    Options = _Options,
+    InnerStrideAtCompileTime = 1,
+    OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime,
+    
+    // FIXME, the following flag in only used to define NeedsToAlign in PlainObjectBase
+    EvaluatorFlags = LinearAccessBit | DirectAccessBit | packet_access_bit | row_major_bit,
+    Alignment = actual_alignment
+  };
+};
+}
+
 /** \class Matrix
   * \ingroup Core_Module
   *
@@ -24,13 +63,13 @@
   * The %Matrix class encompasses \em both fixed-size and dynamic-size objects (\ref fixedsize "note").
   *
   * The first three template parameters are required:
-  * \tparam _Scalar \anchor matrix_tparam_scalar Numeric type, e.g. float, double, int or std::complex<float>.
-  *                 User defined sclar types are supported as well (see \ref user_defined_scalars "here").
+  * \tparam _Scalar Numeric type, e.g. float, double, int or std::complex<float>.
+  *                 User defined scalar types are supported as well (see \ref user_defined_scalars "here").
   * \tparam _Rows Number of rows, or \b Dynamic
   * \tparam _Cols Number of columns, or \b Dynamic
   *
   * The remaining template parameters are optional -- in most cases you don't have to worry about them.
-  * \tparam _Options \anchor matrix_tparam_options A combination of either \b #RowMajor or \b #ColMajor, and of either
+  * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of either
   *                 \b #AutoAlign or \b #DontAlign.
   *                 The former controls \ref TopicStorageOrders "storage order", and defaults to column-major. The latter controls alignment, which is required
   *                 for vectorization. It defaults to aligning matrices except for fixed sizes that aren't a multiple of the packet size.
@@ -67,7 +106,7 @@
   * \endcode
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN.
   *
   * <i><b>Some notes:</b></i>
   *
@@ -97,32 +136,44 @@
   * are the dimensions of the original matrix, while _Rows and _Cols are Dynamic.</dd>
   * </dl>
   *
-  * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy, 
-  * \ref TopicStorageOrders 
+  * <i><b>ABI and storage layout</b></i>
+  *
+  * The table below summarizes the ABI of some possible Matrix instances which is fixed thorough the lifetime of Eigen 3.
+  * <table  class="manual">
+  * <tr><th>Matrix type</th><th>Equivalent C structure</th></tr>
+  * <tr><td>\code Matrix<T,Dynamic,Dynamic> \endcode</td><td>\code
+  * struct {
+  *   T *data;                  // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
+  *   Eigen::Index rows, cols;
+  *  };
+  * \endcode</td></tr>
+  * <tr class="alt"><td>\code
+  * Matrix<T,Dynamic,1>
+  * Matrix<T,1,Dynamic> \endcode</td><td>\code
+  * struct {
+  *   T *data;                  // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0
+  *   Eigen::Index size;
+  *  };
+  * \endcode</td></tr>
+  * <tr><td>\code Matrix<T,Rows,Cols> \endcode</td><td>\code
+  * struct {
+  *   T data[Rows*Cols];        // with (size_t(data)%A(Rows*Cols*sizeof(T)))==0
+  *  };
+  * \endcode</td></tr>
+  * <tr class="alt"><td>\code Matrix<T,Dynamic,Dynamic,0,MaxRows,MaxCols> \endcode</td><td>\code
+  * struct {
+  *   T data[MaxRows*MaxCols];  // with (size_t(data)%A(MaxRows*MaxCols*sizeof(T)))==0
+  *   Eigen::Index rows, cols;
+  *  };
+  * \endcode</td></tr>
+  * </table>
+  * Note that in this table Rows, Cols, MaxRows and MaxCols are all positive integers. A(S) is defined to the largest possible power-of-two
+  * smaller to EIGEN_MAX_STATIC_ALIGN_BYTES.
+  *
+  * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy,
+  * \ref TopicStorageOrders
   */
 
-namespace internal {
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct traits<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-{
-  typedef _Scalar Scalar;
-  typedef Dense StorageKind;
-  typedef DenseIndex Index;
-  typedef MatrixXpr XprKind;
-  enum {
-    RowsAtCompileTime = _Rows,
-    ColsAtCompileTime = _Cols,
-    MaxRowsAtCompileTime = _MaxRows,
-    MaxColsAtCompileTime = _MaxCols,
-    Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret,
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
-    Options = _Options,
-    InnerStrideAtCompileTime = 1,
-    OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime
-  };
-};
-}
-
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 class Matrix
   : public PlainObjectBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
@@ -151,6 +202,7 @@
       *
       * \callgraph
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const Matrix& other)
     {
       return Base::_set(other);
@@ -167,7 +219,8 @@
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix& operator=(const MatrixBase<OtherDerived>& other)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Matrix& operator=(const DenseBase<OtherDerived>& other)
     {
       return Base::_set(other);
     }
@@ -179,12 +232,14 @@
       * \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const EigenBase<OtherDerived> &other)
     {
       return Base::operator=(other);
     }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix& operator=(const ReturnByValue<OtherDerived>& func)
     {
       return Base::operator=(func);
@@ -200,6 +255,7 @@
       *
       * \sa resize(Index,Index)
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix() : Base()
     {
       Base::_check_template_params();
@@ -207,45 +263,85 @@
     }
 
     // FIXME is it still needed
-    Matrix(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC
+    explicit Matrix(internal::constructor_without_unaligned_array_assert)
       : Base(internal::constructor_without_unaligned_array_assert())
     { Base::_check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED }
 
-    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
-      *
-      * Note that this is only useful for dynamic-size vectors. For fixed-size vectors,
-      * it is redundant to pass the dimension here, so it makes more sense to use the default
-      * constructor Matrix() instead.
-      */
-    EIGEN_STRONG_INLINE explicit Matrix(Index dim)
-      : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim)
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    Matrix(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible<Scalar>::value)
+      : Base(std::move(other))
     {
       Base::_check_template_params();
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Matrix)
-      eigen_assert(dim >= 0);
-      eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim);
-      EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
+    EIGEN_DEVICE_FUNC
+    Matrix& operator=(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable<Scalar>::value)
+    {
+      other.swap(*this);
+      return *this;
+    }
+#endif
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
+
+    // This constructor is for both 1x1 matrices and dynamic vectors
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE explicit Matrix(const T& x)
+    {
+      Base::_check_template_params();
+      Base::template _init1<T>(x);
+    }
+
     template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const T0& x, const T1& y)
     {
       Base::_check_template_params();
       Base::template _init2<T0,T1>(x, y);
     }
     #else
+    /** \brief Constructs a fixed-sized matrix initialized with coefficients starting at \a data */
+    EIGEN_DEVICE_FUNC
+    explicit Matrix(const Scalar *data);
+
+    /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors
+      *
+      * This is useful for dynamic-size vectors. For fixed-size vectors,
+      * it is redundant to pass these parameters, so one should use the default constructor
+      * Matrix() instead.
+      * 
+      * \warning This constructor is disabled for fixed-size \c 1x1 matrices. For instance,
+      * calling Matrix<double,1,1>(1) will call the initialization constructor: Matrix(const Scalar&).
+      * For fixed-size \c 1x1 matrices it is therefore recommended to use the default
+      * constructor Matrix() instead, especially when using one of the non standard
+      * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
+      */
+    EIGEN_STRONG_INLINE explicit Matrix(Index dim);
+    /** \brief Constructs an initialized 1x1 matrix with the given coefficient */
+    Matrix(const Scalar& x);
     /** \brief Constructs an uninitialized matrix with \a rows rows and \a cols columns.
       *
       * This is useful for dynamic-size matrices. For fixed-size matrices,
       * it is redundant to pass these parameters, so one should use the default constructor
-      * Matrix() instead. */
+      * Matrix() instead.
+      * 
+      * \warning This constructor is disabled for fixed-size \c 1x2 and \c 2x1 vectors. For instance,
+      * calling Matrix2f(2,1) will call the initialization constructor: Matrix(const Scalar& x, const Scalar& y).
+      * For fixed-size \c 1x2 or \c 2x1 vectors it is therefore recommended to use the default
+      * constructor Matrix() instead, especially when using one of the non standard
+      * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives).
+      */
+    EIGEN_DEVICE_FUNC
     Matrix(Index rows, Index cols);
+    
     /** \brief Constructs an initialized 2D vector with given coefficients */
     Matrix(const Scalar& x, const Scalar& y);
     #endif
 
     /** \brief Constructs an initialized 3D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z)
     {
       Base::_check_template_params();
@@ -255,6 +351,7 @@
       m_storage.data()[2] = z;
     }
     /** \brief Constructs an initialized 4D vector with given coefficients */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w)
     {
       Base::_check_template_params();
@@ -265,76 +362,33 @@
       m_storage.data()[3] = w;
     }
 
-    explicit Matrix(const Scalar *data);
 
-    /** \brief Constructor copying the value of the expression \a other */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix(const MatrixBase<OtherDerived>& other)
-             : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      // This test resides here, to bring the error messages closer to the user. Normally, these checks
-      // are performed deeply within the library, thus causing long and scary error traces.
-      EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
     /** \brief Copy constructor */
-    EIGEN_STRONG_INLINE Matrix(const Matrix& other)
-            : Base(other.rows() * other.cols(), other.rows(), other.cols())
-    {
-      Base::_check_template_params();
-      Base::_set_noalias(other);
-    }
-    /** \brief Copy constructor with in-place evaluation */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE Matrix(const ReturnByValue<OtherDerived>& other)
-    {
-      Base::_check_template_params();
-      Base::resize(other.rows(), other.cols());
-      other.evalTo(*this);
-    }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE Matrix(const Matrix& other) : Base(other)
+    { }
 
     /** \brief Copy constructor for generic expressions.
       * \sa MatrixBase::operator=(const EigenBase<OtherDerived>&)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Matrix(const EigenBase<OtherDerived> &other)
-      : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      Base::_check_template_params();
-      Base::_resize_to_match(other);
-      // FIXME/CHECK: isn't *this = other.derived() more efficient. it allows to
-      //              go for pure _set() implementations, right?
-      *this = other;
-    }
+      : Base(other.derived())
+    { }
 
-    /** \internal
-      * \brief Override MatrixBase::swap() since for dynamic-sized matrices
-      * of same type it is enough to swap the data pointers.
-      */
-    template<typename OtherDerived>
-    void swap(MatrixBase<OtherDerived> const & other)
-    { this->_swap(other.derived()); }
-
-    inline Index innerStride() const { return 1; }
-    inline Index outerStride() const { return this->innerSize(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return 1; }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return this->innerSize(); }
 
     /////////// Geometry module ///////////
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     explicit Matrix(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Matrix& operator=(const RotationBase<OtherDerived,ColsAtCompileTime>& r);
 
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    explicit Matrix(const eigen2_RotationBase<OtherDerived,ColsAtCompileTime>& r);
-    template<typename OtherDerived>
-    Matrix& operator=(const eigen2_RotationBase<OtherDerived,ColsAtCompileTime>& r);
-    #endif
-
     // allow to extend Matrix outside Eigen
     #ifdef EIGEN_MATRIX_PLUGIN
     #include EIGEN_MATRIX_PLUGIN
diff --git a/Eigen/src/Core/MatrixBase.h b/Eigen/src/Core/MatrixBase.h
index b67a7c1..e6c3590 100644
--- a/Eigen/src/Core/MatrixBase.h
+++ b/Eigen/src/Core/MatrixBase.h
@@ -41,9 +41,9 @@
   * \endcode
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN.
   *
-  * \sa \ref TopicClassHierarchy
+  * \sa \blank \ref TopicClassHierarchy
   */
 template<typename Derived> class MatrixBase
   : public DenseBase<Derived>
@@ -52,7 +52,7 @@
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef MatrixBase StorageBaseType;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
@@ -66,7 +66,6 @@
     using Base::MaxSizeAtCompileTime;
     using Base::IsVectorAtCompileTime;
     using Base::Flags;
-    using Base::CoeffReadCost;
 
     using Base::derived;
     using Base::const_cast_derived;
@@ -98,25 +97,14 @@
 
     /** \returns the size of the main diagonal, which is min(rows(),cols()).
       * \sa rows(), cols(), SizeAtCompileTime. */
-    inline Index diagonalSize() const { return (std::min)(rows(),cols()); }
+    EIGEN_DEVICE_FUNC
+    inline Index diagonalSize() const { return (numext::mini)(rows(),cols()); }
 
-    /** \brief The plain matrix type corresponding to this expression.
-      *
-      * This is not necessarily exactly the return type of eval(). In the case of plain matrices,
-      * the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed
-      * that the return type of eval() is either PlainObject or const PlainObject&.
-      */
-    typedef Matrix<typename internal::traits<Derived>::Scalar,
-                internal::traits<Derived>::RowsAtCompileTime,
-                internal::traits<Derived>::ColsAtCompileTime,
-                AutoAlign | (internal::traits<Derived>::Flags&RowMajorBit ? RowMajor : ColMajor),
-                internal::traits<Derived>::MaxRowsAtCompileTime,
-                internal::traits<Derived>::MaxColsAtCompileTime
-          > PlainObject;
+    typedef typename Base::PlainObject PlainObject;
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal Represents a matrix with all coefficients equal to one another*/
-    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Derived> ConstantReturnType;
+    typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,PlainObject> ConstantReturnType;
     /** \internal the return type of MatrixBase::adjoint() */
     typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
                         CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
@@ -125,7 +113,7 @@
     /** \internal Return type of eigenvalues() */
     typedef Matrix<std::complex<RealScalar>, internal::traits<Derived>::ColsAtCompileTime, 1, ColMajor> EigenvaluesReturnType;
     /** \internal the return type of identity */
-    typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,Derived> IdentityReturnType;
+    typedef CwiseNullaryOp<internal::scalar_identity_op<Scalar>,PlainObject> IdentityReturnType;
     /** \internal the return type of unit vectors */
     typedef Block<const CwiseNullaryOp<internal::scalar_identity_op<Scalar>, SquareMatrixType>,
                   internal::traits<Derived>::RowsAtCompileTime,
@@ -133,6 +121,7 @@
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::MatrixBase
+#define EIGEN_DOC_UNARY_ADDONS(X,Y)
 #   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
@@ -141,41 +130,44 @@
 #     include EIGEN_MATRIXBASE_PLUGIN
 #   endif
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_UNARY_ADDONS
 
     /** Special case of the template operator=, in order to prevent the compiler
       * from generating a default operator= (issue hit with g++ 4.1)
       */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const MatrixBase& other);
 
     // We cannot inherit here via Base::operator= since it is causing
     // trouble with MSVC.
 
     template <typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator=(const DenseBase<OtherDerived>& other);
 
     template <typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const EigenBase<OtherDerived>& other);
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     Derived& operator=(const ReturnByValue<OtherDerived>& other);
 
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& lazyAssign(const ProductBase<ProductDerived, Lhs,Rhs>& other);
-
-    template<typename MatrixPower, typename Lhs, typename Rhs>
-    Derived& lazyAssign(const MatrixPowerProduct<MatrixPower, Lhs,Rhs>& other);
-
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator+=(const MatrixBase<OtherDerived>& other);
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
     Derived& operator-=(const MatrixBase<OtherDerived>& other);
 
     template<typename OtherDerived>
-    const typename ProductReturnType<Derived,OtherDerived>::Type
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,OtherDerived>
     operator*(const MatrixBase<OtherDerived> &other) const;
 
     template<typename OtherDerived>
-    const typename LazyProductReturnType<Derived,OtherDerived>::Type
+    EIGEN_DEVICE_FUNC
+    const Product<Derived,OtherDerived,LazyProduct>
     lazyProduct(const MatrixBase<OtherDerived> &other) const;
 
     template<typename OtherDerived>
@@ -188,84 +180,93 @@
     void applyOnTheRight(const EigenBase<OtherDerived>& other);
 
     template<typename DiagonalDerived>
-    const DiagonalProduct<Derived, DiagonalDerived, OnTheRight>
+    EIGEN_DEVICE_FUNC
+    const Product<Derived, DiagonalDerived, LazyProduct>
     operator*(const DiagonalBase<DiagonalDerived> &diagonal) const;
 
     template<typename OtherDerived>
-    typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
+    EIGEN_DEVICE_FUNC
+    typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType
     dot(const MatrixBase<OtherDerived>& other) const;
 
-    #ifdef EIGEN2_SUPPORT
-      template<typename OtherDerived>
-      Scalar eigen2_dot(const MatrixBase<OtherDerived>& other) const;
-    #endif
-
-    RealScalar squaredNorm() const;
-    RealScalar norm() const;
+    EIGEN_DEVICE_FUNC RealScalar squaredNorm() const;
+    EIGEN_DEVICE_FUNC RealScalar norm() const;
     RealScalar stableNorm() const;
     RealScalar blueNorm() const;
     RealScalar hypotNorm() const;
-    const PlainObject normalized() const;
-    void normalize();
+    EIGEN_DEVICE_FUNC const PlainObject normalized() const;
+    EIGEN_DEVICE_FUNC const PlainObject stableNormalized() const;
+    EIGEN_DEVICE_FUNC void normalize();
+    EIGEN_DEVICE_FUNC void stableNormalize();
 
-    const AdjointReturnType adjoint() const;
-    void adjointInPlace();
+    EIGEN_DEVICE_FUNC const AdjointReturnType adjoint() const;
+    EIGEN_DEVICE_FUNC void adjointInPlace();
 
     typedef Diagonal<Derived> DiagonalReturnType;
+    EIGEN_DEVICE_FUNC
     DiagonalReturnType diagonal();
+
     typedef typename internal::add_const<Diagonal<const Derived> >::type ConstDiagonalReturnType;
+    EIGEN_DEVICE_FUNC
     ConstDiagonalReturnType diagonal() const;
 
     template<int Index> struct DiagonalIndexReturnType { typedef Diagonal<Derived,Index> Type; };
     template<int Index> struct ConstDiagonalIndexReturnType { typedef const Diagonal<const Derived,Index> Type; };
 
-    template<int Index> typename DiagonalIndexReturnType<Index>::Type diagonal();
-    template<int Index> typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
-    
+    template<int Index>
+    EIGEN_DEVICE_FUNC
+    typename DiagonalIndexReturnType<Index>::Type diagonal();
+
+    template<int Index>
+    EIGEN_DEVICE_FUNC
+    typename ConstDiagonalIndexReturnType<Index>::Type diagonal() const;
+
     typedef Diagonal<Derived,DynamicIndex> DiagonalDynamicIndexReturnType;
     typedef typename internal::add_const<Diagonal<const Derived,DynamicIndex> >::type ConstDiagonalDynamicIndexReturnType;
 
+    EIGEN_DEVICE_FUNC
     DiagonalDynamicIndexReturnType diagonal(Index index);
+    EIGEN_DEVICE_FUNC
     ConstDiagonalDynamicIndexReturnType diagonal(Index index) const;
 
-    #ifdef EIGEN2_SUPPORT
-    template<unsigned int Mode> typename internal::eigen2_part_return_type<Derived, Mode>::type part();
-    template<unsigned int Mode> const typename internal::eigen2_part_return_type<Derived, Mode>::type part() const;
-    
-    // huuuge hack. make Eigen2's matrix.part<Diagonal>() work in eigen3. Problem: Diagonal is now a class template instead
-    // of an integer constant. Solution: overload the part() method template wrt template parameters list.
-    template<template<typename T, int N> class U>
-    const DiagonalWrapper<ConstDiagonalReturnType> part() const
-    { return diagonal().asDiagonal(); }
-    #endif // EIGEN2_SUPPORT
-
     template<unsigned int Mode> struct TriangularViewReturnType { typedef TriangularView<Derived, Mode> Type; };
     template<unsigned int Mode> struct ConstTriangularViewReturnType { typedef const TriangularView<const Derived, Mode> Type; };
 
-    template<unsigned int Mode> typename TriangularViewReturnType<Mode>::Type triangularView();
-    template<unsigned int Mode> typename ConstTriangularViewReturnType<Mode>::Type triangularView() const;
+    template<unsigned int Mode>
+    EIGEN_DEVICE_FUNC
+    typename TriangularViewReturnType<Mode>::Type triangularView();
+    template<unsigned int Mode>
+    EIGEN_DEVICE_FUNC
+    typename ConstTriangularViewReturnType<Mode>::Type triangularView() const;
 
     template<unsigned int UpLo> struct SelfAdjointViewReturnType { typedef SelfAdjointView<Derived, UpLo> Type; };
     template<unsigned int UpLo> struct ConstSelfAdjointViewReturnType { typedef const SelfAdjointView<const Derived, UpLo> Type; };
 
-    template<unsigned int UpLo> typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
-    template<unsigned int UpLo> typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
+    template<unsigned int UpLo>
+    EIGEN_DEVICE_FUNC
+    typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
+    template<unsigned int UpLo>
+    EIGEN_DEVICE_FUNC
+    typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
 
     const SparseView<Derived> sparseView(const Scalar& m_reference = Scalar(0),
                                          const typename NumTraits<Scalar>::Real& m_epsilon = NumTraits<Scalar>::dummy_precision()) const;
-    static const IdentityReturnType Identity();
-    static const IdentityReturnType Identity(Index rows, Index cols);
-    static const BasisReturnType Unit(Index size, Index i);
-    static const BasisReturnType Unit(Index i);
-    static const BasisReturnType UnitX();
-    static const BasisReturnType UnitY();
-    static const BasisReturnType UnitZ();
-    static const BasisReturnType UnitW();
+    EIGEN_DEVICE_FUNC static const IdentityReturnType Identity();
+    EIGEN_DEVICE_FUNC static const IdentityReturnType Identity(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index size, Index i);
+    EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index i);
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitX();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitY();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitZ();
+    EIGEN_DEVICE_FUNC static const BasisReturnType UnitW();
 
+    EIGEN_DEVICE_FUNC
     const DiagonalWrapper<const Derived> asDiagonal() const;
     const PermutationWrapper<const Derived> asPermutation() const;
 
+    EIGEN_DEVICE_FUNC
     Derived& setIdentity();
+    EIGEN_DEVICE_FUNC
     Derived& setIdentity(Index rows, Index cols);
 
     bool isIdentity(const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
@@ -284,7 +285,7 @@
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator!= */
     template<typename OtherDerived>
-    inline bool operator==(const MatrixBase<OtherDerived>& other) const
+    EIGEN_DEVICE_FUNC inline bool operator==(const MatrixBase<OtherDerived>& other) const
     { return cwiseEqual(other).all(); }
 
     /** \returns true if at least one pair of coefficients of \c *this and \a other are not exactly equal to each other.
@@ -292,64 +293,50 @@
       *          fuzzy comparison such as isApprox()
       * \sa isApprox(), operator== */
     template<typename OtherDerived>
-    inline bool operator!=(const MatrixBase<OtherDerived>& other) const
+    EIGEN_DEVICE_FUNC inline bool operator!=(const MatrixBase<OtherDerived>& other) const
     { return cwiseNotEqual(other).any(); }
 
     NoAlias<Derived,Eigen::MatrixBase > noalias();
 
-    inline const ForceAlignedAccess<Derived> forceAlignedAccess() const;
-    inline ForceAlignedAccess<Derived> forceAlignedAccess();
-    template<bool Enable> inline typename internal::add_const_on_value_type<typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type>::type forceAlignedAccessIf() const;
-    template<bool Enable> inline typename internal::conditional<Enable,ForceAlignedAccess<Derived>,Derived&>::type forceAlignedAccessIf();
+    // TODO forceAlignedAccess is temporarily disabled
+    // Need to find a nicer workaround.
+    inline const Derived& forceAlignedAccess() const { return derived(); }
+    inline Derived& forceAlignedAccess() { return derived(); }
+    template<bool Enable> inline const Derived& forceAlignedAccessIf() const { return derived(); }
+    template<bool Enable> inline Derived& forceAlignedAccessIf() { return derived(); }
 
-    Scalar trace() const;
+    EIGEN_DEVICE_FUNC Scalar trace() const;
 
-/////////// Array module ///////////
+    template<int p> EIGEN_DEVICE_FUNC RealScalar lpNorm() const;
 
-    template<int p> RealScalar lpNorm() const;
-
-    MatrixBase<Derived>& matrix() { return *this; }
-    const MatrixBase<Derived>& matrix() const { return *this; }
+    EIGEN_DEVICE_FUNC MatrixBase<Derived>& matrix() { return *this; }
+    EIGEN_DEVICE_FUNC const MatrixBase<Derived>& matrix() const { return *this; }
 
     /** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix
       * \sa ArrayBase::matrix() */
-    ArrayWrapper<Derived> array() { return derived(); }
-    const ArrayWrapper<const Derived> array() const { return derived(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ArrayWrapper<Derived> array() { return ArrayWrapper<Derived>(derived()); }
+    /** \returns a const \link Eigen::ArrayBase Array \endlink expression of this matrix
+      * \sa ArrayBase::matrix() */
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ArrayWrapper<const Derived> array() const { return ArrayWrapper<const Derived>(derived()); }
 
 /////////// LU module ///////////
 
-    const FullPivLU<PlainObject> fullPivLu() const;
-    const PartialPivLU<PlainObject> partialPivLu() const;
+    inline const FullPivLU<PlainObject> fullPivLu() const;
+    inline const PartialPivLU<PlainObject> partialPivLu() const;
 
-    #if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS
-    const LU<PlainObject> lu() const;
-    #endif
+    inline const PartialPivLU<PlainObject> lu() const;
 
-    #ifdef EIGEN2_SUPPORT
-    const LU<PlainObject> eigen2_lu() const;
-    #endif
+    inline const Inverse<Derived> inverse() const;
 
-    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-    const PartialPivLU<PlainObject> lu() const;
-    #endif
-    
-    #ifdef EIGEN2_SUPPORT
     template<typename ResultType>
-    void computeInverse(MatrixBase<ResultType> *result) const {
-      *result = this->inverse();
-    }
-    #endif
-
-    const internal::inverse_impl<Derived> inverse() const;
-    template<typename ResultType>
-    void computeInverseAndDetWithCheck(
+    inline void computeInverseAndDetWithCheck(
       ResultType& inverse,
       typename ResultType::Scalar& determinant,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
     ) const;
     template<typename ResultType>
-    void computeInverseWithCheck(
+    inline void computeInverseWithCheck(
       ResultType& inverse,
       bool& invertible,
       const RealScalar& absDeterminantThreshold = NumTraits<Scalar>::dummy_precision()
@@ -358,65 +345,70 @@
 
 /////////// Cholesky module ///////////
 
-    const LLT<PlainObject>  llt() const;
-    const LDLT<PlainObject> ldlt() const;
+    inline const LLT<PlainObject>  llt() const;
+    inline const LDLT<PlainObject> ldlt() const;
 
 /////////// QR module ///////////
 
-    const HouseholderQR<PlainObject> householderQr() const;
-    const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
-    const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
-    
-    #ifdef EIGEN2_SUPPORT
-    const QR<PlainObject> qr() const;
-    #endif
+    inline const HouseholderQR<PlainObject> householderQr() const;
+    inline const ColPivHouseholderQR<PlainObject> colPivHouseholderQr() const;
+    inline const FullPivHouseholderQR<PlainObject> fullPivHouseholderQr() const;
+    inline const CompleteOrthogonalDecomposition<PlainObject> completeOrthogonalDecomposition() const;
 
-    EigenvaluesReturnType eigenvalues() const;
-    RealScalar operatorNorm() const;
+/////////// Eigenvalues module ///////////
+
+    inline EigenvaluesReturnType eigenvalues() const;
+    inline RealScalar operatorNorm() const;
 
 /////////// SVD module ///////////
 
-    JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
-
-    #ifdef EIGEN2_SUPPORT
-    SVD<PlainObject> svd() const;
-    #endif
+    inline JacobiSVD<PlainObject> jacobiSvd(unsigned int computationOptions = 0) const;
+    inline BDCSVD<PlainObject>    bdcSvd(unsigned int computationOptions = 0) const;
 
 /////////// Geometry module ///////////
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /// \internal helper struct to form the return type of the cross product
     template<typename OtherDerived> struct cross_product_return_type {
-      typedef typename internal::scalar_product_traits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType Scalar;
+      typedef typename ScalarBinaryOpTraits<typename internal::traits<Derived>::Scalar,typename internal::traits<OtherDerived>::Scalar>::ReturnType Scalar;
       typedef Matrix<Scalar,MatrixBase::RowsAtCompileTime,MatrixBase::ColsAtCompileTime> type;
     };
     #endif // EIGEN_PARSED_BY_DOXYGEN
     template<typename OtherDerived>
-    typename cross_product_return_type<OtherDerived>::type
+    EIGEN_DEVICE_FUNC
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    inline typename cross_product_return_type<OtherDerived>::type
+#else
+    inline PlainObject
+#endif
     cross(const MatrixBase<OtherDerived>& other) const;
+
     template<typename OtherDerived>
-    PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
-    PlainObject unitOrthogonal(void) const;
-    Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
-    
-    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-    ScalarMultipleReturnType operator*(const UniformScaling<Scalar>& s) const;
+    EIGEN_DEVICE_FUNC
+    inline PlainObject cross3(const MatrixBase<OtherDerived>& other) const;
+
+    EIGEN_DEVICE_FUNC
+    inline PlainObject unitOrthogonal(void) const;
+
+    EIGEN_DEVICE_FUNC
+    inline Matrix<Scalar,3,1> eulerAngles(Index a0, Index a1, Index a2) const;
+
     // put this as separate enum value to work around possible GCC 4.3 bug (?)
-    enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1?Vertical:Horizontal };
+    enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1&&RowsAtCompileTime==1 ? ((internal::traits<Derived>::Flags&RowMajorBit)==RowMajorBit ? Horizontal : Vertical)
+                                          : ColsAtCompileTime==1 ? Vertical : Horizontal };
     typedef Homogeneous<Derived, HomogeneousReturnTypeDirection> HomogeneousReturnType;
-    HomogeneousReturnType homogeneous() const;
-    #endif
-    
+    EIGEN_DEVICE_FUNC
+    inline HomogeneousReturnType homogeneous() const;
+
     enum {
       SizeMinusOne = SizeAtCompileTime==Dynamic ? Dynamic : SizeAtCompileTime-1
     };
     typedef Block<const Derived,
                   internal::traits<Derived>::ColsAtCompileTime==1 ? SizeMinusOne : 1,
                   internal::traits<Derived>::ColsAtCompileTime==1 ? 1 : SizeMinusOne> ConstStartMinusOne;
-    typedef CwiseUnaryOp<internal::scalar_quotient1_op<typename internal::traits<Derived>::Scalar>,
-                const ConstStartMinusOne > HNormalizedReturnType;
-
-    const HNormalizedReturnType hnormalized() const;
+    typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(ConstStartMinusOne,Scalar,quotient) HNormalizedReturnType;
+    EIGEN_DEVICE_FUNC
+    inline const HNormalizedReturnType hnormalized() const;
 
 ////////// Householder module ///////////
 
@@ -440,61 +432,44 @@
     template<typename OtherScalar>
     void applyOnTheRight(Index p, Index q, const JacobiRotation<OtherScalar>& j);
 
+///////// SparseCore module /////////
+
+    template<typename OtherDerived>
+    EIGEN_STRONG_INLINE const typename SparseMatrixBase<OtherDerived>::template CwiseProductDenseReturnType<Derived>::Type
+    cwiseProduct(const SparseMatrixBase<OtherDerived> &other) const
+    {
+      return other.cwiseProduct(derived());
+    }
+
 ///////// MatrixFunctions module /////////
 
     typedef typename internal::stem_function<Scalar>::type StemFunction;
-    const MatrixExponentialReturnValue<Derived> exp() const;
+#define EIGEN_MATRIX_FUNCTION(ReturnType, Name, Description) \
+    /** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
+    const ReturnType<Derived> Name() const;
+#define EIGEN_MATRIX_FUNCTION_1(ReturnType, Name, Description, Argument) \
+    /** \returns an expression of the matrix Description of \c *this. \brief This function requires the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>. To compute the coefficient-wise Description use ArrayBase::##Name . */ \
+    const ReturnType<Derived> Name(Argument) const;
+
+    EIGEN_MATRIX_FUNCTION(MatrixExponentialReturnValue, exp, exponential)
+    /** \brief Helper function for the <a href="unsupported/group__MatrixFunctions__Module.html"> unsupported MatrixFunctions module</a>.*/
     const MatrixFunctionReturnValue<Derived> matrixFunction(StemFunction f) const;
-    const MatrixFunctionReturnValue<Derived> cosh() const;
-    const MatrixFunctionReturnValue<Derived> sinh() const;
-    const MatrixFunctionReturnValue<Derived> cos() const;
-    const MatrixFunctionReturnValue<Derived> sin() const;
-    const MatrixSquareRootReturnValue<Derived> sqrt() const;
-    const MatrixLogarithmReturnValue<Derived> log() const;
-    const MatrixPowerReturnValue<Derived> pow(const RealScalar& p) const;
-
-#ifdef EIGEN2_SUPPORT
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& operator+=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                      EvalBeforeAssigningBit>& other);
-
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    Derived& operator-=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                      EvalBeforeAssigningBit>& other);
-
-    /** \deprecated because .lazy() is deprecated
-      * Overloaded for cache friendly product evaluation */
-    template<typename OtherDerived>
-    Derived& lazyAssign(const Flagged<OtherDerived, 0, EvalBeforeAssigningBit>& other)
-    { return lazyAssign(other._expression()); }
-
-    template<unsigned int Added>
-    const Flagged<Derived, Added, 0> marked() const;
-    const Flagged<Derived, 0, EvalBeforeAssigningBit> lazy() const;
-
-    inline const Cwise<Derived> cwise() const;
-    inline Cwise<Derived> cwise();
-
-    VectorBlock<Derived> start(Index size);
-    const VectorBlock<const Derived> start(Index size) const;
-    VectorBlock<Derived> end(Index size);
-    const VectorBlock<const Derived> end(Index size) const;
-    template<int Size> VectorBlock<Derived,Size> start();
-    template<int Size> const VectorBlock<const Derived,Size> start() const;
-    template<int Size> VectorBlock<Derived,Size> end();
-    template<int Size> const VectorBlock<const Derived,Size> end() const;
-
-    Minor<Derived> minor(Index row, Index col);
-    const Minor<Derived> minor(Index row, Index col) const;
-#endif
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cosh, hyperbolic cosine)
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sinh, hyperbolic sine)
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cos, cosine)
+    EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sin, sine)
+    EIGEN_MATRIX_FUNCTION(MatrixSquareRootReturnValue, sqrt, square root)
+    EIGEN_MATRIX_FUNCTION(MatrixLogarithmReturnValue, log, logarithm)
+    EIGEN_MATRIX_FUNCTION_1(MatrixPowerReturnValue,        pow, power to \c p, const RealScalar& p)
+    EIGEN_MATRIX_FUNCTION_1(MatrixComplexPowerReturnValue, pow, power to \c p, const std::complex<RealScalar>& p)
 
   protected:
-    MatrixBase() : Base() {}
+    EIGEN_DEVICE_FUNC MatrixBase() : Base() {}
 
   private:
-    explicit MatrixBase(int);
-    MatrixBase(int,int);
-    template<typename OtherDerived> explicit MatrixBase(const MatrixBase<OtherDerived>&);
+    EIGEN_DEVICE_FUNC explicit MatrixBase(int);
+    EIGEN_DEVICE_FUNC MatrixBase(int,int);
+    template<typename OtherDerived> EIGEN_DEVICE_FUNC explicit MatrixBase(const MatrixBase<OtherDerived>&);
   protected:
     // mixing arrays and matrices is not legal
     template<typename OtherDerived> Derived& operator+=(const ArrayBase<OtherDerived>& )
diff --git a/Eigen/src/Core/NestByValue.h b/Eigen/src/Core/NestByValue.h
index a893b17..13adf07 100644
--- a/Eigen/src/Core/NestByValue.h
+++ b/Eigen/src/Core/NestByValue.h
@@ -13,25 +13,24 @@
 
 namespace Eigen {
 
-/** \class NestByValue
-  * \ingroup Core_Module
-  *
-  * \brief Expression which must be nested by value
-  *
-  * \param ExpressionType the type of the object of which we are requiring nesting-by-value
-  *
-  * This class is the return type of MatrixBase::nestByValue()
-  * and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::nestByValue()
-  */
-
 namespace internal {
 template<typename ExpressionType>
 struct traits<NestByValue<ExpressionType> > : public traits<ExpressionType>
 {};
 }
 
+/** \class NestByValue
+  * \ingroup Core_Module
+  *
+  * \brief Expression which must be nested by value
+  *
+  * \tparam ExpressionType the type of the object of which we are requiring nesting-by-value
+  *
+  * This class is the return type of MatrixBase::nestByValue()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::nestByValue()
+  */
 template<typename ExpressionType> class NestByValue
   : public internal::dense_xpr_base< NestByValue<ExpressionType> >::type
 {
@@ -40,29 +39,29 @@
     typedef typename internal::dense_xpr_base<NestByValue>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(NestByValue)
 
-    inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
+    EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {}
 
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_expression.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_expression.cols(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return m_expression.outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return m_expression.innerStride(); }
 
-    inline const CoeffReturnType coeff(Index row, Index col) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const
     {
       return m_expression.coeff(row, col);
     }
 
-    inline Scalar& coeffRef(Index row, Index col)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col)
     {
       return m_expression.const_cast_derived().coeffRef(row, col);
     }
 
-    inline const CoeffReturnType coeff(Index index) const
+    EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const
     {
       return m_expression.coeff(index);
     }
 
-    inline Scalar& coeffRef(Index index)
+    EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index)
     {
       return m_expression.const_cast_derived().coeffRef(index);
     }
@@ -91,7 +90,7 @@
       m_expression.const_cast_derived().template writePacket<LoadMode>(index, x);
     }
 
-    operator const ExpressionType&() const { return m_expression; }
+    EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; }
 
   protected:
     const ExpressionType m_expression;
diff --git a/Eigen/src/Core/NoAlias.h b/Eigen/src/Core/NoAlias.h
index 768bfb1..3390801 100644
--- a/Eigen/src/Core/NoAlias.h
+++ b/Eigen/src/Core/NoAlias.h
@@ -17,7 +17,7 @@
   *
   * \brief Pseudo expression providing an operator = assuming no aliasing
   *
-  * \param ExpressionType the type of the object on which to do the lazy assignment
+  * \tparam ExpressionType the type of the object on which to do the lazy assignment
   *
   * This class represents an expression with special assignment operators
   * assuming no aliasing between the target expression and the source expression.
@@ -30,62 +30,36 @@
 template<typename ExpressionType, template <typename> class StorageBase>
 class NoAlias
 {
-    typedef typename ExpressionType::Scalar Scalar;
   public:
-    NoAlias(ExpressionType& expression) : m_expression(expression) {}
-
-    /** Behaves like MatrixBase::lazyAssign(other)
-      * \sa MatrixBase::lazyAssign() */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase<OtherDerived>& other)
-    { return internal::assign_selector<ExpressionType,OtherDerived,false>::run(m_expression,other.derived()); }
-
-    /** \sa MatrixBase::operator+= */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase<OtherDerived>& other)
-    {
-      typedef SelfCwiseBinaryOp<internal::scalar_sum_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
-      SelfAdder tmp(m_expression);
-      typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
-      typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
-      internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
-      return m_expression;
-    }
-
-    /** \sa MatrixBase::operator-= */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
-    {
-      typedef SelfCwiseBinaryOp<internal::scalar_difference_op<Scalar>, ExpressionType, OtherDerived> SelfAdder;
-      SelfAdder tmp(m_expression);
-      typedef typename internal::nested<OtherDerived>::type OtherDerivedNested;
-      typedef typename internal::remove_all<OtherDerivedNested>::type _OtherDerivedNested;
-      internal::assign_selector<SelfAdder,_OtherDerivedNested,false>::run(tmp,OtherDerivedNested(other.derived()));
-      return m_expression;
-    }
-
-#ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE ExpressionType& operator+=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    { other.derived().addTo(m_expression); return m_expression; }
-
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    { other.derived().subTo(m_expression); return m_expression; }
-
-    template<typename Lhs, typename Rhs, int NestingFlags>
-    EIGEN_STRONG_INLINE ExpressionType& operator+=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
-    { return m_expression.derived() += CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
-
-    template<typename Lhs, typename Rhs, int NestingFlags>
-    EIGEN_STRONG_INLINE ExpressionType& operator-=(const CoeffBasedProduct<Lhs,Rhs,NestingFlags>& other)
-    { return m_expression.derived() -= CoeffBasedProduct<Lhs,Rhs,NestByRefBit>(other.lhs(), other.rhs()); }
+    typedef typename ExpressionType::Scalar Scalar;
+    
+    explicit NoAlias(ExpressionType& expression) : m_expression(expression) {}
     
     template<typename OtherDerived>
-    ExpressionType& operator=(const ReturnByValue<OtherDerived>& func)
-    { return m_expression = func; }
-#endif
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase<OtherDerived>& other)
+    {
+      call_assignment_no_alias(m_expression, other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+      return m_expression;
+    }
+    
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase<OtherDerived>& other)
+    {
+      call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+      return m_expression;
+    }
+    
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase<OtherDerived>& other)
+    {
+      call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+      return m_expression;
+    }
 
+    EIGEN_DEVICE_FUNC
     ExpressionType& expression() const
     {
       return m_expression;
@@ -126,7 +100,7 @@
 template<typename Derived>
 NoAlias<Derived,MatrixBase> MatrixBase<Derived>::noalias()
 {
-  return derived();
+  return NoAlias<Derived, Eigen::MatrixBase >(derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/NumTraits.h b/Eigen/src/Core/NumTraits.h
index bac9e50..daf4898 100644
--- a/Eigen/src/Core/NumTraits.h
+++ b/Eigen/src/Core/NumTraits.h
@@ -12,24 +12,57 @@
 
 namespace Eigen {
 
+namespace internal {
+
+// default implementation of digits10(), based on numeric_limits if specialized,
+// 0 for integer types, and log10(epsilon()) otherwise.
+template< typename T,
+          bool use_numeric_limits = std::numeric_limits<T>::is_specialized,
+          bool is_integer = NumTraits<T>::IsInteger>
+struct default_digits10_impl
+{
+  static int run() { return std::numeric_limits<T>::digits10; }
+};
+
+template<typename T>
+struct default_digits10_impl<T,false,false> // Floating point
+{
+  static int run() {
+    using std::log10;
+    using std::ceil;
+    typedef typename NumTraits<T>::Real Real;
+    return int(ceil(-log10(NumTraits<Real>::epsilon())));
+  }
+};
+
+template<typename T>
+struct default_digits10_impl<T,false,true> // Integer
+{
+  static int run() { return 0; }
+};
+
+} // end namespace internal
+
 /** \class NumTraits
   * \ingroup Core_Module
   *
   * \brief Holds information about the various numeric (i.e. scalar) types allowed by Eigen.
   *
-  * \param T the numeric type at hand
+  * \tparam T the numeric type at hand
   *
   * This class stores enums, typedefs and static methods giving information about a numeric type.
   *
   * The provided data consists of:
-  * \li A typedef \a Real, giving the "real part" type of \a T. If \a T is already real,
-  *     then \a Real is just a typedef to \a T. If \a T is \c std::complex<U> then \a Real
+  * \li A typedef \c Real, giving the "real part" type of \a T. If \a T is already real,
+  *     then \c Real is just a typedef to \a T. If \a T is \c std::complex<U> then \c Real
   *     is a typedef to \a U.
-  * \li A typedef \a NonInteger, giving the type that should be used for operations producing non-integral values,
+  * \li A typedef \c NonInteger, giving the type that should be used for operations producing non-integral values,
   *     such as quotients, square roots, etc. If \a T is a floating-point type, then this typedef just gives
   *     \a T again. Note however that many Eigen functions such as internal::sqrt simply refuse to
   *     take integers. Outside of a few cases, Eigen doesn't do automatic type promotion. Thus, this typedef is
   *     only intended as a helper for code that needs to explicitly promote types.
+  * \li A typedef \c Literal giving the type to use for numeric literals such as "2" or "0.5". For instance, for \c std::complex<U>, Literal is defined as \c U.
+  *     Of course, this type must be fully compatible with \a T. In doubt, just use \a T here.
   * \li A typedef \a Nested giving the type to use to nest a value inside of the expression tree. If you don't know what
   *     this means, just use \a T here.
   * \li An enum value \a IsComplex. It is equal to 1 if \a T is a \c std::complex
@@ -42,10 +75,14 @@
   * \li An enum value \a IsSigned. It is equal to \c 1 if \a T is a signed type and to 0 if \a T is unsigned.
   * \li An enum value \a RequireInitialization. It is equal to \c 1 if the constructor of the numeric type \a T must
   *     be called, and to 0 if it is safe not to call it. Default is 0 if \a T is an arithmetic type, and 1 otherwise.
-  * \li An epsilon() function which, unlike std::numeric_limits::epsilon(), returns a \a Real instead of a \a T.
+  * \li An epsilon() function which, unlike <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/epsilon">std::numeric_limits::epsilon()</a>,
+  *     it returns a \a Real instead of a \a T.
   * \li A dummy_precision() function returning a weak epsilon value. It is mainly used as a default
   *     value by the fuzzy comparison operators.
   * \li highest() and lowest() functions returning the highest and lowest possible values respectively.
+  * \li digits10() function returning the number of decimal digits that can be represented without change. This is
+  *     the analogue of <a href="http://en.cppreference.com/w/cpp/types/numeric_limits/digits10">std::numeric_limits<T>::digits10</a>
+  *     which is used as the default implementation if specialized.
   */
 
 template<typename T> struct GenericNumTraits
@@ -67,22 +104,47 @@
                      T
                    >::type NonInteger;
   typedef T Nested;
+  typedef T Literal;
 
-  static inline Real epsilon() { return std::numeric_limits<T>::epsilon(); }
+  EIGEN_DEVICE_FUNC
+  static inline Real epsilon()
+  {
+    return numext::numeric_limits<T>::epsilon();
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline int digits10()
+  {
+    return internal::default_digits10_impl<T>::run();
+  }
+
+  EIGEN_DEVICE_FUNC
   static inline Real dummy_precision()
   {
     // make sure to override this for floating-point types
     return Real(0);
   }
-  static inline T highest() { return (std::numeric_limits<T>::max)(); }
-  static inline T lowest()  { return IsInteger ? (std::numeric_limits<T>::min)() : (-(std::numeric_limits<T>::max)()); }
-  
-#ifdef EIGEN2_SUPPORT
-  enum {
-    HasFloatingPoint = !IsInteger
-  };
-  typedef NonInteger FloatingPoint;
-#endif
+
+
+  EIGEN_DEVICE_FUNC
+  static inline T highest() {
+    return (numext::numeric_limits<T>::max)();
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline T lowest()  {
+    return IsInteger ? (numext::numeric_limits<T>::min)() : (-(numext::numeric_limits<T>::max)());
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline T infinity() {
+    return numext::numeric_limits<T>::infinity();
+  }
+
+  EIGEN_DEVICE_FUNC
+  static inline T quiet_NaN() {
+    return numext::numeric_limits<T>::quiet_NaN();
+  }
 };
 
 template<typename T> struct NumTraits : GenericNumTraits<T>
@@ -91,11 +153,13 @@
 template<> struct NumTraits<float>
   : GenericNumTraits<float>
 {
+  EIGEN_DEVICE_FUNC
   static inline float dummy_precision() { return 1e-5f; }
 };
 
 template<> struct NumTraits<double> : GenericNumTraits<double>
 {
+  EIGEN_DEVICE_FUNC
   static inline double dummy_precision() { return 1e-12; }
 };
 
@@ -109,6 +173,7 @@
   : GenericNumTraits<std::complex<_Real> >
 {
   typedef _Real Real;
+  typedef typename NumTraits<_Real>::Literal Literal;
   enum {
     IsComplex = 1,
     RequireInitialization = NumTraits<_Real>::RequireInitialization,
@@ -117,8 +182,12 @@
     MulCost = 4 * NumTraits<Real>::MulCost + 2 * NumTraits<Real>::AddCost
   };
 
+  EIGEN_DEVICE_FUNC
   static inline Real epsilon() { return NumTraits<Real>::epsilon(); }
+  EIGEN_DEVICE_FUNC
   static inline Real dummy_precision() { return NumTraits<Real>::dummy_precision(); }
+  EIGEN_DEVICE_FUNC
+  static inline int digits10() { return NumTraits<Real>::digits10(); }
 };
 
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
@@ -130,21 +199,50 @@
   typedef typename NumTraits<Scalar>::NonInteger NonIntegerScalar;
   typedef Array<NonIntegerScalar, Rows, Cols, Options, MaxRows, MaxCols> NonInteger;
   typedef ArrayType & Nested;
-  
+  typedef typename NumTraits<Scalar>::Literal Literal;
+
   enum {
     IsComplex = NumTraits<Scalar>::IsComplex,
     IsInteger = NumTraits<Scalar>::IsInteger,
     IsSigned  = NumTraits<Scalar>::IsSigned,
     RequireInitialization = 1,
-    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::ReadCost,
-    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::AddCost,
-    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::MulCost
+    ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::ReadCost,
+    AddCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::AddCost,
+    MulCost  = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * NumTraits<Scalar>::MulCost
   };
-  
+
+  EIGEN_DEVICE_FUNC
   static inline RealScalar epsilon() { return NumTraits<RealScalar>::epsilon(); }
+  EIGEN_DEVICE_FUNC
   static inline RealScalar dummy_precision() { return NumTraits<RealScalar>::dummy_precision(); }
+
+  static inline int digits10() { return NumTraits<Scalar>::digits10(); }
 };
 
+template<> struct NumTraits<std::string>
+  : GenericNumTraits<std::string>
+{
+  enum {
+    RequireInitialization = 1,
+    ReadCost = HugeCost,
+    AddCost  = HugeCost,
+    MulCost  = HugeCost
+  };
+
+  static inline int digits10() { return 0; }
+
+private:
+  static inline std::string epsilon();
+  static inline std::string dummy_precision();
+  static inline std::string lowest();
+  static inline std::string highest();
+  static inline std::string infinity();
+  static inline std::string quiet_NaN();
+};
+
+// Empty specialization for void to allow template specialization based on NumTraits<T>::Real with T==void and SFINAE.
+template<> struct NumTraits<void> {};
+
 } // end namespace Eigen
 
 #endif // EIGEN_NUMTRAITS_H
diff --git a/Eigen/src/Core/PermutationMatrix.h b/Eigen/src/Core/PermutationMatrix.h
index 85ffae2..b1fb455 100644
--- a/Eigen/src/Core/PermutationMatrix.h
+++ b/Eigen/src/Core/PermutationMatrix.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2009-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -13,14 +13,18 @@
 
 namespace Eigen { 
 
-template<int RowCol,typename IndicesType,typename MatrixType, typename StorageKind> class PermutedImpl;
+namespace internal {
+
+enum PermPermProduct_t {PermPermProduct};
+
+} // end namespace internal
 
 /** \class PermutationBase
   * \ingroup Core_Module
   *
   * \brief Base class for permutations
   *
-  * \param Derived the derived class
+  * \tparam Derived the derived class
   *
   * This class is the base class for all expressions representing a permutation matrix,
   * internally stored as a vector of integers.
@@ -38,17 +42,6 @@
   *
   * \sa class PermutationMatrix, class PermutationWrapper
   */
-
-namespace internal {
-
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed=false>
-struct permut_matrix_product_retval;
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed=false>
-struct permut_sparsematrix_product_retval;
-enum PermPermProduct_t {PermPermProduct};
-
-} // end namespace internal
-
 template<typename Derived>
 class PermutationBase : public EigenBase<Derived>
 {
@@ -60,19 +53,20 @@
     typedef typename Traits::IndicesType IndicesType;
     enum {
       Flags = Traits::Flags,
-      CoeffReadCost = Traits::CoeffReadCost,
       RowsAtCompileTime = Traits::RowsAtCompileTime,
       ColsAtCompileTime = Traits::ColsAtCompileTime,
       MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
     };
-    typedef typename Traits::Scalar Scalar;
-    typedef typename Traits::Index Index;
-    typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
+    typedef typename Traits::StorageIndex StorageIndex;
+    typedef Matrix<StorageIndex,RowsAtCompileTime,ColsAtCompileTime,0,MaxRowsAtCompileTime,MaxColsAtCompileTime>
             DenseMatrixType;
-    typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,Index>
+    typedef PermutationMatrix<IndicesType::SizeAtCompileTime,IndicesType::MaxSizeAtCompileTime,StorageIndex>
             PlainPermutationType;
+    typedef PlainPermutationType PlainObject;
     using Base::derived;
+    typedef Inverse<Derived> InverseReturnType;
+    typedef void Scalar;
     #endif
 
     /** Copies the other permutation into *this */
@@ -118,7 +112,7 @@
     void evalTo(MatrixBase<DenseDerived>& other) const
     {
       other.setZero();
-      for (int i=0; i<rows();++i)
+      for (Index i=0; i<rows(); ++i)
         other.coeffRef(indices().coeff(i),i) = typename DenseDerived::Scalar(1);
     }
     #endif
@@ -147,7 +141,8 @@
     /** Sets *this to be the identity permutation matrix */
     void setIdentity()
     {
-      for(Index i = 0; i < size(); ++i)
+      StorageIndex n = StorageIndex(size());
+      for(StorageIndex i = 0; i < n; ++i)
         indices().coeffRef(i) = i;
     }
 
@@ -163,18 +158,18 @@
       *
       * \returns a reference to *this.
       *
-      * \warning This is much slower than applyTranspositionOnTheRight(int,int):
+      * \warning This is much slower than applyTranspositionOnTheRight(Index,Index):
       * this has linear complexity and requires a lot of branching.
       *
-      * \sa applyTranspositionOnTheRight(int,int)
+      * \sa applyTranspositionOnTheRight(Index,Index)
       */
     Derived& applyTranspositionOnTheLeft(Index i, Index j)
     {
       eigen_assert(i>=0 && j>=0 && i<size() && j<size());
       for(Index k = 0; k < size(); ++k)
       {
-        if(indices().coeff(k) == i) indices().coeffRef(k) = j;
-        else if(indices().coeff(k) == j) indices().coeffRef(k) = i;
+        if(indices().coeff(k) == i) indices().coeffRef(k) = StorageIndex(j);
+        else if(indices().coeff(k) == j) indices().coeffRef(k) = StorageIndex(i);
       }
       return derived();
     }
@@ -185,7 +180,7 @@
       *
       * This is a fast operation, it only consists in swapping two indices.
       *
-      * \sa applyTranspositionOnTheLeft(int,int)
+      * \sa applyTranspositionOnTheLeft(Index,Index)
       */
     Derived& applyTranspositionOnTheRight(Index i, Index j)
     {
@@ -196,16 +191,16 @@
 
     /** \returns the inverse permutation matrix.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
-    inline Transpose<PermutationBase> inverse() const
-    { return derived(); }
+    inline InverseReturnType inverse() const
+    { return InverseReturnType(derived()); }
     /** \returns the tranpose permutation matrix.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
-    inline Transpose<PermutationBase> transpose() const
-    { return derived(); }
+    inline InverseReturnType transpose() const
+    { return InverseReturnType(derived()); }
 
     /**** multiplication helpers to hopefully get RVO ****/
 
@@ -215,13 +210,13 @@
     template<typename OtherDerived>
     void assignTranspose(const PermutationBase<OtherDerived>& other)
     {
-      for (int i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
+      for (Index i=0; i<rows();++i) indices().coeffRef(other.indices().coeff(i)) = i;
     }
     template<typename Lhs,typename Rhs>
     void assignProduct(const Lhs& lhs, const Rhs& rhs)
     {
       eigen_assert(lhs.cols() == rhs.rows());
-      for (int i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
+      for (Index i=0; i<rows();++i) indices().coeffRef(i) = lhs.indices().coeff(rhs.indices().coeff(i));
     }
 #endif
 
@@ -229,7 +224,7 @@
 
     /** \returns the product permutation matrix.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
     template<typename Other>
     inline PlainPermutationType operator*(const PermutationBase<Other>& other) const
@@ -237,18 +232,18 @@
 
     /** \returns the product of a permutation with another inverse permutation.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
     template<typename Other>
-    inline PlainPermutationType operator*(const Transpose<PermutationBase<Other> >& other) const
+    inline PlainPermutationType operator*(const InverseImpl<Other,PermutationStorage>& other) const
     { return PlainPermutationType(internal::PermPermProduct, *this, other.eval()); }
 
     /** \returns the product of an inverse permutation with another permutation.
       *
-      * \note \note_try_to_help_rvo
+      * \note \blank \note_try_to_help_rvo
       */
     template<typename Other> friend
-    inline PlainPermutationType operator*(const Transpose<PermutationBase<Other> >& other, const PermutationBase& perm)
+    inline PlainPermutationType operator*(const InverseImpl<Other, PermutationStorage>& other, const PermutationBase& perm)
     { return PlainPermutationType(internal::PermPermProduct, other.eval(), perm); }
     
     /** \returns the determinant of the permutation matrix, which is either 1 or -1 depending on the parity of the permutation.
@@ -284,39 +279,43 @@
 
 };
 
+namespace internal {
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
+ : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+{
+  typedef PermutationStorage StorageKind;
+  typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef void Scalar;
+};
+}
+
 /** \class PermutationMatrix
   * \ingroup Core_Module
   *
   * \brief Permutation matrix
   *
-  * \param SizeAtCompileTime the number of rows/cols, or Dynamic
-  * \param MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
-  * \param IndexType the interger type of the indices
+  * \tparam SizeAtCompileTime the number of rows/cols, or Dynamic
+  * \tparam MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
+  * \tparam _StorageIndex the integer type of the indices
   *
   * This class represents a permutation matrix, internally stored as a vector of integers.
   *
   * \sa class PermutationBase, class PermutationWrapper, class DiagonalMatrix
   */
-
-namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-struct traits<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType> >
- : traits<Matrix<IndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
-{
-  typedef IndexType Index;
-  typedef Matrix<IndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
-};
-}
-
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+class PermutationMatrix : public PermutationBase<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex> >
 {
     typedef PermutationBase<PermutationMatrix> Base;
     typedef internal::traits<PermutationMatrix> Traits;
   public:
 
+    typedef const PermutationMatrix& Nested;
+
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef typename Traits::IndicesType IndicesType;
+    typedef typename Traits::StorageIndex StorageIndex;
     #endif
 
     inline PermutationMatrix()
@@ -324,8 +323,10 @@
 
     /** Constructs an uninitialized permutation matrix of given size.
       */
-    inline PermutationMatrix(int size) : m_indices(size)
-    {}
+    explicit inline PermutationMatrix(Index size) : m_indices(size)
+    {
+      eigen_internal_assert(size <= NumTraits<StorageIndex>::highest());
+    }
 
     /** Copy constructor. */
     template<typename OtherDerived>
@@ -346,7 +347,7 @@
       * array's size.
       */
     template<typename Other>
-    explicit inline PermutationMatrix(const MatrixBase<Other>& a_indices) : m_indices(a_indices)
+    explicit inline PermutationMatrix(const MatrixBase<Other>& indices) : m_indices(indices)
     {}
 
     /** Convert the Transpositions \a tr to a permutation matrix */
@@ -393,10 +394,13 @@
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename Other>
-    PermutationMatrix(const Transpose<PermutationBase<Other> >& other)
-      : m_indices(other.nestedPermutation().size())
+    PermutationMatrix(const InverseImpl<Other,PermutationStorage>& other)
+      : m_indices(other.derived().nestedExpression().size())
     {
-      for (int i=0; i<m_indices.size();++i) m_indices.coeffRef(other.nestedPermutation().indices().coeff(i)) = i;
+      eigen_internal_assert(m_indices.size() <= NumTraits<StorageIndex>::highest());
+      StorageIndex end = StorageIndex(m_indices.size());
+      for (StorageIndex i=0; i<end;++i)
+        m_indices.coeffRef(other.derived().nestedExpression().indices().coeff(i)) = i;
     }
     template<typename Lhs,typename Rhs>
     PermutationMatrix(internal::PermPermProduct_t, const Lhs& lhs, const Rhs& rhs)
@@ -413,18 +417,20 @@
 
 
 namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
-struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess> >
- : traits<Matrix<IndexType,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+struct traits<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
+ : traits<Matrix<_StorageIndex,SizeAtCompileTime,SizeAtCompileTime,0,MaxSizeAtCompileTime,MaxSizeAtCompileTime> >
 {
-  typedef IndexType Index;
-  typedef Map<const Matrix<IndexType, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
+  typedef PermutationStorage StorageKind;
+  typedef Map<const Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1>, _PacketAccess> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef void Scalar;
 };
 }
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
-class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess>
-  : public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, IndexType>,_PacketAccess> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+class Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess>
+  : public PermutationBase<Map<PermutationMatrix<SizeAtCompileTime, MaxSizeAtCompileTime, _StorageIndex>,_PacketAccess> >
 {
     typedef PermutationBase<Map> Base;
     typedef internal::traits<Map> Traits;
@@ -432,14 +438,14 @@
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
     #endif
 
-    inline Map(const Index* indicesPtr)
+    inline Map(const StorageIndex* indicesPtr)
       : m_indices(indicesPtr)
     {}
 
-    inline Map(const Index* indicesPtr, Index size)
+    inline Map(const StorageIndex* indicesPtr, Index size)
       : m_indices(indicesPtr,size)
     {}
 
@@ -474,40 +480,36 @@
     IndicesType m_indices;
 };
 
-/** \class PermutationWrapper
-  * \ingroup Core_Module
-  *
-  * \brief Class to view a vector of integers as a permutation matrix
-  *
-  * \param _IndicesType the type of the vector of integer (can be any compatible expression)
-  *
-  * This class allows to view any vector expression of integers as a permutation matrix.
-  *
-  * \sa class PermutationBase, class PermutationMatrix
-  */
-
-struct PermutationStorage {};
-
 template<typename _IndicesType> class TranspositionsWrapper;
 namespace internal {
 template<typename _IndicesType>
 struct traits<PermutationWrapper<_IndicesType> >
 {
   typedef PermutationStorage StorageKind;
-  typedef typename _IndicesType::Scalar Scalar;
-  typedef typename _IndicesType::Scalar Index;
+  typedef void Scalar;
+  typedef typename _IndicesType::Scalar StorageIndex;
   typedef _IndicesType IndicesType;
   enum {
     RowsAtCompileTime = _IndicesType::SizeAtCompileTime,
     ColsAtCompileTime = _IndicesType::SizeAtCompileTime,
-    MaxRowsAtCompileTime = IndicesType::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = IndicesType::MaxColsAtCompileTime,
-    Flags = 0,
-    CoeffReadCost = _IndicesType::CoeffReadCost
+    MaxRowsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
+    MaxColsAtCompileTime = IndicesType::MaxSizeAtCompileTime,
+    Flags = 0
   };
 };
 }
 
+/** \class PermutationWrapper
+  * \ingroup Core_Module
+  *
+  * \brief Class to view a vector of integers as a permutation matrix
+  *
+  * \tparam _IndicesType the type of the vector of integer (can be any compatible expression)
+  *
+  * This class allows to view any vector expression of integers as a permutation matrix.
+  *
+  * \sa class PermutationBase, class PermutationMatrix
+  */
 template<typename _IndicesType>
 class PermutationWrapper : public PermutationBase<PermutationWrapper<_IndicesType> >
 {
@@ -519,8 +521,8 @@
     typedef typename Traits::IndicesType IndicesType;
     #endif
 
-    inline PermutationWrapper(const IndicesType& a_indices)
-      : m_indices(a_indices)
+    inline PermutationWrapper(const IndicesType& indices)
+      : m_indices(indices)
     {}
 
     /** const version of indices(). */
@@ -532,182 +534,86 @@
     typename IndicesType::Nested m_indices;
 };
 
+
 /** \returns the matrix with the permutation applied to the columns.
   */
-template<typename Derived, typename PermutationDerived>
-inline const internal::permut_matrix_product_retval<PermutationDerived, Derived, OnTheRight>
-operator*(const MatrixBase<Derived>& matrix,
-          const PermutationBase<PermutationDerived> &permutation)
+template<typename MatrixDerived, typename PermutationDerived>
+EIGEN_DEVICE_FUNC
+const Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
+operator*(const MatrixBase<MatrixDerived> &matrix,
+          const PermutationBase<PermutationDerived>& permutation)
 {
-  return internal::permut_matrix_product_retval
-           <PermutationDerived, Derived, OnTheRight>
-           (permutation.derived(), matrix.derived());
+  return Product<MatrixDerived, PermutationDerived, AliasFreeProduct>
+            (matrix.derived(), permutation.derived());
 }
 
 /** \returns the matrix with the permutation applied to the rows.
   */
-template<typename Derived, typename PermutationDerived>
-inline const internal::permut_matrix_product_retval
-               <PermutationDerived, Derived, OnTheLeft>
+template<typename PermutationDerived, typename MatrixDerived>
+EIGEN_DEVICE_FUNC
+const Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
 operator*(const PermutationBase<PermutationDerived> &permutation,
-          const MatrixBase<Derived>& matrix)
+          const MatrixBase<MatrixDerived>& matrix)
 {
-  return internal::permut_matrix_product_retval
-           <PermutationDerived, Derived, OnTheLeft>
-           (permutation.derived(), matrix.derived());
+  return Product<PermutationDerived, MatrixDerived, AliasFreeProduct>
+            (permutation.derived(), matrix.derived());
 }
 
-namespace internal {
 
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct traits<permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
+template<typename PermutationType>
+class InverseImpl<PermutationType, PermutationStorage>
+  : public EigenBase<Inverse<PermutationType> >
 {
-  typedef typename MatrixType::PlainObject ReturnType;
-};
-
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct permut_matrix_product_retval
- : public ReturnByValue<permut_matrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
-{
-    typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-    typedef typename MatrixType::Index Index;
-
-    permut_matrix_product_retval(const PermutationType& perm, const MatrixType& matrix)
-      : m_permutation(perm), m_matrix(matrix)
-    {}
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-
-    template<typename Dest> inline void evalTo(Dest& dst) const
-    {
-      const Index n = Side==OnTheLeft ? rows() : cols();
-      // FIXME we need an is_same for expression that is not sensitive to constness. For instance
-      // is_same_xpr<Block<const Matrix>, Block<Matrix> >::value should be true.
-      if(    is_same<MatrixTypeNestedCleaned,Dest>::value
-          && blas_traits<MatrixTypeNestedCleaned>::HasUsableDirectAccess
-          && blas_traits<Dest>::HasUsableDirectAccess
-          && extract_data(dst) == extract_data(m_matrix))
-      {
-        // apply the permutation inplace
-        Matrix<bool,PermutationType::RowsAtCompileTime,1,0,PermutationType::MaxRowsAtCompileTime> mask(m_permutation.size());
-        mask.fill(false);
-        Index r = 0;
-        while(r < m_permutation.size())
-        {
-          // search for the next seed
-          while(r<m_permutation.size() && mask[r]) r++;
-          if(r>=m_permutation.size())
-            break;
-          // we got one, let's follow it until we are back to the seed
-          Index k0 = r++;
-          Index kPrev = k0;
-          mask.coeffRef(k0) = true;
-          for(Index k=m_permutation.indices().coeff(k0); k!=k0; k=m_permutation.indices().coeff(k))
-          {
-                  Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>(dst, k)
-            .swap(Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
-                       (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev));
-
-            mask.coeffRef(k) = true;
-            kPrev = k;
-          }
-        }
-      }
-      else
-      {
-        for(int i = 0; i < n; ++i)
-        {
-          Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
-               (dst, ((Side==OnTheLeft) ^ Transposed) ? m_permutation.indices().coeff(i) : i)
-
-          =
-
-          Block<const MatrixTypeNestedCleaned,Side==OnTheLeft ? 1 : MatrixType::RowsAtCompileTime,Side==OnTheRight ? 1 : MatrixType::ColsAtCompileTime>
-               (m_matrix, ((Side==OnTheRight) ^ Transposed) ? m_permutation.indices().coeff(i) : i);
-        }
-      }
-    }
-
-  protected:
-    const PermutationType& m_permutation;
-    typename MatrixType::Nested m_matrix;
-};
-
-/* Template partial specialization for transposed/inverse permutations */
-
-template<typename Derived>
-struct traits<Transpose<PermutationBase<Derived> > >
- : traits<Derived>
-{};
-
-} // end namespace internal
-
-template<typename Derived>
-class Transpose<PermutationBase<Derived> >
-  : public EigenBase<Transpose<PermutationBase<Derived> > >
-{
-    typedef Derived PermutationType;
-    typedef typename PermutationType::IndicesType IndicesType;
     typedef typename PermutationType::PlainPermutationType PlainPermutationType;
+    typedef internal::traits<PermutationType> PermTraits;
+  protected:
+    InverseImpl() {}
   public:
+    typedef Inverse<PermutationType> InverseType;
+    using EigenBase<Inverse<PermutationType> >::derived;
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
-    typedef internal::traits<PermutationType> Traits;
-    typedef typename Derived::DenseMatrixType DenseMatrixType;
+    typedef typename PermutationType::DenseMatrixType DenseMatrixType;
     enum {
-      Flags = Traits::Flags,
-      CoeffReadCost = Traits::CoeffReadCost,
-      RowsAtCompileTime = Traits::RowsAtCompileTime,
-      ColsAtCompileTime = Traits::ColsAtCompileTime,
-      MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = Traits::MaxColsAtCompileTime
+      RowsAtCompileTime = PermTraits::RowsAtCompileTime,
+      ColsAtCompileTime = PermTraits::ColsAtCompileTime,
+      MaxRowsAtCompileTime = PermTraits::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = PermTraits::MaxColsAtCompileTime
     };
-    typedef typename Traits::Scalar Scalar;
     #endif
 
-    Transpose(const PermutationType& p) : m_permutation(p) {}
-
-    inline int rows() const { return m_permutation.rows(); }
-    inline int cols() const { return m_permutation.cols(); }
-
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename DenseDerived>
     void evalTo(MatrixBase<DenseDerived>& other) const
     {
       other.setZero();
-      for (int i=0; i<rows();++i)
-        other.coeffRef(i, m_permutation.indices().coeff(i)) = typename DenseDerived::Scalar(1);
+      for (Index i=0; i<derived().rows();++i)
+        other.coeffRef(i, derived().nestedExpression().indices().coeff(i)) = typename DenseDerived::Scalar(1);
     }
     #endif
 
     /** \return the equivalent permutation matrix */
-    PlainPermutationType eval() const { return *this; }
+    PlainPermutationType eval() const { return derived(); }
 
-    DenseMatrixType toDenseMatrix() const { return *this; }
+    DenseMatrixType toDenseMatrix() const { return derived(); }
 
     /** \returns the matrix with the inverse permutation applied to the columns.
       */
     template<typename OtherDerived> friend
-    inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>
-    operator*(const MatrixBase<OtherDerived>& matrix, const Transpose& trPerm)
+    const Product<OtherDerived, InverseType, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix, const InverseType& trPerm)
     {
-      return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheRight, true>(trPerm.m_permutation, matrix.derived());
+      return Product<OtherDerived, InverseType, AliasFreeProduct>(matrix.derived(), trPerm.derived());
     }
 
     /** \returns the matrix with the inverse permutation applied to the rows.
       */
     template<typename OtherDerived>
-    inline const internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>
+    const Product<InverseType, OtherDerived, AliasFreeProduct>
     operator*(const MatrixBase<OtherDerived>& matrix) const
     {
-      return internal::permut_matrix_product_retval<PermutationType, OtherDerived, OnTheLeft, true>(m_permutation, matrix.derived());
+      return Product<InverseType, OtherDerived, AliasFreeProduct>(derived(), matrix.derived());
     }
-
-    const PermutationType& nestedPermutation() const { return m_permutation; }
-
-  protected:
-    const PermutationType& m_permutation;
 };
 
 template<typename Derived>
@@ -716,6 +622,12 @@
   return derived();
 }
 
+namespace internal {
+
+template<> struct AssignmentKind<DenseShape,PermutationShape> { typedef EigenBase2EigenBase Kind; };
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_PERMUTATIONMATRIX_H
diff --git a/Eigen/src/Core/PlainObjectBase.h b/Eigen/src/Core/PlainObjectBase.h
index 8a3e454..1dc7e22 100644
--- a/Eigen/src/Core/PlainObjectBase.h
+++ b/Eigen/src/Core/PlainObjectBase.h
@@ -28,6 +28,7 @@
 
 template<int MaxSizeAtCompileTime> struct check_rows_cols_for_overflow {
   template<typename Index>
+  EIGEN_DEVICE_FUNC
   static EIGEN_ALWAYS_INLINE void run(Index, Index)
   {
   }
@@ -35,11 +36,12 @@
 
 template<> struct check_rows_cols_for_overflow<Dynamic> {
   template<typename Index>
+  EIGEN_DEVICE_FUNC
   static EIGEN_ALWAYS_INLINE void run(Index rows, Index cols)
   {
     // http://hg.mozilla.org/mozilla-central/file/6c8a909977d3/xpcom/ds/CheckedInt.h#l242
     // we assume Index is signed
-    Index max_index = (size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
+    Index max_index = (std::size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed
     bool error = (rows == 0 || cols == 0) ? false
                : (rows > max_index / cols);
     if (error)
@@ -56,33 +58,41 @@
 
 } // end namespace internal
 
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+namespace doxygen {
+
+// This is a workaround to doxygen not being able to understand the inheritance logic
+// when it is hidden by the dense_xpr_base helper struct.
+// Moreover, doxygen fails to include members that are not documented in the declaration body of
+// MatrixBase if we inherits MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >,
+// this is why we simply inherits MatrixBase, though this does not make sense.
+
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename Derived> struct dense_xpr_base_dispatcher;
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct dense_xpr_base_dispatcher<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+    : public MatrixBase {};
+/** This class is just a workaround for Doxygen and it does not not actually exist. */
+template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
+struct dense_xpr_base_dispatcher<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
+    : public ArrayBase {};
+
+} // namespace doxygen
+
 /** \class PlainObjectBase
+  * \ingroup Core_Module
   * \brief %Dense storage base class for matrices and arrays.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN.
+  *
+  * \tparam Derived is the derived type, e.g., a Matrix or Array
   *
   * \sa \ref TopicClassHierarchy
   */
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-namespace internal {
-
-// this is a warkaround to doxygen not being able to understand the inheritence logic
-// when it is hidden by the dense_xpr_base helper struct.
-template<typename Derived> struct dense_xpr_base_dispatcher_for_doxygen;// : public MatrixBase<Derived> {};
-/** This class is just a workaround for Doxygen and it does not not actually exist. */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct dense_xpr_base_dispatcher_for_doxygen<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-    : public MatrixBase<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
-/** This class is just a workaround for Doxygen and it does not not actually exist. */
-template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
-struct dense_xpr_base_dispatcher_for_doxygen<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> >
-    : public ArrayBase<Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> > {};
-
-} // namespace internal
-
 template<typename Derived>
-class PlainObjectBase : public internal::dense_xpr_base_dispatcher_for_doxygen<Derived>
+class PlainObjectBase : public doxygen::dense_xpr_base_dispatcher<Derived>
 #else
 template<typename Derived>
 class PlainObjectBase : public internal::dense_xpr_base<Derived>::type
@@ -93,8 +103,8 @@
     typedef typename internal::dense_xpr_base<Derived>::type Base;
 
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
     typedef typename internal::traits<Derived>::Scalar Scalar;
+    
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
     typedef Derived DenseType;
@@ -113,28 +123,40 @@
     typedef Eigen::Map<Derived, Unaligned>  MapType;
     friend  class Eigen::Map<const Derived, Unaligned>;
     typedef const Eigen::Map<const Derived, Unaligned> ConstMapType;
-    friend  class Eigen::Map<Derived, Aligned>;
-    typedef Eigen::Map<Derived, Aligned> AlignedMapType;
-    friend  class Eigen::Map<const Derived, Aligned>;
-    typedef const Eigen::Map<const Derived, Aligned> ConstAlignedMapType;
+#if EIGEN_MAX_ALIGN_BYTES>0
+    // for EIGEN_MAX_ALIGN_BYTES==0, AlignedMax==Unaligned, and many compilers generate warnings for friend-ing a class twice.
+    friend  class Eigen::Map<Derived, AlignedMax>;
+    friend  class Eigen::Map<const Derived, AlignedMax>;
+#endif
+    typedef Eigen::Map<Derived, AlignedMax> AlignedMapType;
+    typedef const Eigen::Map<const Derived, AlignedMax> ConstAlignedMapType;
     template<typename StrideType> struct StridedMapType { typedef Eigen::Map<Derived, Unaligned, StrideType> type; };
     template<typename StrideType> struct StridedConstMapType { typedef Eigen::Map<const Derived, Unaligned, StrideType> type; };
-    template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, Aligned, StrideType> type; };
-    template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, Aligned, StrideType> type; };
+    template<typename StrideType> struct StridedAlignedMapType { typedef Eigen::Map<Derived, AlignedMax, StrideType> type; };
+    template<typename StrideType> struct StridedConstAlignedMapType { typedef Eigen::Map<const Derived, AlignedMax, StrideType> type; };
 
   protected:
     DenseStorage<Scalar, Base::MaxSizeAtCompileTime, Base::RowsAtCompileTime, Base::ColsAtCompileTime, Options> m_storage;
 
   public:
-    enum { NeedsToAlign = SizeAtCompileTime != Dynamic && (internal::traits<Derived>::Flags & AlignedBit) != 0 };
+    enum { NeedsToAlign = (SizeAtCompileTime != Dynamic) && (internal::traits<Derived>::Alignment>0) };
     EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign)
 
+    EIGEN_DEVICE_FUNC
     Base& base() { return *static_cast<Base*>(this); }
+    EIGEN_DEVICE_FUNC
     const Base& base() const { return *static_cast<const Base*>(this); }
 
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index rows() const { return m_storage.rows(); }
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Index cols() const { return m_storage.cols(); }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index,Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeff(Index rowId, Index colId) const
     {
       if(Flags & RowMajorBit)
@@ -143,11 +165,21 @@
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,ReadOnlyAccessors>::coeff(Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeff(Index index) const
     {
       return m_storage.data()[index];
     }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index,Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index,Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index rowId, Index colId)
     {
       if(Flags & RowMajorBit)
@@ -156,11 +188,19 @@
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
+    /** This is an overloaded version of DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index) const
+      * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts.
+      *
+      * See DenseCoeffsBase<Derived,WriteAccessors>::coeffRef(Index) const for details. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Scalar& coeffRef(Index index)
     {
       return m_storage.data()[index];
     }
 
+    /** This is the const version of coeffRef(Index,Index) which is thus synonym of coeff(Index,Index).
+      * It is provided for convenience. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeffRef(Index rowId, Index colId) const
     {
       if(Flags & RowMajorBit)
@@ -169,6 +209,9 @@
         return m_storage.data()[rowId + colId * m_storage.rows()];
     }
 
+    /** This is the const version of coeffRef(Index) which is thus synonym of coeff(Index).
+      * It is provided for convenience. */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE const Scalar& coeffRef(Index index) const
     {
       return m_storage.data()[index];
@@ -209,11 +252,11 @@
     }
 
     /** \returns a const pointer to the data array of this matrix */
-    EIGEN_STRONG_INLINE const Scalar *data() const
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar *data() const
     { return m_storage.data(); }
 
     /** \returns a pointer to the data array of this matrix */
-    EIGEN_STRONG_INLINE Scalar *data()
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar *data()
     { return m_storage.data(); }
 
     /** Resizes \c *this to a \a rows x \a cols matrix.
@@ -232,22 +275,22 @@
       *
       * \sa resize(Index) for vectors, resize(NoChange_t, Index), resize(Index, NoChange_t)
       */
-    EIGEN_STRONG_INLINE void resize(Index nbRows, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void resize(Index rows, Index cols)
     {
-      eigen_assert(   EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,nbRows==RowsAtCompileTime)
-                   && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,nbCols==ColsAtCompileTime)
-                   && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,nbRows<=MaxRowsAtCompileTime)
-                   && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,nbCols<=MaxColsAtCompileTime)
-                   && nbRows>=0 && nbCols>=0 && "Invalid sizes when resizing a matrix or array.");
-      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(nbRows, nbCols);
+      eigen_assert(   EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,rows==RowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,cols==ColsAtCompileTime)
+                   && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,rows<=MaxRowsAtCompileTime)
+                   && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,cols<=MaxColsAtCompileTime)
+                   && rows>=0 && cols>=0 && "Invalid sizes when resizing a matrix or array.");
+      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(rows, cols);
       #ifdef EIGEN_INITIALIZE_COEFFS
-        Index size = nbRows*nbCols;
+        Index size = rows*cols;
         bool size_changed = size != this->size();
-        m_storage.resize(size, nbRows, nbCols);
+        m_storage.resize(size, rows, cols);
         if(size_changed) EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
       #else
-        internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(nbRows, nbCols);
-        m_storage.resize(nbRows*nbCols, nbRows, nbCols);
+        m_storage.resize(rows*cols, rows, cols);
       #endif
     }
 
@@ -262,6 +305,7 @@
       *
       * \sa resize(Index,Index), resize(NoChange_t, Index), resize(Index, NoChange_t)
       */
+    EIGEN_DEVICE_FUNC
     inline void resize(Index size)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(PlainObjectBase)
@@ -286,9 +330,10 @@
       *
       * \sa resize(Index,Index)
       */
-    inline void resize(NoChange_t, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    inline void resize(NoChange_t, Index cols)
     {
-      resize(rows(), nbCols);
+      resize(rows(), cols);
     }
 
     /** Resizes the matrix, changing only the number of rows. For the parameter of type NoChange_t, just pass the special value \c NoChange
@@ -299,9 +344,10 @@
       *
       * \sa resize(Index,Index)
       */
-    inline void resize(Index nbRows, NoChange_t)
+    EIGEN_DEVICE_FUNC
+    inline void resize(Index rows, NoChange_t)
     {
-      resize(nbRows, cols());
+      resize(rows, cols());
     }
 
     /** Resizes \c *this to have the same dimensions as \a other.
@@ -312,6 +358,7 @@
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE void resizeLike(const EigenBase<OtherDerived>& _other)
     {
       const OtherDerived& other = _other.derived();
@@ -339,9 +386,10 @@
       * Matrices are resized relative to the top-left element. In case values need to be 
       * appended to the matrix they will be uninitialized.
       */
-    EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(Index rows, Index cols)
     {
-      internal::conservative_resize_like_impl<Derived>::run(*this, nbRows, nbCols);
+      internal::conservative_resize_like_impl<Derived>::run(*this, rows, cols);
     }
 
     /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
@@ -351,10 +399,11 @@
       *
       * In case the matrix is growing, new rows will be uninitialized.
       */
-    EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, NoChange_t)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(Index rows, NoChange_t)
     {
       // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(nbRows, cols());
+      conservativeResize(rows, cols());
     }
 
     /** Resizes the matrix to \a rows x \a cols while leaving old values untouched.
@@ -364,10 +413,11 @@
       *
       * In case the matrix is growing, new columns will be uninitialized.
       */
-    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index nbCols)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index cols)
     {
       // Note: see the comment in conservativeResize(Index,Index)
-      conservativeResize(rows(), nbCols);
+      conservativeResize(rows(), cols);
     }
 
     /** Resizes the vector to \a size while retaining old values.
@@ -378,6 +428,7 @@
       *
       * When values are appended, they will be uninitialized.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void conservativeResize(Index size)
     {
       internal::conservative_resize_like_impl<Derived>::run(*this, size);
@@ -393,6 +444,7 @@
       * appended to the matrix they will copied from \c other.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void conservativeResizeLike(const DenseBase<OtherDerived>& other)
     {
       internal::conservative_resize_like_impl<Derived,OtherDerived>::run(*this, other);
@@ -401,6 +453,7 @@
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
       */
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& operator=(const PlainObjectBase& other)
     {
       return _set(other);
@@ -408,6 +461,7 @@
 
     /** \sa MatrixBase::lazyAssign() */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& lazyAssign(const DenseBase<OtherDerived>& other)
     {
       _resize_to_match(other);
@@ -415,12 +469,18 @@
     }
 
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE Derived& operator=(const ReturnByValue<OtherDerived>& func)
     {
       resize(func.rows(), func.cols());
       return Base::operator=(func);
     }
 
+    // Prevent user from trying to instantiate PlainObjectBase objects
+    // by making all its constructor protected. See bug 1074.
+  protected:
+
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE PlainObjectBase() : m_storage()
     {
 //       _check_template_params();
@@ -430,23 +490,81 @@
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     // FIXME is it still needed ?
     /** \internal */
-    PlainObjectBase(internal::constructor_without_unaligned_array_assert)
+    EIGEN_DEVICE_FUNC
+    explicit PlainObjectBase(internal::constructor_without_unaligned_array_assert)
       : m_storage(internal::constructor_without_unaligned_array_assert())
     {
 //       _check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 #endif
 
-    EIGEN_STRONG_INLINE PlainObjectBase(Index a_size, Index nbRows, Index nbCols)
-      : m_storage(a_size, nbRows, nbCols)
+#if EIGEN_HAS_RVALUE_REFERENCES
+    EIGEN_DEVICE_FUNC
+    PlainObjectBase(PlainObjectBase&& other) EIGEN_NOEXCEPT
+      : m_storage( std::move(other.m_storage) )
+    {
+    }
+
+    EIGEN_DEVICE_FUNC
+    PlainObjectBase& operator=(PlainObjectBase&& other) EIGEN_NOEXCEPT
+    {
+      using std::swap;
+      swap(m_storage, other.m_storage);
+      return *this;
+    }
+#endif
+
+    /** Copy constructor */
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const PlainObjectBase& other)
+      : Base(), m_storage(other.m_storage) { }
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(Index size, Index rows, Index cols)
+      : m_storage(size, rows, cols)
     {
 //       _check_template_params();
 //       EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED
     }
 
-    /** \copydoc MatrixBase::operator=(const EigenBase<OtherDerived>&)
+    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const DenseBase<OtherDerived> &other)
+      : m_storage()
+    {
+      _check_template_params();
+      resizeLike(other);
+      _set_noalias(other);
+    }
+
+    /** \sa PlainObjectBase::operator=(const EigenBase<OtherDerived>&) */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase<OtherDerived> &other)
+      : m_storage()
+    {
+      _check_template_params();
+      resizeLike(other);
+      *this = other.derived();
+    }
+    /** \brief Copy constructor with in-place evaluation */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE PlainObjectBase(const ReturnByValue<OtherDerived>& other)
+    {
+      _check_template_params();
+      // FIXME this does not automatically transpose vectors if necessary
+      resize(other.rows(), other.cols());
+      other.evalTo(this->derived());
+    }
+
+  public:
+
+    /** \brief Copies the generic expression \a other into *this.
+      * \copydetails DenseBase::operator=(const EigenBase<OtherDerived> &other)
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& operator=(const EigenBase<OtherDerived> &other)
     {
       _resize_to_match(other);
@@ -454,21 +572,15 @@
       return this->derived();
     }
 
-    /** \sa MatrixBase::operator=(const EigenBase<OtherDerived>&) */
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase<OtherDerived> &other)
-      : m_storage(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols())
-    {
-      _check_template_params();
-      internal::check_rows_cols_for_overflow<MaxSizeAtCompileTime>::run(other.derived().rows(), other.derived().cols());
-      Base::operator=(other.derived());
-    }
-
     /** \name Map
       * These are convenience functions returning Map objects. The Map() static functions return unaligned Map objects,
       * while the AlignedMap() functions return aligned Map objects and thus should be called only with 16-byte-aligned
       * \a data pointers.
       *
+      * Here is an example using strides:
+      * \include Matrix_Map_stride.cpp
+      * Output: \verbinclude Matrix_Map_stride.out
+      *
       * \see class Map
       */
     //@{
@@ -538,16 +650,16 @@
     //@}
 
     using Base::setConstant;
-    Derived& setConstant(Index size, const Scalar& value);
-    Derived& setConstant(Index rows, Index cols, const Scalar& value);
+    EIGEN_DEVICE_FUNC Derived& setConstant(Index size, const Scalar& val);
+    EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, Index cols, const Scalar& val);
 
     using Base::setZero;
-    Derived& setZero(Index size);
-    Derived& setZero(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC Derived& setZero(Index size);
+    EIGEN_DEVICE_FUNC Derived& setZero(Index rows, Index cols);
 
     using Base::setOnes;
-    Derived& setOnes(Index size);
-    Derived& setOnes(Index rows, Index cols);
+    EIGEN_DEVICE_FUNC Derived& setOnes(Index size);
+    EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, Index cols);
 
     using Base::setRandom;
     Derived& setRandom(Index size);
@@ -566,6 +678,7 @@
       * remain row-vectors and vectors remain vectors.
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE void _resize_to_match(const EigenBase<OtherDerived>& other)
     {
       #ifdef EIGEN_NO_AUTOMATIC_RESIZING
@@ -573,8 +686,6 @@
                  : (rows() == other.rows() && cols() == other.cols())))
         && "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined");
       EIGEN_ONLY_USED_FOR_DEBUG(other);
-      if(this->size()==0)
-        resizeLike(other);
       #else
       resizeLike(other);
       #endif
@@ -594,25 +705,23 @@
       *
       * \internal
       */
+    // aliasing is dealt once in internall::call_assignment
+    // so at this stage we have to assume aliasing... and resising has to be done later.
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& _set(const DenseBase<OtherDerived>& other)
     {
-      _set_selector(other.derived(), typename internal::conditional<static_cast<bool>(int(OtherDerived::Flags) & EvalBeforeAssigningBit), internal::true_type, internal::false_type>::type());
+      internal::call_assignment(this->derived(), other.derived());
       return this->derived();
     }
 
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::true_type&) { _set_noalias(other.eval()); }
-
-    template<typename OtherDerived>
-    EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::false_type&) { _set_noalias(other); }
-
     /** \internal Like _set() but additionally makes the assumption that no aliasing effect can happen (which
       * is the case when creating a new matrix) so one can enforce lazy evaluation.
       *
       * \sa operator=(const MatrixBase<OtherDerived>&), _set()
       */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC 
     EIGEN_STRONG_INLINE Derived& _set_noalias(const DenseBase<OtherDerived>& other)
     {
       // I don't think we need this resize call since the lazyAssign will anyways resize
@@ -620,40 +729,175 @@
       //_resize_to_match(other);
       // the 'false' below means to enforce lazy evaluation. We don't use lazyAssign() because
       // it wouldn't allow to copy a row-vector into a column-vector.
-      return internal::assign_selector<Derived,OtherDerived,false>::run(this->derived(), other.derived());
+      internal::call_assignment_no_alias(this->derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+      return this->derived();
     }
 
     template<typename T0, typename T1>
-    EIGEN_STRONG_INLINE void _init2(Index nbRows, Index nbCols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename internal::enable_if<Base::SizeAtCompileTime!=2,T0>::type* = 0)
     {
       EIGEN_STATIC_ASSERT(bool(NumTraits<T0>::IsInteger) &&
                           bool(NumTraits<T1>::IsInteger),
                           FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
-      resize(nbRows,nbCols);
+      resize(rows,cols);
     }
+    
     template<typename T0, typename T1>
-    EIGEN_STRONG_INLINE void _init2(const Scalar& val0, const Scalar& val1, typename internal::enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE void _init2(const T0& val0, const T1& val1, typename internal::enable_if<Base::SizeAtCompileTime==2,T0>::type* = 0)
     {
       EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
-      m_storage.data()[0] = val0;
-      m_storage.data()[1] = val1;
+      m_storage.data()[0] = Scalar(val0);
+      m_storage.data()[1] = Scalar(val1);
+    }
+    
+    template<typename T0, typename T1>
+    EIGEN_DEVICE_FUNC 
+    EIGEN_STRONG_INLINE void _init2(const Index& val0, const Index& val1,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<T0,Index>::value)
+                                                                  && (internal::is_same<T1,Index>::value)
+                                                                  && Base::SizeAtCompileTime==2,T1>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2)
+      m_storage.data()[0] = Scalar(val0);
+      m_storage.data()[1] = Scalar(val1);
     }
 
+    // The argument is convertible to the Index type and we either have a non 1x1 Matrix, or a dynamic-sized Array,
+    // then the argument is meant to be the size of the object.
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(Index size, typename internal::enable_if<    (Base::SizeAtCompileTime!=1 || !internal::is_convertible<T, Scalar>::value)
+                                                                              && ((!internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value || Base::SizeAtCompileTime==Dynamic)),T>::type* = 0)
+    {
+      // NOTE MSVC 2008 complains if we directly put bool(NumTraits<T>::IsInteger) as the EIGEN_STATIC_ASSERT argument.
+      const bool is_integer = NumTraits<T>::IsInteger;
+      EIGEN_UNUSED_VARIABLE(is_integer);
+      EIGEN_STATIC_ASSERT(is_integer,
+                          FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED)
+      resize(size);
+    }
+    
+    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitely converted)
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar& val0, typename internal::enable_if<Base::SizeAtCompileTime==1 && internal::is_convertible<T, Scalar>::value,T>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
+      m_storage.data()[0] = val0;
+    }
+    
+    // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type match the index type)
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Index& val0,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<Index,T>::value)
+                                                                  && Base::SizeAtCompileTime==1
+                                                                  && internal::is_convertible<T, Scalar>::value,T*>::type* = 0)
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1)
+      m_storage.data()[0] = Scalar(val0);
+    }
+
+    // Initialize a fixed size matrix from a pointer to raw data
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar* data){
+      this->_set_noalias(ConstMapType(data));
+    }
+
+    // Initialize an arbitrary matrix from a dense expression
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const DenseBase<OtherDerived>& other){
+      this->_set_noalias(other);
+    }
+
+    // Initialize an arbitrary matrix from an object convertible to the Derived type.
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Derived& other){
+      this->_set_noalias(other);
+    }
+
+    // Initialize an arbitrary matrix from a generic Eigen expression
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const EigenBase<OtherDerived>& other){
+      this->derived() = other;
+    }
+
+    template<typename T, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const ReturnByValue<OtherDerived>& other)
+    {
+      resize(other.rows(), other.cols());
+      other.evalTo(this->derived());
+    }
+
+    template<typename T, typename OtherDerived, int ColsAtCompileTime>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
+    {
+      this->derived() = r;
+    }
+    
+    // For fixed-size Array<Scalar,...>
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Scalar& val0,
+                                    typename internal::enable_if<    Base::SizeAtCompileTime!=Dynamic
+                                                                  && Base::SizeAtCompileTime!=1
+                                                                  && internal::is_convertible<T, Scalar>::value
+                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T>::type* = 0)
+    {
+      Base::setConstant(val0);
+    }
+    
+    // For fixed-size Array<Index,...>
+    template<typename T>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _init1(const Index& val0,
+                                    typename internal::enable_if<    (!internal::is_same<Index,Scalar>::value)
+                                                                  && (internal::is_same<Index,T>::value)
+                                                                  && Base::SizeAtCompileTime!=Dynamic
+                                                                  && Base::SizeAtCompileTime!=1
+                                                                  && internal::is_convertible<T, Scalar>::value
+                                                                  && internal::is_same<typename internal::traits<Derived>::XprKind,ArrayXpr>::value,T*>::type* = 0)
+    {
+      Base::setConstant(val0);
+    }
+    
     template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
     friend struct internal::matrix_swap_impl;
 
-    /** \internal generic implementation of swap for dense storage since for dynamic-sized matrices of same type it is enough to swap the
-      * data pointers.
+  public:
+    
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** \internal
+      * \brief Override DenseBase::swap() since for dynamic-sized matrices
+      * of same type it is enough to swap the data pointers.
       */
     template<typename OtherDerived>
-    void _swap(DenseBase<OtherDerived> const & other)
+    EIGEN_DEVICE_FUNC
+    void swap(DenseBase<OtherDerived> & other)
     {
       enum { SwapPointers = internal::is_same<Derived, OtherDerived>::value && Base::SizeAtCompileTime==Dynamic };
-      internal::matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.const_cast_derived());
+      internal::matrix_swap_impl<Derived, OtherDerived, bool(SwapPointers)>::run(this->derived(), other.derived());
     }
-
-  public:
-#ifndef EIGEN_PARSED_BY_DOXYGEN
+    
+    /** \internal
+      * \brief const version forwarded to DenseBase::swap
+      */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void swap(DenseBase<OtherDerived> const & other)
+    { Base::swap(other.derived()); }
+    
+    EIGEN_DEVICE_FUNC 
     static EIGEN_STRONG_INLINE void _check_template_params()
     {
       EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (Options&RowMajor)==RowMajor)
@@ -667,10 +911,9 @@
                         && (Options & (DontAlign|RowMajor)) == Options),
         INVALID_MATRIX_TEMPLATE_PARAMETERS)
     }
-#endif
 
-private:
-    enum { ThisConstantIsPrivateInPlainObjectBase };
+    enum { IsPlainObjectBase = 1 };
+#endif
 };
 
 namespace internal {
@@ -678,7 +921,6 @@
 template <typename Derived, typename OtherDerived, bool IsVector>
 struct conservative_resize_like_impl
 {
-  typedef typename Derived::Index Index;
   static void run(DenseBase<Derived>& _this, Index rows, Index cols)
   {
     if (_this.rows() == rows && _this.cols() == cols) return;
@@ -694,8 +936,8 @@
     {
       // The storage order does not allow us to use reallocation.
       typename Derived::PlainObject tmp(rows,cols);
-      const Index common_rows = (std::min)(rows, _this.rows());
-      const Index common_cols = (std::min)(cols, _this.cols());
+      const Index common_rows = numext::mini(rows, _this.rows());
+      const Index common_cols = numext::mini(cols, _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
       _this.derived().swap(tmp);
     }
@@ -728,8 +970,8 @@
     {
       // The storage order does not allow us to use reallocation.
       typename Derived::PlainObject tmp(other);
-      const Index common_rows = (std::min)(tmp.rows(), _this.rows());
-      const Index common_cols = (std::min)(tmp.cols(), _this.cols());
+      const Index common_rows = numext::mini(tmp.rows(), _this.rows());
+      const Index common_cols = numext::mini(tmp.cols(), _this.cols());
       tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols);
       _this.derived().swap(tmp);
     }
@@ -744,7 +986,6 @@
 {
   using conservative_resize_like_impl<Derived,OtherDerived,false>::run;
   
-  typedef typename Derived::Index Index;
   static void run(DenseBase<Derived>& _this, Index size)
   {
     const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : size;
@@ -770,6 +1011,7 @@
 template<typename MatrixTypeA, typename MatrixTypeB, bool SwapPointers>
 struct matrix_swap_impl
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(MatrixTypeA& a, MatrixTypeB& b)
   {
     a.base().swap(b);
@@ -779,6 +1021,7 @@
 template<typename MatrixTypeA, typename MatrixTypeB>
 struct matrix_swap_impl<MatrixTypeA, MatrixTypeB, true>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(MatrixTypeA& a, MatrixTypeB& b)
   {
     static_cast<typename MatrixTypeA::Base&>(a).m_storage.swap(static_cast<typename MatrixTypeB::Base&>(b).m_storage);
diff --git a/Eigen/src/Core/Product.h b/Eigen/src/Core/Product.h
new file mode 100644
index 0000000..676c480
--- /dev/null
+++ b/Eigen/src/Core/Product.h
@@ -0,0 +1,186 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PRODUCT_H
+#define EIGEN_PRODUCT_H
+
+namespace Eigen {
+
+template<typename Lhs, typename Rhs, int Option, typename StorageKind> class ProductImpl;
+
+namespace internal {
+
+template<typename Lhs, typename Rhs, int Option>
+struct traits<Product<Lhs, Rhs, Option> >
+{
+  typedef typename remove_all<Lhs>::type LhsCleaned;
+  typedef typename remove_all<Rhs>::type RhsCleaned;
+  typedef traits<LhsCleaned> LhsTraits;
+  typedef traits<RhsCleaned> RhsTraits;
+  
+  typedef MatrixXpr XprKind;
+  
+  typedef typename ScalarBinaryOpTraits<typename traits<LhsCleaned>::Scalar, typename traits<RhsCleaned>::Scalar>::ReturnType Scalar;
+  typedef typename product_promote_storage_type<typename LhsTraits::StorageKind,
+                                                typename RhsTraits::StorageKind,
+                                                internal::product_type<Lhs,Rhs>::ret>::ret StorageKind;
+  typedef typename promote_index_type<typename LhsTraits::StorageIndex,
+                                      typename RhsTraits::StorageIndex>::type StorageIndex;
+  
+  enum {
+    RowsAtCompileTime    = LhsTraits::RowsAtCompileTime,
+    ColsAtCompileTime    = RhsTraits::ColsAtCompileTime,
+    MaxRowsAtCompileTime = LhsTraits::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = RhsTraits::MaxColsAtCompileTime,
+    
+    // FIXME: only needed by GeneralMatrixMatrixTriangular
+    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsTraits::ColsAtCompileTime, RhsTraits::RowsAtCompileTime),
+    
+    // The storage order is somewhat arbitrary here. The correct one will be determined through the evaluator.
+    Flags = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? RowMajorBit
+          : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0
+          : (   ((LhsTraits::Flags&NoPreferredStorageOrderBit) && (RhsTraits::Flags&RowMajorBit))
+             || ((RhsTraits::Flags&NoPreferredStorageOrderBit) && (LhsTraits::Flags&RowMajorBit)) ) ? RowMajorBit
+          : NoPreferredStorageOrderBit
+  };
+};
+
+} // end namespace internal
+
+/** \class Product
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the product of two arbitrary matrices or vectors
+  *
+  * \tparam _Lhs the type of the left-hand side expression
+  * \tparam _Rhs the type of the right-hand side expression
+  *
+  * This class represents an expression of the product of two arbitrary matrices.
+  *
+  * The other template parameters are:
+  * \tparam Option     can be DefaultProduct, AliasFreeProduct, or LazyProduct
+  *
+  */
+template<typename _Lhs, typename _Rhs, int Option>
+class Product : public ProductImpl<_Lhs,_Rhs,Option,
+                                   typename internal::product_promote_storage_type<typename internal::traits<_Lhs>::StorageKind,
+                                                                                   typename internal::traits<_Rhs>::StorageKind,
+                                                                                   internal::product_type<_Lhs,_Rhs>::ret>::ret>
+{
+  public:
+    
+    typedef _Lhs Lhs;
+    typedef _Rhs Rhs;
+    
+    typedef typename ProductImpl<
+        Lhs, Rhs, Option,
+        typename internal::product_promote_storage_type<typename internal::traits<Lhs>::StorageKind,
+                                                        typename internal::traits<Rhs>::StorageKind,
+                                                        internal::product_type<Lhs,Rhs>::ret>::ret>::Base Base;
+    EIGEN_GENERIC_PUBLIC_INTERFACE(Product)
+
+    typedef typename internal::ref_selector<Lhs>::type LhsNested;
+    typedef typename internal::ref_selector<Rhs>::type RhsNested;
+    typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
+    typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
+
+    EIGEN_DEVICE_FUNC Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs)
+    {
+      eigen_assert(lhs.cols() == rhs.rows()
+        && "invalid matrix product"
+        && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
+    }
+
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
+
+    EIGEN_DEVICE_FUNC const LhsNestedCleaned& lhs() const { return m_lhs; }
+    EIGEN_DEVICE_FUNC const RhsNestedCleaned& rhs() const { return m_rhs; }
+
+  protected:
+
+    LhsNested m_lhs;
+    RhsNested m_rhs;
+};
+
+namespace internal {
+  
+template<typename Lhs, typename Rhs, int Option, int ProductTag = internal::product_type<Lhs,Rhs>::ret>
+class dense_product_base
+ : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
+{};
+
+/** Convertion to scalar for inner-products */
+template<typename Lhs, typename Rhs, int Option>
+class dense_product_base<Lhs, Rhs, Option, InnerProduct>
+ : public internal::dense_xpr_base<Product<Lhs,Rhs,Option> >::type
+{
+  typedef Product<Lhs,Rhs,Option> ProductXpr;
+  typedef typename internal::dense_xpr_base<ProductXpr>::type Base;
+public:
+  using Base::derived;
+  typedef typename Base::Scalar Scalar;
+  
+  EIGEN_STRONG_INLINE operator const Scalar() const
+  {
+    return internal::evaluator<ProductXpr>(derived()).coeff(0,0);
+  }
+};
+
+} // namespace internal
+
+// Generic API dispatcher
+template<typename Lhs, typename Rhs, int Option, typename StorageKind>
+class ProductImpl : public internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type
+{
+  public:
+    typedef typename internal::generic_xpr_base<Product<Lhs,Rhs,Option>, MatrixXpr, StorageKind>::type Base;
+};
+
+template<typename Lhs, typename Rhs, int Option>
+class ProductImpl<Lhs,Rhs,Option,Dense>
+  : public internal::dense_product_base<Lhs,Rhs,Option>
+{
+    typedef Product<Lhs, Rhs, Option> Derived;
+    
+  public:
+    
+    typedef typename internal::dense_product_base<Lhs, Rhs, Option> Base;
+    EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
+  protected:
+    enum {
+      IsOneByOne = (RowsAtCompileTime == 1 || RowsAtCompileTime == Dynamic) && 
+                   (ColsAtCompileTime == 1 || ColsAtCompileTime == Dynamic),
+      EnableCoeff = IsOneByOne || Option==LazyProduct
+    };
+    
+  public:
+  
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index row, Index col) const
+    {
+      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
+      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
+      
+      return internal::evaluator<Derived>(derived()).coeff(row,col);
+    }
+
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index i) const
+    {
+      EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS);
+      eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) );
+      
+      return internal::evaluator<Derived>(derived()).coeff(i);
+    }
+    
+  
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_PRODUCT_H
diff --git a/Eigen/src/Core/ProductBase.h b/Eigen/src/Core/ProductBase.h
deleted file mode 100644
index cf74470..0000000
--- a/Eigen/src/Core/ProductBase.h
+++ /dev/null
@@ -1,290 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_PRODUCTBASE_H
-#define EIGEN_PRODUCTBASE_H
-
-namespace Eigen { 
-
-/** \class ProductBase
-  * \ingroup Core_Module
-  *
-  */
-
-namespace internal {
-template<typename Derived, typename _Lhs, typename _Rhs>
-struct traits<ProductBase<Derived,_Lhs,_Rhs> >
-{
-  typedef MatrixXpr XprKind;
-  typedef typename remove_all<_Lhs>::type Lhs;
-  typedef typename remove_all<_Rhs>::type Rhs;
-  typedef typename scalar_product_traits<typename Lhs::Scalar, typename Rhs::Scalar>::ReturnType Scalar;
-  typedef typename promote_storage_type<typename traits<Lhs>::StorageKind,
-                                           typename traits<Rhs>::StorageKind>::ret StorageKind;
-  typedef typename promote_index_type<typename traits<Lhs>::Index,
-                                         typename traits<Rhs>::Index>::type Index;
-  enum {
-    RowsAtCompileTime = traits<Lhs>::RowsAtCompileTime,
-    ColsAtCompileTime = traits<Rhs>::ColsAtCompileTime,
-    MaxRowsAtCompileTime = traits<Lhs>::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = traits<Rhs>::MaxColsAtCompileTime,
-    Flags = (MaxRowsAtCompileTime==1 ? RowMajorBit : 0)
-          | EvalBeforeNestingBit | EvalBeforeAssigningBit | NestByRefBit,
-                  // Note that EvalBeforeNestingBit and NestByRefBit
-                  // are not used in practice because nested is overloaded for products
-    CoeffReadCost = 0 // FIXME why is it needed ?
-  };
-};
-}
-
-#define EIGEN_PRODUCT_PUBLIC_INTERFACE(Derived) \
-  typedef ProductBase<Derived, Lhs, Rhs > Base; \
-  EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
-  typedef typename Base::LhsNested LhsNested; \
-  typedef typename Base::_LhsNested _LhsNested; \
-  typedef typename Base::LhsBlasTraits LhsBlasTraits; \
-  typedef typename Base::ActualLhsType ActualLhsType; \
-  typedef typename Base::_ActualLhsType _ActualLhsType; \
-  typedef typename Base::RhsNested RhsNested; \
-  typedef typename Base::_RhsNested _RhsNested; \
-  typedef typename Base::RhsBlasTraits RhsBlasTraits; \
-  typedef typename Base::ActualRhsType ActualRhsType; \
-  typedef typename Base::_ActualRhsType _ActualRhsType; \
-  using Base::m_lhs; \
-  using Base::m_rhs;
-
-template<typename Derived, typename Lhs, typename Rhs>
-class ProductBase : public MatrixBase<Derived>
-{
-  public:
-    typedef MatrixBase<Derived> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(ProductBase)
-    
-    typedef typename Lhs::Nested LhsNested;
-    typedef typename internal::remove_all<LhsNested>::type _LhsNested;
-    typedef internal::blas_traits<_LhsNested> LhsBlasTraits;
-    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
-    typedef typename internal::remove_all<ActualLhsType>::type _ActualLhsType;
-    typedef typename internal::traits<Lhs>::Scalar LhsScalar;
-
-    typedef typename Rhs::Nested RhsNested;
-    typedef typename internal::remove_all<RhsNested>::type _RhsNested;
-    typedef internal::blas_traits<_RhsNested> RhsBlasTraits;
-    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
-    typedef typename internal::remove_all<ActualRhsType>::type _ActualRhsType;
-    typedef typename internal::traits<Rhs>::Scalar RhsScalar;
-
-    // Diagonal of a product: no need to evaluate the arguments because they are going to be evaluated only once
-    typedef CoeffBasedProduct<LhsNested, RhsNested, 0> FullyLazyCoeffBaseProductType;
-
-  public:
-
-#ifndef EIGEN_NO_MALLOC
-    typedef typename Base::PlainObject BasePlainObject;
-    typedef Matrix<Scalar,RowsAtCompileTime==1?1:Dynamic,ColsAtCompileTime==1?1:Dynamic,BasePlainObject::Options> DynPlainObject;
-    typedef typename internal::conditional<(BasePlainObject::SizeAtCompileTime==Dynamic) || (BasePlainObject::SizeAtCompileTime*int(sizeof(Scalar)) < int(EIGEN_STACK_ALLOCATION_LIMIT)),
-                                           BasePlainObject, DynPlainObject>::type PlainObject;
-#else
-    typedef typename Base::PlainObject PlainObject;
-#endif
-
-    ProductBase(const Lhs& a_lhs, const Rhs& a_rhs)
-      : m_lhs(a_lhs), m_rhs(a_rhs)
-    {
-      eigen_assert(a_lhs.cols() == a_rhs.rows()
-        && "invalid matrix product"
-        && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
-    }
-
-    inline Index rows() const { return m_lhs.rows(); }
-    inline Index cols() const { return m_rhs.cols(); }
-
-    template<typename Dest>
-    inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst,Scalar(1)); }
-
-    template<typename Dest>
-    inline void addTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(1)); }
-
-    template<typename Dest>
-    inline void subTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(-1)); }
-
-    template<typename Dest>
-    inline void scaleAndAddTo(Dest& dst, const Scalar& alpha) const { derived().scaleAndAddTo(dst,alpha); }
-
-    const _LhsNested& lhs() const { return m_lhs; }
-    const _RhsNested& rhs() const { return m_rhs; }
-
-    // Implicit conversion to the nested type (trigger the evaluation of the product)
-    operator const PlainObject& () const
-    {
-      m_result.resize(m_lhs.rows(), m_rhs.cols());
-      derived().evalTo(m_result);
-      return m_result;
-    }
-
-    const Diagonal<const FullyLazyCoeffBaseProductType,0> diagonal() const
-    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
-
-    template<int Index>
-    const Diagonal<FullyLazyCoeffBaseProductType,Index> diagonal() const
-    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); }
-
-    const Diagonal<FullyLazyCoeffBaseProductType,Dynamic> diagonal(Index index) const
-    { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs).diagonal(index); }
-
-    // restrict coeff accessors to 1x1 expressions. No need to care about mutators here since this isnt a Lvalue expression
-    typename Base::CoeffReturnType coeff(Index row, Index col) const
-    {
-#ifdef EIGEN2_SUPPORT
-      return lhs().row(row).cwiseProduct(rhs().col(col).transpose()).sum();
-#else
-      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
-      eigen_assert(this->rows() == 1 && this->cols() == 1);
-      Matrix<Scalar,1,1> result = *this;
-      return result.coeff(row,col);
-#endif
-    }
-
-    typename Base::CoeffReturnType coeff(Index i) const
-    {
-      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
-      eigen_assert(this->rows() == 1 && this->cols() == 1);
-      Matrix<Scalar,1,1> result = *this;
-      return result.coeff(i);
-    }
-
-    const Scalar& coeffRef(Index row, Index col) const
-    {
-      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
-      eigen_assert(this->rows() == 1 && this->cols() == 1);
-      return derived().coeffRef(row,col);
-    }
-
-    const Scalar& coeffRef(Index i) const
-    {
-      EIGEN_STATIC_ASSERT_SIZE_1x1(Derived)
-      eigen_assert(this->rows() == 1 && this->cols() == 1);
-      return derived().coeffRef(i);
-    }
-
-  protected:
-
-    LhsNested m_lhs;
-    RhsNested m_rhs;
-
-    mutable PlainObject m_result;
-};
-
-// here we need to overload the nested rule for products
-// such that the nested type is a const reference to a plain matrix
-namespace internal {
-template<typename Lhs, typename Rhs, int Mode, int N, typename PlainObject>
-struct nested<GeneralProduct<Lhs,Rhs,Mode>, N, PlainObject>
-{
-  typedef typename GeneralProduct<Lhs,Rhs,Mode>::PlainObject const& type;
-};
-template<typename Lhs, typename Rhs, int Mode, int N, typename PlainObject>
-struct nested<const GeneralProduct<Lhs,Rhs,Mode>, N, PlainObject>
-{
-  typedef typename GeneralProduct<Lhs,Rhs,Mode>::PlainObject const& type;
-};
-}
-
-template<typename NestedProduct>
-class ScaledProduct;
-
-// Note that these two operator* functions are not defined as member
-// functions of ProductBase, because, otherwise we would have to
-// define all overloads defined in MatrixBase. Furthermore, Using
-// "using Base::operator*" would not work with MSVC.
-//
-// Also note that here we accept any compatible scalar types
-template<typename Derived,typename Lhs,typename Rhs>
-const ScaledProduct<Derived>
-operator*(const ProductBase<Derived,Lhs,Rhs>& prod, const typename Derived::Scalar& x)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-template<typename Derived,typename Lhs,typename Rhs>
-typename internal::enable_if<!internal::is_same<typename Derived::Scalar,typename Derived::RealScalar>::value,
-                      const ScaledProduct<Derived> >::type
-operator*(const ProductBase<Derived,Lhs,Rhs>& prod, const typename Derived::RealScalar& x)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-
-template<typename Derived,typename Lhs,typename Rhs>
-const ScaledProduct<Derived>
-operator*(const typename Derived::Scalar& x,const ProductBase<Derived,Lhs,Rhs>& prod)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-template<typename Derived,typename Lhs,typename Rhs>
-typename internal::enable_if<!internal::is_same<typename Derived::Scalar,typename Derived::RealScalar>::value,
-                      const ScaledProduct<Derived> >::type
-operator*(const typename Derived::RealScalar& x,const ProductBase<Derived,Lhs,Rhs>& prod)
-{ return ScaledProduct<Derived>(prod.derived(), x); }
-
-namespace internal {
-template<typename NestedProduct>
-struct traits<ScaledProduct<NestedProduct> >
- : traits<ProductBase<ScaledProduct<NestedProduct>,
-                         typename NestedProduct::_LhsNested,
-                         typename NestedProduct::_RhsNested> >
-{
-  typedef typename traits<NestedProduct>::StorageKind StorageKind;
-};
-}
-
-template<typename NestedProduct>
-class ScaledProduct
-  : public ProductBase<ScaledProduct<NestedProduct>,
-                       typename NestedProduct::_LhsNested,
-                       typename NestedProduct::_RhsNested>
-{
-  public:
-    typedef ProductBase<ScaledProduct<NestedProduct>,
-                       typename NestedProduct::_LhsNested,
-                       typename NestedProduct::_RhsNested> Base;
-    typedef typename Base::Scalar Scalar;
-    typedef typename Base::PlainObject PlainObject;
-//     EIGEN_PRODUCT_PUBLIC_INTERFACE(ScaledProduct)
-
-    ScaledProduct(const NestedProduct& prod, const Scalar& x)
-    : Base(prod.lhs(),prod.rhs()), m_prod(prod), m_alpha(x) {}
-
-    template<typename Dest>
-    inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst, Scalar(1)); }
-
-    template<typename Dest>
-    inline void addTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(1)); }
-
-    template<typename Dest>
-    inline void subTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(-1)); }
-
-    template<typename Dest>
-    inline void scaleAndAddTo(Dest& dst, const Scalar& a_alpha) const { m_prod.derived().scaleAndAddTo(dst,a_alpha * m_alpha); }
-
-    const Scalar& alpha() const { return m_alpha; }
-    
-  protected:
-    const NestedProduct& m_prod;
-    Scalar m_alpha;
-};
-
-/** \internal
-  * Overloaded to perform an efficient C = (A*B).lazy() */
-template<typename Derived>
-template<typename ProductDerived, typename Lhs, typename Rhs>
-Derived& MatrixBase<Derived>::lazyAssign(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-{
-  other.derived().evalTo(derived());
-  return derived();
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_PRODUCTBASE_H
diff --git a/Eigen/src/Core/ProductEvaluators.h b/Eigen/src/Core/ProductEvaluators.h
new file mode 100644
index 0000000..9b99bd7
--- /dev/null
+++ b/Eigen/src/Core/ProductEvaluators.h
@@ -0,0 +1,1112 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011 Jitse Niesen <jitse@maths.leeds.ac.uk>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+
+#ifndef EIGEN_PRODUCTEVALUATORS_H
+#define EIGEN_PRODUCTEVALUATORS_H
+
+namespace Eigen {
+  
+namespace internal {
+
+/** \internal
+  * Evaluator of a product expression.
+  * Since products require special treatments to handle all possible cases,
+  * we simply deffer the evaluation logic to a product_evaluator class
+  * which offers more partial specialization possibilities.
+  * 
+  * \sa class product_evaluator
+  */
+template<typename Lhs, typename Rhs, int Options>
+struct evaluator<Product<Lhs, Rhs, Options> > 
+ : public product_evaluator<Product<Lhs, Rhs, Options> >
+{
+  typedef Product<Lhs, Rhs, Options> XprType;
+  typedef product_evaluator<XprType> Base;
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr) : Base(xpr) {}
+};
+ 
+// Catch "scalar * ( A * B )" and transform it to "(A*scalar) * B"
+// TODO we should apply that rule only if that's really helpful
+template<typename Lhs, typename Rhs, typename Scalar1, typename Scalar2, typename Plain1>
+struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
+                                               const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>,
+                                               const Product<Lhs, Rhs, DefaultProduct> > >
+{
+  static const bool value = true;
+};
+template<typename Lhs, typename Rhs, typename Scalar1, typename Scalar2, typename Plain1>
+struct evaluator<CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
+                               const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>,
+                               const Product<Lhs, Rhs, DefaultProduct> > >
+ : public evaluator<Product<EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar1,Lhs,product), Rhs, DefaultProduct> >
+{
+  typedef CwiseBinaryOp<internal::scalar_product_op<Scalar1,Scalar2>,
+                               const CwiseNullaryOp<internal::scalar_constant_op<Scalar1>, Plain1>,
+                               const Product<Lhs, Rhs, DefaultProduct> > XprType;
+  typedef evaluator<Product<EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar1,Lhs,product), Rhs, DefaultProduct> > Base;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr)
+    : Base(xpr.lhs().functor().m_other * xpr.rhs().lhs() * xpr.rhs().rhs())
+  {}
+};
+
+
+template<typename Lhs, typename Rhs, int DiagIndex>
+struct evaluator<Diagonal<const Product<Lhs, Rhs, DefaultProduct>, DiagIndex> > 
+ : public evaluator<Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex> >
+{
+  typedef Diagonal<const Product<Lhs, Rhs, DefaultProduct>, DiagIndex> XprType;
+  typedef evaluator<Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex> > Base;
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr)
+    : Base(Diagonal<const Product<Lhs, Rhs, LazyProduct>, DiagIndex>(
+        Product<Lhs, Rhs, LazyProduct>(xpr.nestedExpression().lhs(), xpr.nestedExpression().rhs()),
+        xpr.index() ))
+  {}
+};
+
+
+// Helper class to perform a matrix product with the destination at hand.
+// Depending on the sizes of the factors, there are different evaluation strategies
+// as controlled by internal::product_type.
+template< typename Lhs, typename Rhs,
+          typename LhsShape = typename evaluator_traits<Lhs>::Shape,
+          typename RhsShape = typename evaluator_traits<Rhs>::Shape,
+          int ProductType = internal::product_type<Lhs,Rhs>::value>
+struct generic_product_impl;
+
+template<typename Lhs, typename Rhs>
+struct evaluator_assume_aliasing<Product<Lhs, Rhs, DefaultProduct> > {
+  static const bool value = true;
+};
+
+// This is the default evaluator implementation for products:
+// It creates a temporary and call generic_product_impl
+template<typename Lhs, typename Rhs, int Options, int ProductTag, typename LhsShape, typename RhsShape>
+struct product_evaluator<Product<Lhs, Rhs, Options>, ProductTag, LhsShape, RhsShape>
+  : public evaluator<typename Product<Lhs, Rhs, Options>::PlainObject>
+{
+  typedef Product<Lhs, Rhs, Options> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  explicit product_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    
+// FIXME shall we handle nested_eval here?,
+// if so, then we must take care at removing the call to nested_eval in the specializations (e.g., in permutation_matrix_product, transposition_matrix_product, etc.)
+//     typedef typename internal::nested_eval<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
+//     typedef typename internal::nested_eval<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
+//     typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
+//     typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
+//     
+//     const LhsNested lhs(xpr.lhs());
+//     const RhsNested rhs(xpr.rhs());
+//   
+//     generic_product_impl<LhsNestedCleaned, RhsNestedCleaned>::evalTo(m_result, lhs, rhs);
+
+    generic_product_impl<Lhs, Rhs, LhsShape, RhsShape, ProductTag>::evalTo(m_result, xpr.lhs(), xpr.rhs());
+  }
+  
+protected:  
+  PlainObject m_result;
+};
+
+// The following three shortcuts are enabled only if the scalar types match excatly.
+// TODO: we could enable them for different scalar types when the product is not vectorized.
+
+// Dense = Product
+template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::assign_op<Scalar,Scalar>, Dense2Dense,
+  typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
+{
+  typedef Product<Lhs,Rhs,Options> SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    // FIXME shall we handle nested_eval here?
+    generic_product_impl<Lhs, Rhs>::evalTo(dst, src.lhs(), src.rhs());
+  }
+};
+
+// Dense += Product
+template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::add_assign_op<Scalar,Scalar>, Dense2Dense,
+  typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
+{
+  typedef Product<Lhs,Rhs,Options> SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,Scalar> &)
+  {
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    // FIXME shall we handle nested_eval here?
+    generic_product_impl<Lhs, Rhs>::addTo(dst, src.lhs(), src.rhs());
+  }
+};
+
+// Dense -= Product
+template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,Options>, internal::sub_assign_op<Scalar,Scalar>, Dense2Dense,
+  typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type>
+{
+  typedef Product<Lhs,Rhs,Options> SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,Scalar> &)
+  {
+    eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols());
+    // FIXME shall we handle nested_eval here?
+    generic_product_impl<Lhs, Rhs>::subTo(dst, src.lhs(), src.rhs());
+  }
+};
+
+
+// Dense ?= scalar * Product
+// TODO we should apply that rule if that's really helpful
+// for instance, this is not good for inner products
+template< typename DstXprType, typename Lhs, typename Rhs, typename AssignFunc, typename Scalar, typename ScalarBis, typename Plain>
+struct Assignment<DstXprType, CwiseBinaryOp<internal::scalar_product_op<ScalarBis,Scalar>, const CwiseNullaryOp<internal::scalar_constant_op<ScalarBis>,Plain>,
+                                           const Product<Lhs,Rhs,DefaultProduct> >, AssignFunc, Dense2Dense>
+{
+  typedef CwiseBinaryOp<internal::scalar_product_op<ScalarBis,Scalar>,
+                        const CwiseNullaryOp<internal::scalar_constant_op<ScalarBis>,Plain>,
+                        const Product<Lhs,Rhs,DefaultProduct> > SrcXprType;
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const AssignFunc& func)
+  {
+    call_assignment_no_alias(dst, (src.lhs().functor().m_other * src.rhs().lhs())*src.rhs().rhs(), func);
+  }
+};
+
+//----------------------------------------
+// Catch "Dense ?= xpr + Product<>" expression to save one temporary
+// FIXME we could probably enable these rules for any product, i.e., not only Dense and DefaultProduct
+
+template<typename OtherXpr, typename Lhs, typename Rhs>
+struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_sum_op<typename OtherXpr::Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, const OtherXpr,
+                                               const Product<Lhs,Rhs,DefaultProduct> >, DenseShape > {
+  static const bool value = true;
+};
+
+template<typename OtherXpr, typename Lhs, typename Rhs>
+struct evaluator_assume_aliasing<CwiseBinaryOp<internal::scalar_difference_op<typename OtherXpr::Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, const OtherXpr,
+                                               const Product<Lhs,Rhs,DefaultProduct> >, DenseShape > {
+  static const bool value = true;
+};
+
+template<typename DstXprType, typename OtherXpr, typename ProductType, typename Func1, typename Func2>
+struct assignment_from_xpr_op_product
+{
+  template<typename SrcXprType, typename InitialFunc>
+  static EIGEN_STRONG_INLINE
+  void run(DstXprType &dst, const SrcXprType &src, const InitialFunc& /*func*/)
+  {
+    call_assignment_no_alias(dst, src.lhs(), Func1());
+    call_assignment_no_alias(dst, src.rhs(), Func2());
+  }
+};
+
+#define EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(ASSIGN_OP,BINOP,ASSIGN_OP2) \
+  template< typename DstXprType, typename OtherXpr, typename Lhs, typename Rhs, typename DstScalar, typename SrcScalar, typename OtherScalar,typename ProdScalar> \
+  struct Assignment<DstXprType, CwiseBinaryOp<internal::BINOP<OtherScalar,ProdScalar>, const OtherXpr, \
+                                            const Product<Lhs,Rhs,DefaultProduct> >, internal::ASSIGN_OP<DstScalar,SrcScalar>, Dense2Dense> \
+    : assignment_from_xpr_op_product<DstXprType, OtherXpr, Product<Lhs,Rhs,DefaultProduct>, internal::ASSIGN_OP<DstScalar,OtherScalar>, internal::ASSIGN_OP2<DstScalar,ProdScalar> > \
+  {}
+
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(assign_op,    scalar_sum_op,add_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(add_assign_op,scalar_sum_op,add_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(sub_assign_op,scalar_sum_op,sub_assign_op);
+
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(assign_op,    scalar_difference_op,sub_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(add_assign_op,scalar_difference_op,sub_assign_op);
+EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(sub_assign_op,scalar_difference_op,add_assign_op);
+
+//----------------------------------------
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,InnerProduct>
+{
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    dst.coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum();
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    dst.coeffRef(0,0) += (lhs.transpose().cwiseProduct(rhs)).sum();
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { dst.coeffRef(0,0) -= (lhs.transpose().cwiseProduct(rhs)).sum(); }
+};
+
+
+/***********************************************************************
+*  Implementation of outer dense * dense vector product
+***********************************************************************/
+
+// Column major result
+template<typename Dst, typename Lhs, typename Rhs, typename Func>
+void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const false_type&)
+{
+  evaluator<Rhs> rhsEval(rhs);
+  typename nested_eval<Lhs,Rhs::SizeAtCompileTime>::type actual_lhs(lhs);
+  // FIXME if cols is large enough, then it might be useful to make sure that lhs is sequentially stored
+  // FIXME not very good if rhs is real and lhs complex while alpha is real too
+  const Index cols = dst.cols();
+  for (Index j=0; j<cols; ++j)
+    func(dst.col(j), rhsEval.coeff(Index(0),j) * actual_lhs);
+}
+
+// Row major result
+template<typename Dst, typename Lhs, typename Rhs, typename Func>
+void outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const true_type&)
+{
+  evaluator<Lhs> lhsEval(lhs);
+  typename nested_eval<Rhs,Lhs::SizeAtCompileTime>::type actual_rhs(rhs);
+  // FIXME if rows is large enough, then it might be useful to make sure that rhs is sequentially stored
+  // FIXME not very good if lhs is real and rhs complex while alpha is real too
+  const Index rows = dst.rows();
+  for (Index i=0; i<rows; ++i)
+    func(dst.row(i), lhsEval.coeff(i,Index(0)) * actual_rhs);
+}
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,OuterProduct>
+{
+  template<typename T> struct is_row_major : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {};
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  // TODO it would be nice to be able to exploit our *_assign_op functors for that purpose
+  struct set  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived()  = src; } };
+  struct add  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } };
+  struct sub  { template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } };
+  struct adds {
+    Scalar m_scale;
+    explicit adds(const Scalar& s) : m_scale(s) {}
+    template<typename Dst, typename Src> void operator()(const Dst& dst, const Src& src) const {
+      dst.const_cast_derived() += m_scale * src;
+    }
+  };
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, set(), is_row_major<Dst>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, add(), is_row_major<Dst>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, sub(), is_row_major<Dst>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    internal::outer_product_selector_run(dst, lhs, rhs, adds(alpha), is_row_major<Dst>());
+  }
+  
+};
+
+
+// This base class provides default implementations for evalTo, addTo, subTo, in terms of scaleAndAddTo
+template<typename Lhs, typename Rhs, typename Derived>
+struct generic_product_impl_base
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { dst.setZero(); scaleAndAddTo(dst, lhs, rhs, Scalar(1)); }
+
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { scaleAndAddTo(dst,lhs, rhs, Scalar(1)); }
+
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  { scaleAndAddTo(dst, lhs, rhs, Scalar(-1)); }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  { Derived::scaleAndAddTo(dst,lhs,rhs,alpha); }
+
+};
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemvProduct>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemvProduct> >
+{
+  typedef typename nested_eval<Lhs,1>::type LhsNested;
+  typedef typename nested_eval<Rhs,1>::type RhsNested;
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight };
+  typedef typename internal::remove_all<typename internal::conditional<int(Side)==OnTheRight,LhsNested,RhsNested>::type>::type MatrixType;
+
+  template<typename Dest>
+  static EIGEN_STRONG_INLINE void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    LhsNested actual_lhs(lhs);
+    RhsNested actual_rhs(rhs);
+    internal::gemv_dense_selector<Side,
+                            (int(MatrixType::Flags)&RowMajorBit) ? RowMajor : ColMajor,
+                            bool(internal::blas_traits<MatrixType>::HasUsableDirectAccess)
+                           >::run(actual_lhs, actual_rhs, dst, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> 
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    // Same as: dst.noalias() = lhs.lazyProduct(rhs);
+    // but easier on the compiler side
+    call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::assign_op<typename Dst::Scalar,Scalar>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    // dst.noalias() += lhs.lazyProduct(rhs);
+    call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::add_assign_op<typename Dst::Scalar,Scalar>());
+  }
+  
+  template<typename Dst>
+  static EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    // dst.noalias() -= lhs.lazyProduct(rhs);
+    call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::sub_assign_op<typename Dst::Scalar,Scalar>());
+  }
+  
+//   template<typename Dst>
+//   static inline void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+//   { dst.noalias() += alpha * lhs.lazyProduct(rhs); }
+};
+
+// This specialization enforces the use of a coefficient-based evaluation strategy
+template<typename Lhs, typename Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,LazyCoeffBasedProductMode>
+  : generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> {};
+
+// Case 2: Evaluate coeff by coeff
+//
+// This is mostly taken from CoeffBasedProduct.h
+// The main difference is that we add an extra argument to the etor_product_*_impl::run() function
+// for the inner dimension of the product, because evaluator object do not know their size.
+
+template<int Traversal, int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
+struct etor_product_coeff_impl;
+
+template<int StorageOrder, int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl;
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape, DenseShape>
+    : evaluator_base<Product<Lhs, Rhs, LazyProduct> >
+{
+  typedef Product<Lhs, Rhs, LazyProduct> XprType;
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  explicit product_evaluator(const XprType& xpr)
+    : m_lhs(xpr.lhs()),
+      m_rhs(xpr.rhs()),
+      m_lhsImpl(m_lhs),     // FIXME the creation of the evaluator objects should result in a no-op, but check that!
+      m_rhsImpl(m_rhs),     //       Moreover, they are only useful for the packet path, so we could completely disable them when not needed,
+                            //       or perhaps declare them on the fly on the packet method... We have experiment to check what's best.
+      m_innerDim(xpr.lhs().cols())
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits<Scalar>::MulCost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits<Scalar>::AddCost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+#if 0
+    std::cerr << "LhsOuterStrideBytes=  " << LhsOuterStrideBytes << "\n";
+    std::cerr << "RhsOuterStrideBytes=  " << RhsOuterStrideBytes << "\n";
+    std::cerr << "LhsAlignment=         " << LhsAlignment << "\n";
+    std::cerr << "RhsAlignment=         " << RhsAlignment << "\n";
+    std::cerr << "CanVectorizeLhs=      " << CanVectorizeLhs << "\n";
+    std::cerr << "CanVectorizeRhs=      " << CanVectorizeRhs << "\n";
+    std::cerr << "CanVectorizeInner=    " << CanVectorizeInner << "\n";
+    std::cerr << "EvalToRowMajor=       " << EvalToRowMajor << "\n";
+    std::cerr << "Alignment=            " << Alignment << "\n";
+    std::cerr << "Flags=                " << Flags << "\n";
+#endif
+  }
+
+  // Everything below here is taken from CoeffBasedProduct.h
+
+  typedef typename internal::nested_eval<Lhs,Rhs::ColsAtCompileTime>::type LhsNested;
+  typedef typename internal::nested_eval<Rhs,Lhs::RowsAtCompileTime>::type RhsNested;
+  
+  typedef typename internal::remove_all<LhsNested>::type LhsNestedCleaned;
+  typedef typename internal::remove_all<RhsNested>::type RhsNestedCleaned;
+
+  typedef evaluator<LhsNestedCleaned> LhsEtorType;
+  typedef evaluator<RhsNestedCleaned> RhsEtorType;
+
+  enum {
+    RowsAtCompileTime = LhsNestedCleaned::RowsAtCompileTime,
+    ColsAtCompileTime = RhsNestedCleaned::ColsAtCompileTime,
+    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsNestedCleaned::ColsAtCompileTime, RhsNestedCleaned::RowsAtCompileTime),
+    MaxRowsAtCompileTime = LhsNestedCleaned::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = RhsNestedCleaned::MaxColsAtCompileTime
+  };
+
+  typedef typename find_best_packet<Scalar,RowsAtCompileTime>::type LhsVecPacketType;
+  typedef typename find_best_packet<Scalar,ColsAtCompileTime>::type RhsVecPacketType;
+
+  enum {
+      
+    LhsCoeffReadCost = LhsEtorType::CoeffReadCost,
+    RhsCoeffReadCost = RhsEtorType::CoeffReadCost,
+    CoeffReadCost = InnerSize==0 ? NumTraits<Scalar>::ReadCost
+                  : InnerSize == Dynamic ? HugeCost
+                  : InnerSize * (NumTraits<Scalar>::MulCost + LhsCoeffReadCost + RhsCoeffReadCost)
+                    + (InnerSize - 1) * NumTraits<Scalar>::AddCost,
+
+    Unroll = CoeffReadCost <= EIGEN_UNROLLING_LIMIT,
+    
+    LhsFlags = LhsEtorType::Flags,
+    RhsFlags = RhsEtorType::Flags,
+    
+    LhsRowMajor = LhsFlags & RowMajorBit,
+    RhsRowMajor = RhsFlags & RowMajorBit,
+
+    LhsVecPacketSize = unpacket_traits<LhsVecPacketType>::size,
+    RhsVecPacketSize = unpacket_traits<RhsVecPacketType>::size,
+
+    // Here, we don't care about alignment larger than the usable packet size.
+    LhsAlignment = EIGEN_PLAIN_ENUM_MIN(LhsEtorType::Alignment,LhsVecPacketSize*int(sizeof(typename LhsNestedCleaned::Scalar))),
+    RhsAlignment = EIGEN_PLAIN_ENUM_MIN(RhsEtorType::Alignment,RhsVecPacketSize*int(sizeof(typename RhsNestedCleaned::Scalar))),
+      
+    SameType = is_same<typename LhsNestedCleaned::Scalar,typename RhsNestedCleaned::Scalar>::value,
+
+    CanVectorizeRhs = bool(RhsRowMajor) && (RhsFlags & PacketAccessBit) && (ColsAtCompileTime!=1),
+    CanVectorizeLhs = (!LhsRowMajor) && (LhsFlags & PacketAccessBit) && (RowsAtCompileTime!=1),
+
+    EvalToRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
+                    : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
+                    : (bool(RhsRowMajor) && !CanVectorizeLhs),
+
+    Flags = ((unsigned int)(LhsFlags | RhsFlags) & HereditaryBits & ~RowMajorBit)
+          | (EvalToRowMajor ? RowMajorBit : 0)
+          // TODO enable vectorization for mixed types
+          | (SameType && (CanVectorizeLhs || CanVectorizeRhs) ? PacketAccessBit : 0)
+          | (XprType::IsVectorAtCompileTime ? LinearAccessBit : 0),
+          
+    LhsOuterStrideBytes = int(LhsNestedCleaned::OuterStrideAtCompileTime) * int(sizeof(typename LhsNestedCleaned::Scalar)),
+    RhsOuterStrideBytes = int(RhsNestedCleaned::OuterStrideAtCompileTime) * int(sizeof(typename RhsNestedCleaned::Scalar)),
+
+    Alignment = bool(CanVectorizeLhs) ? (LhsOuterStrideBytes<=0 || (int(LhsOuterStrideBytes) % EIGEN_PLAIN_ENUM_MAX(1,LhsAlignment))!=0 ? 0 : LhsAlignment)
+              : bool(CanVectorizeRhs) ? (RhsOuterStrideBytes<=0 || (int(RhsOuterStrideBytes) % EIGEN_PLAIN_ENUM_MAX(1,RhsAlignment))!=0 ? 0 : RhsAlignment)
+              : 0,
+
+    /* CanVectorizeInner deserves special explanation. It does not affect the product flags. It is not used outside
+     * of Product. If the Product itself is not a packet-access expression, there is still a chance that the inner
+     * loop of the product might be vectorized. This is the meaning of CanVectorizeInner. Since it doesn't affect
+     * the Flags, it is safe to make this value depend on ActualPacketAccessBit, that doesn't affect the ABI.
+     */
+    CanVectorizeInner =    SameType
+                        && LhsRowMajor
+                        && (!RhsRowMajor)
+                        && (LhsFlags & RhsFlags & ActualPacketAccessBit)
+                        && (InnerSize % packet_traits<Scalar>::size == 0)
+  };
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index row, Index col) const
+  {
+    return (m_lhs.row(row).transpose().cwiseProduct( m_rhs.col(col) )).sum();
+  }
+
+  /* Allow index-based non-packet access. It is impossible though to allow index-based packed access,
+   * which is why we don't set the LinearAccessBit.
+   * TODO: this seems possible when the result is a vector
+   */
+  EIGEN_DEVICE_FUNC const CoeffReturnType coeff(Index index) const
+  {
+    const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index;
+    const Index col = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? index : 0;
+    return (m_lhs.row(row).transpose().cwiseProduct( m_rhs.col(col) )).sum();
+  }
+
+  template<int LoadMode, typename PacketType>
+  const PacketType packet(Index row, Index col) const
+  {
+    PacketType res;
+    typedef etor_product_packet_impl<bool(int(Flags)&RowMajorBit) ? RowMajor : ColMajor,
+                                     Unroll ? int(InnerSize) : Dynamic,
+                                     LhsEtorType, RhsEtorType, PacketType, LoadMode> PacketImpl;
+    PacketImpl::run(row, col, m_lhsImpl, m_rhsImpl, m_innerDim, res);
+    return res;
+  }
+
+  template<int LoadMode, typename PacketType>
+  const PacketType packet(Index index) const
+  {
+    const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index;
+    const Index col = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? index : 0;
+    return packet<LoadMode,PacketType>(row,col);
+  }
+
+protected:
+  typename internal::add_const_on_value_type<LhsNested>::type m_lhs;
+  typename internal::add_const_on_value_type<RhsNested>::type m_rhs;
+  
+  LhsEtorType m_lhsImpl;
+  RhsEtorType m_rhsImpl;
+
+  // TODO: Get rid of m_innerDim if known at compile time
+  Index m_innerDim;
+};
+
+template<typename Lhs, typename Rhs>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, LazyCoeffBasedProductMode, DenseShape, DenseShape>
+  : product_evaluator<Product<Lhs, Rhs, LazyProduct>, CoeffBasedProductMode, DenseShape, DenseShape>
+{
+  typedef Product<Lhs, Rhs, DefaultProduct> XprType;
+  typedef Product<Lhs, Rhs, LazyProduct> BaseProduct;
+  typedef product_evaluator<BaseProduct, CoeffBasedProductMode, DenseShape, DenseShape> Base;
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(BaseProduct(xpr.lhs(),xpr.rhs()))
+  {}
+};
+
+/****************************************
+*** Coeff based product, Packet path  ***
+****************************************/
+
+template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
+  {
+    etor_product_packet_impl<RowMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, innerDim, res);
+    res =  pmadd(pset1<Packet>(lhs.coeff(row, Index(UnrollingIndex-1))), rhs.template packet<LoadMode,Packet>(Index(UnrollingIndex-1), col), res);
+  }
+};
+
+template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res)
+  {
+    etor_product_packet_impl<ColMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, innerDim, res);
+    res =  pmadd(lhs.template packet<LoadMode,Packet>(row, Index(UnrollingIndex-1)), pset1<Packet>(rhs.coeff(Index(UnrollingIndex-1), col)), res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, 1, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
+  {
+    res = pmul(pset1<Packet>(lhs.coeff(row, Index(0))),rhs.template packet<LoadMode,Packet>(Index(0), col));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, 1, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res)
+  {
+    res = pmul(lhs.template packet<LoadMode,Packet>(row, Index(0)), pset1<Packet>(rhs.coeff(Index(0), col)));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, 0, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, 0, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<RowMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+    for(Index i = 0; i < innerDim; ++i)
+      res =  pmadd(pset1<Packet>(lhs.coeff(row, i)), rhs.template packet<LoadMode,Packet>(i, col), res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
+struct etor_product_packet_impl<ColMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
+{
+  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res)
+  {
+    res = pset1<Packet>(typename unpacket_traits<Packet>::type(0));
+    for(Index i = 0; i < innerDim; ++i)
+      res =  pmadd(lhs.template packet<LoadMode,Packet>(row, i), pset1<Packet>(rhs.coeff(i, col)), res);
+  }
+};
+
+
+/***************************************************************************
+* Triangular products
+***************************************************************************/
+template<int Mode, bool LhsIsTriangular,
+         typename Lhs, bool LhsIsVector,
+         typename Rhs, bool RhsIsVector>
+struct triangular_product_impl;
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,TriangularShape,DenseShape,ProductTag>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,TriangularShape,DenseShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    triangular_product_impl<Lhs::Mode,true,typename Lhs::MatrixType,false,Rhs, Rhs::ColsAtCompileTime==1>
+        ::run(dst, lhs.nestedExpression(), rhs, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,DenseShape,TriangularShape,ProductTag>
+: generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,TriangularShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    triangular_product_impl<Rhs::Mode,false,Lhs,Lhs::RowsAtCompileTime==1, typename Rhs::MatrixType, false>::run(dst, lhs, rhs.nestedExpression(), alpha);
+  }
+};
+
+
+/***************************************************************************
+* SelfAdjoint products
+***************************************************************************/
+template <typename Lhs, int LhsMode, bool LhsIsVector,
+          typename Rhs, int RhsMode, bool RhsIsVector>
+struct selfadjoint_product_impl;
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,SelfAdjointShape,DenseShape,ProductTag>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,SelfAdjointShape,DenseShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    selfadjoint_product_impl<typename Lhs::MatrixType,Lhs::Mode,false,Rhs,0,Rhs::IsVectorAtCompileTime>::run(dst, lhs.nestedExpression(), rhs, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct generic_product_impl<Lhs,Rhs,DenseShape,SelfAdjointShape,ProductTag>
+: generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,SelfAdjointShape,ProductTag> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    selfadjoint_product_impl<Lhs,0,Lhs::IsVectorAtCompileTime,typename Rhs::MatrixType,Rhs::Mode,false>::run(dst, lhs, rhs.nestedExpression(), alpha);
+  }
+};
+
+
+/***************************************************************************
+* Diagonal products
+***************************************************************************/
+  
+template<typename MatrixType, typename DiagonalType, typename Derived, int ProductOrder>
+struct diagonal_product_evaluator_base
+  : evaluator_base<Derived>
+{
+   typedef typename ScalarBinaryOpTraits<typename MatrixType::Scalar, typename DiagonalType::Scalar>::ReturnType Scalar;
+public:
+  enum {
+    CoeffReadCost = NumTraits<Scalar>::MulCost + evaluator<MatrixType>::CoeffReadCost + evaluator<DiagonalType>::CoeffReadCost,
+    
+    MatrixFlags = evaluator<MatrixType>::Flags,
+    DiagFlags = evaluator<DiagonalType>::Flags,
+    _StorageOrder = MatrixFlags & RowMajorBit ? RowMajor : ColMajor,
+    _ScalarAccessOnDiag =  !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft)
+                           ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)),
+    _SameTypes = is_same<typename MatrixType::Scalar, typename DiagonalType::Scalar>::value,
+    // FIXME currently we need same types, but in the future the next rule should be the one
+    //_Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagFlags)&PacketAccessBit))),
+    _Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagFlags)&PacketAccessBit))),
+    _LinearAccessMask = (MatrixType::RowsAtCompileTime==1 || MatrixType::ColsAtCompileTime==1) ? LinearAccessBit : 0,
+    Flags = ((HereditaryBits|_LinearAccessMask) & (unsigned int)(MatrixFlags)) | (_Vectorizable ? PacketAccessBit : 0),
+    Alignment = evaluator<MatrixType>::Alignment,
+
+    AsScalarProduct =     (DiagonalType::SizeAtCompileTime==1)
+                      ||  (DiagonalType::SizeAtCompileTime==Dynamic && MatrixType::RowsAtCompileTime==1 && ProductOrder==OnTheLeft)
+                      ||  (DiagonalType::SizeAtCompileTime==Dynamic && MatrixType::ColsAtCompileTime==1 && ProductOrder==OnTheRight)
+  };
+  
+  diagonal_product_evaluator_base(const MatrixType &mat, const DiagonalType &diag)
+    : m_diagImpl(diag), m_matImpl(mat)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits<Scalar>::MulCost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const
+  {
+    if(AsScalarProduct)
+      return m_diagImpl.coeff(0) * m_matImpl.coeff(idx);
+    else
+      return m_diagImpl.coeff(idx) * m_matImpl.coeff(idx);
+  }
+  
+protected:
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet_impl(Index row, Index col, Index id, internal::true_type) const
+  {
+    return internal::pmul(m_matImpl.template packet<LoadMode,PacketType>(row, col),
+                          internal::pset1<PacketType>(m_diagImpl.coeff(id)));
+  }
+  
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet_impl(Index row, Index col, Index id, internal::false_type) const
+  {
+    enum {
+      InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime,
+      DiagonalPacketLoadMode = EIGEN_PLAIN_ENUM_MIN(LoadMode,((InnerSize%16) == 0) ? int(Aligned16) : int(evaluator<DiagonalType>::Alignment)) // FIXME hardcoded 16!!
+    };
+    return internal::pmul(m_matImpl.template packet<LoadMode,PacketType>(row, col),
+                          m_diagImpl.template packet<DiagonalPacketLoadMode,PacketType>(id));
+  }
+  
+  evaluator<DiagonalType> m_diagImpl;
+  evaluator<MatrixType>   m_matImpl;
+};
+
+// diagonal * dense
+template<typename Lhs, typename Rhs, int ProductKind, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DiagonalShape, DenseShape>
+  : diagonal_product_evaluator_base<Rhs, typename Lhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheLeft>
+{
+  typedef diagonal_product_evaluator_base<Rhs, typename Lhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheLeft> Base;
+  using Base::m_diagImpl;
+  using Base::m_matImpl;
+  using Base::coeff;
+  typedef typename Base::Scalar Scalar;
+  
+  typedef Product<Lhs, Rhs, ProductKind> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  
+  enum {
+    StorageOrder = int(Rhs::Flags) & RowMajorBit ? RowMajor : ColMajor
+  };
+
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.rhs(), xpr.lhs().diagonal())
+  {
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
+  {
+    return m_diagImpl.coeff(row) * m_matImpl.coeff(row, col);
+  }
+  
+#ifndef __CUDACC__
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const
+  {
+    // FIXME: NVCC used to complain about the template keyword, but we have to check whether this is still the case.
+    // See also similar calls below.
+    return this->template packet_impl<LoadMode,PacketType>(row,col, row,
+                                 typename internal::conditional<int(StorageOrder)==RowMajor, internal::true_type, internal::false_type>::type());
+  }
+  
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index idx) const
+  {
+    return packet<LoadMode,PacketType>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
+  }
+#endif
+};
+
+// dense * diagonal
+template<typename Lhs, typename Rhs, int ProductKind, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, ProductKind>, ProductTag, DenseShape, DiagonalShape>
+  : diagonal_product_evaluator_base<Lhs, typename Rhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheRight>
+{
+  typedef diagonal_product_evaluator_base<Lhs, typename Rhs::DiagonalVectorType, Product<Lhs, Rhs, LazyProduct>, OnTheRight> Base;
+  using Base::m_diagImpl;
+  using Base::m_matImpl;
+  using Base::coeff;
+  typedef typename Base::Scalar Scalar;
+  
+  typedef Product<Lhs, Rhs, ProductKind> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  
+  enum { StorageOrder = int(Lhs::Flags) & RowMajorBit ? RowMajor : ColMajor };
+
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.lhs(), xpr.rhs().diagonal())
+  {
+  }
+  
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
+  {
+    return m_matImpl.coeff(row, col) * m_diagImpl.coeff(col);
+  }
+  
+#ifndef __CUDACC__
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const
+  {
+    return this->template packet_impl<LoadMode,PacketType>(row,col, col,
+                                 typename internal::conditional<int(StorageOrder)==ColMajor, internal::true_type, internal::false_type>::type());
+  }
+  
+  template<int LoadMode,typename PacketType>
+  EIGEN_STRONG_INLINE PacketType packet(Index idx) const
+  {
+    return packet<LoadMode,PacketType>(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx);
+  }
+#endif
+};
+
+/***************************************************************************
+* Products with permutation matrices
+***************************************************************************/
+
+/** \internal
+  * \class permutation_matrix_product
+  * Internal helper class implementing the product between a permutation matrix and a matrix.
+  * This class is specialized for DenseShape below and for SparseShape in SparseCore/SparsePermutation.h
+  */
+template<typename ExpressionType, int Side, bool Transposed, typename ExpressionShape>
+struct permutation_matrix_product;
+
+template<typename ExpressionType, int Side, bool Transposed>
+struct permutation_matrix_product<ExpressionType, Side, Transposed, DenseShape>
+{
+    typedef typename nested_eval<ExpressionType, 1>::type MatrixType;
+    typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
+
+    template<typename Dest, typename PermutationType>
+    static inline void run(Dest& dst, const PermutationType& perm, const ExpressionType& xpr)
+    {
+      MatrixType mat(xpr);
+      const Index n = Side==OnTheLeft ? mat.rows() : mat.cols();
+      // FIXME we need an is_same for expression that is not sensitive to constness. For instance
+      // is_same_xpr<Block<const Matrix>, Block<Matrix> >::value should be true.
+      //if(is_same<MatrixTypeCleaned,Dest>::value && extract_data(dst) == extract_data(mat))
+      if(is_same_dense(dst, mat))
+      {
+        // apply the permutation inplace
+        Matrix<bool,PermutationType::RowsAtCompileTime,1,0,PermutationType::MaxRowsAtCompileTime> mask(perm.size());
+        mask.fill(false);
+        Index r = 0;
+        while(r < perm.size())
+        {
+          // search for the next seed
+          while(r<perm.size() && mask[r]) r++;
+          if(r>=perm.size())
+            break;
+          // we got one, let's follow it until we are back to the seed
+          Index k0 = r++;
+          Index kPrev = k0;
+          mask.coeffRef(k0) = true;
+          for(Index k=perm.indices().coeff(k0); k!=k0; k=perm.indices().coeff(k))
+          {
+                  Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>(dst, k)
+            .swap(Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
+                       (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev));
+
+            mask.coeffRef(k) = true;
+            kPrev = k;
+          }
+        }
+      }
+      else
+      {
+        for(Index i = 0; i < n; ++i)
+        {
+          Block<Dest, Side==OnTheLeft ? 1 : Dest::RowsAtCompileTime, Side==OnTheRight ? 1 : Dest::ColsAtCompileTime>
+               (dst, ((Side==OnTheLeft) ^ Transposed) ? perm.indices().coeff(i) : i)
+
+          =
+
+          Block<const MatrixTypeCleaned,Side==OnTheLeft ? 1 : MatrixTypeCleaned::RowsAtCompileTime,Side==OnTheRight ? 1 : MatrixTypeCleaned::ColsAtCompileTime>
+               (mat, ((Side==OnTheRight) ^ Transposed) ? perm.indices().coeff(i) : i);
+        }
+      }
+    }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, PermutationShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    permutation_matrix_product<Rhs, OnTheLeft, false, MatrixShape>::run(dst, lhs, rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, MatrixShape, PermutationShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    permutation_matrix_product<Lhs, OnTheRight, false, MatrixShape>::run(dst, rhs, lhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Inverse<Lhs>, Rhs, PermutationShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Inverse<Lhs>& lhs, const Rhs& rhs)
+  {
+    permutation_matrix_product<Rhs, OnTheLeft, true, MatrixShape>::run(dst, lhs.nestedExpression(), rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Inverse<Rhs>, MatrixShape, PermutationShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Inverse<Rhs>& rhs)
+  {
+    permutation_matrix_product<Lhs, OnTheRight, true, MatrixShape>::run(dst, rhs.nestedExpression(), lhs);
+  }
+};
+
+
+/***************************************************************************
+* Products with transpositions matrices
+***************************************************************************/
+
+// FIXME could we unify Transpositions and Permutation into a single "shape"??
+
+/** \internal
+  * \class transposition_matrix_product
+  * Internal helper class implementing the product between a permutation matrix and a matrix.
+  */
+template<typename ExpressionType, int Side, bool Transposed, typename ExpressionShape>
+struct transposition_matrix_product
+{
+  typedef typename nested_eval<ExpressionType, 1>::type MatrixType;
+  typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
+  
+  template<typename Dest, typename TranspositionType>
+  static inline void run(Dest& dst, const TranspositionType& tr, const ExpressionType& xpr)
+  {
+    MatrixType mat(xpr);
+    typedef typename TranspositionType::StorageIndex StorageIndex;
+    const Index size = tr.size();
+    StorageIndex j = 0;
+
+    if(!is_same_dense(dst,mat))
+      dst = mat;
+
+    for(Index k=(Transposed?size-1:0) ; Transposed?k>=0:k<size ; Transposed?--k:++k)
+      if(Index(j=tr.coeff(k))!=k)
+      {
+        if(Side==OnTheLeft)        dst.row(k).swap(dst.row(j));
+        else if(Side==OnTheRight)  dst.col(k).swap(dst.col(j));
+      }
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, TranspositionsShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    transposition_matrix_product<Rhs, OnTheLeft, false, MatrixShape>::run(dst, lhs, rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Rhs, MatrixShape, TranspositionsShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    transposition_matrix_product<Lhs, OnTheRight, false, MatrixShape>::run(dst, rhs, lhs);
+  }
+};
+
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Transpose<Lhs>, Rhs, TranspositionsShape, MatrixShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Transpose<Lhs>& lhs, const Rhs& rhs)
+  {
+    transposition_matrix_product<Rhs, OnTheLeft, true, MatrixShape>::run(dst, lhs.nestedExpression(), rhs);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductTag, typename MatrixShape>
+struct generic_product_impl<Lhs, Transpose<Rhs>, MatrixShape, TranspositionsShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Transpose<Rhs>& rhs)
+  {
+    transposition_matrix_product<Lhs, OnTheRight, true, MatrixShape>::run(dst, rhs.nestedExpression(), lhs);
+  }
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PRODUCT_EVALUATORS_H
diff --git a/Eigen/src/Core/Random.h b/Eigen/src/Core/Random.h
index 480fea4..6faf789 100644
--- a/Eigen/src/Core/Random.h
+++ b/Eigen/src/Core/Random.h
@@ -16,8 +16,7 @@
 
 template<typename Scalar> struct scalar_random_op {
   EIGEN_EMPTY_STRUCT_CTOR(scalar_random_op)
-  template<typename Index>
-  inline const Scalar operator() (Index, Index = 0) const { return random<Scalar>(); }
+  inline const Scalar operator() () const { return random<Scalar>(); }
 };
 
 template<typename Scalar>
@@ -28,12 +27,18 @@
 
 /** \returns a random matrix expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * The parameters \a rows and \a cols are the number of rows and of columns of
   * the returned matrix. Must be compatible with this MatrixBase type.
   *
+  * \not_reentrant
+  * 
   * This variant is meant to be used for dynamic-size matrix types. For fixed-size types,
   * it is redundant to pass \a rows and \a cols as arguments, so Random() should be used
   * instead.
+  * 
   *
   * Example: \include MatrixBase_random_int_int.cpp
   * Output: \verbinclude MatrixBase_random_int_int.out
@@ -41,11 +46,13 @@
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
+  * 
+  * See DenseBase::NullaryExpr(Index, const CustomNullaryOp&) for an example using C++11 random generators.
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index), MatrixBase::Random()
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index), DenseBase::Random()
   */
 template<typename Derived>
-inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random(Index rows, Index cols)
 {
   return NullaryExpr(rows, cols, internal::scalar_random_op<Scalar>());
@@ -53,10 +60,14 @@
 
 /** \returns a random vector expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  *
   * The parameter \a size is the size of the returned vector.
   * Must be compatible with this MatrixBase type.
   *
   * \only_for_vectors
+  * \not_reentrant
   *
   * This variant is meant to be used for dynamic-size vector types. For fixed-size types,
   * it is redundant to pass \a size as argument, so Random() should be used
@@ -69,10 +80,10 @@
   * a temporary vector whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random()
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random()
   */
 template<typename Derived>
-inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random(Index size)
 {
   return NullaryExpr(size, internal::scalar_random_op<Scalar>());
@@ -80,6 +91,9 @@
 
 /** \returns a fixed-size random matrix or vector expression
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you
   * need to use the variants taking size arguments.
   *
@@ -89,11 +103,13 @@
   * This expression has the "evaluate before nesting" flag so that it will be evaluated into
   * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected
   * behavior with expressions involving random matrices.
+  * 
+  * \not_reentrant
   *
-  * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random(Index)
+  * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random(Index)
   */
 template<typename Derived>
-inline const CwiseNullaryOp<internal::scalar_random_op<typename internal::traits<Derived>::Scalar>, Derived>
+inline const typename DenseBase<Derived>::RandomReturnType
 DenseBase<Derived>::Random()
 {
   return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_random_op<Scalar>());
@@ -101,6 +117,11 @@
 
 /** Sets all coefficients in this expression to random values.
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
+  * \not_reentrant
+  * 
   * Example: \include MatrixBase_setRandom.cpp
   * Output: \verbinclude MatrixBase_setRandom.out
   *
@@ -114,12 +135,16 @@
 
 /** Resizes to the given \a newSize, and sets all coefficients in this expression to random values.
   *
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  * 
   * \only_for_vectors
+  * \not_reentrant
   *
   * Example: \include Matrix_setRandom_int.cpp
   * Output: \verbinclude Matrix_setRandom_int.out
   *
-  * \sa MatrixBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, MatrixBase::Random()
+  * \sa DenseBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, DenseBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
@@ -131,19 +156,24 @@
 
 /** Resizes to the given size, and sets all coefficients in this expression to random values.
   *
-  * \param nbRows the new number of rows
-  * \param nbCols the new number of columns
+  * Numbers are uniformly spread through their whole definition range for integer types,
+  * and in the [-1:1] range for floating point scalar types.
+  *
+  * \not_reentrant
+  * 
+  * \param rows the new number of rows
+  * \param cols the new number of columns
   *
   * Example: \include Matrix_setRandom_int_int.cpp
   * Output: \verbinclude Matrix_setRandom_int_int.out
   *
-  * \sa MatrixBase::setRandom(), setRandom(Index), class CwiseNullaryOp, MatrixBase::Random()
+  * \sa DenseBase::setRandom(), setRandom(Index), class CwiseNullaryOp, DenseBase::Random()
   */
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
-PlainObjectBase<Derived>::setRandom(Index nbRows, Index nbCols)
+PlainObjectBase<Derived>::setRandom(Index rows, Index cols)
 {
-  resize(nbRows, nbCols);
+  resize(rows, cols);
   return setRandom();
 }
 
diff --git a/Eigen/src/Core/Redux.h b/Eigen/src/Core/Redux.h
index 50548fa..760e9f8 100644
--- a/Eigen/src/Core/Redux.h
+++ b/Eigen/src/Core/Redux.h
@@ -27,8 +27,9 @@
 struct redux_traits
 {
 public:
+    typedef typename find_best_packet<typename Derived::Scalar,Derived::SizeAtCompileTime>::type PacketType;
   enum {
-    PacketSize = packet_traits<typename Derived::Scalar>::size,
+    PacketSize = unpacket_traits<PacketType>::size,
     InnerMaxSize = int(Derived::IsRowMajor)
                  ? Derived::MaxColsAtCompileTime
                  : Derived::MaxRowsAtCompileTime
@@ -37,8 +38,8 @@
   enum {
     MightVectorize = (int(Derived::Flags)&ActualPacketAccessBit)
                   && (functor_traits<Func>::PacketAccess),
-    MayLinearVectorize = MightVectorize && (int(Derived::Flags)&LinearAccessBit),
-    MaySliceVectorize  = MightVectorize && int(InnerMaxSize)>=3*PacketSize
+    MayLinearVectorize = bool(MightVectorize) && (int(Derived::Flags)&LinearAccessBit),
+    MaySliceVectorize  = bool(MightVectorize) && int(InnerMaxSize)>=3*PacketSize
   };
 
 public:
@@ -50,21 +51,34 @@
 
 public:
   enum {
-    Cost = (  Derived::SizeAtCompileTime == Dynamic
-           || Derived::CoeffReadCost == Dynamic
-           || (Derived::SizeAtCompileTime!=1 && functor_traits<Func>::Cost == Dynamic)
-           ) ? Dynamic
-           : Derived::SizeAtCompileTime * Derived::CoeffReadCost
-               + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
+    Cost = Derived::SizeAtCompileTime == Dynamic ? HugeCost
+         : Derived::SizeAtCompileTime * Derived::CoeffReadCost + (Derived::SizeAtCompileTime-1) * functor_traits<Func>::Cost,
     UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Traversal) == int(DefaultTraversal) ? 1 : int(PacketSize))
   };
 
 public:
   enum {
-    Unrolling = Cost != Dynamic && Cost <= UnrollingLimit
-              ? CompleteUnrolling
-              : NoUnrolling
+    Unrolling = Cost <= UnrollingLimit ? CompleteUnrolling : NoUnrolling
   };
+  
+#ifdef EIGEN_DEBUG_ASSIGN
+  static void debug()
+  {
+    std::cerr << "Xpr: " << typeid(typename Derived::XprType).name() << std::endl;
+    std::cerr.setf(std::ios::hex, std::ios::basefield);
+    EIGEN_DEBUG_VAR(Derived::Flags)
+    std::cerr.unsetf(std::ios::hex);
+    EIGEN_DEBUG_VAR(InnerMaxSize)
+    EIGEN_DEBUG_VAR(PacketSize)
+    EIGEN_DEBUG_VAR(MightVectorize)
+    EIGEN_DEBUG_VAR(MayLinearVectorize)
+    EIGEN_DEBUG_VAR(MaySliceVectorize)
+    EIGEN_DEBUG_VAR(Traversal)
+    EIGEN_DEBUG_VAR(UnrollingLimit)
+    EIGEN_DEBUG_VAR(Unrolling)
+    std::cerr << std::endl;
+  }
+#endif
 };
 
 /***************************************************************************
@@ -82,6 +96,7 @@
 
   typedef typename Derived::Scalar Scalar;
 
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
     return func(redux_novec_unroller<Func, Derived, Start, HalfLength>::run(mat,func),
@@ -99,6 +114,7 @@
 
   typedef typename Derived::Scalar Scalar;
 
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func&)
   {
     return mat.coeffByOuterInner(outer, inner);
@@ -112,6 +128,7 @@
 struct redux_novec_unroller<Func, Derived, Start, 0>
 {
   typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC 
   static EIGEN_STRONG_INLINE Scalar run(const Derived&, const Func&) { return Scalar(); }
 };
 
@@ -121,12 +138,12 @@
 struct redux_vec_unroller
 {
   enum {
-    PacketSize = packet_traits<typename Derived::Scalar>::size,
+    PacketSize = redux_traits<Func, Derived>::PacketSize,
     HalfLength = Length/2
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
 
   static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func& func)
   {
@@ -140,18 +157,18 @@
 struct redux_vec_unroller<Func, Derived, Start, 1>
 {
   enum {
-    index = Start * packet_traits<typename Derived::Scalar>::size,
+    index = Start * redux_traits<Func, Derived>::PacketSize,
     outer = index / int(Derived::InnerSizeAtCompileTime),
     inner = index % int(Derived::InnerSizeAtCompileTime),
-    alignment = (Derived::Flags & AlignedBit) ? Aligned : Unaligned
+    alignment = Derived::Alignment
   };
 
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
 
   static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func&)
   {
-    return mat.template packetByOuterInner<alignment>(outer, inner);
+    return mat.template packetByOuterInner<alignment,PacketScalar>(outer, inner);
   }
 };
 
@@ -169,8 +186,8 @@
 struct redux_impl<Func, Derived, DefaultTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename Derived::Index Index;
-  static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC
+  static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     Scalar res;
@@ -193,19 +210,19 @@
 struct redux_impl<Func, Derived, LinearVectorizedTraversal, NoUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
-  typedef typename Derived::Index Index;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
 
-  static Scalar run(const Derived& mat, const Func& func)
+  static Scalar run(const Derived &mat, const Func& func)
   {
     const Index size = mat.size();
-    eigen_assert(size && "you are using an empty matrix");
-    const Index packetSize = packet_traits<Scalar>::size;
-    const Index alignedStart = internal::first_aligned(mat);
+    
+    const Index packetSize = redux_traits<Func, Derived>::PacketSize;
+    const int packetAlignment = unpacket_traits<PacketScalar>::alignment;
     enum {
-      alignment = bool(Derived::Flags & DirectAccessBit) || bool(Derived::Flags & AlignedBit)
-                ? Aligned : Unaligned
+      alignment0 = (bool(Derived::Flags & DirectAccessBit) && bool(packet_traits<Scalar>::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned),
+      alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Derived::Alignment)
     };
+    const Index alignedStart = internal::first_default_aligned(mat.nestedExpression());
     const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize);
     const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize);
     const Index alignedEnd2 = alignedStart + alignedSize2;
@@ -213,19 +230,19 @@
     Scalar res;
     if(alignedSize)
     {
-      PacketScalar packet_res0 = mat.template packet<alignment>(alignedStart);
+      PacketScalar packet_res0 = mat.template packet<alignment,PacketScalar>(alignedStart);
       if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop
       {
-        PacketScalar packet_res1 = mat.template packet<alignment>(alignedStart+packetSize);
+        PacketScalar packet_res1 = mat.template packet<alignment,PacketScalar>(alignedStart+packetSize);
         for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize)
         {
-          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(index));
-          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment>(index+packetSize));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(index));
+          packet_res1 = func.packetOp(packet_res1, mat.template packet<alignment,PacketScalar>(index+packetSize));
         }
 
         packet_res0 = func.packetOp(packet_res0,packet_res1);
         if(alignedEnd>alignedEnd2)
-          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment>(alignedEnd2));
+          packet_res0 = func.packetOp(packet_res0, mat.template packet<alignment,PacketScalar>(alignedEnd2));
       }
       res = func.predux(packet_res0);
 
@@ -247,29 +264,29 @@
   }
 };
 
-template<typename Func, typename Derived>
-struct redux_impl<Func, Derived, SliceVectorizedTraversal, NoUnrolling>
+// NOTE: for SliceVectorizedTraversal we simply bypass unrolling
+template<typename Func, typename Derived, int Unrolling>
+struct redux_impl<Func, Derived, SliceVectorizedTraversal, Unrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
-  typedef typename Derived::Index Index;
+  typedef typename redux_traits<Func, Derived>::PacketType PacketType;
 
-  static Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC static Scalar run(const Derived &mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
     const Index innerSize = mat.innerSize();
     const Index outerSize = mat.outerSize();
     enum {
-      packetSize = packet_traits<Scalar>::size
+      packetSize = redux_traits<Func, Derived>::PacketSize
     };
     const Index packetedInnerSize = ((innerSize)/packetSize)*packetSize;
     Scalar res;
     if(packetedInnerSize)
     {
-      PacketScalar packet_res = mat.template packet<Unaligned>(0,0);
+      PacketType packet_res = mat.template packet<Unaligned,PacketType>(0,0);
       for(Index j=0; j<outerSize; ++j)
         for(Index i=(j==0?packetSize:0); i<packetedInnerSize; i+=Index(packetSize))
-          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned>(j,i));
+          packet_res = func.packetOp(packet_res, mat.template packetByOuterInner<Unaligned,PacketType>(j,i));
 
       res = func.predux(packet_res);
       for(Index j=0; j<outerSize; ++j)
@@ -290,22 +307,90 @@
 struct redux_impl<Func, Derived, LinearVectorizedTraversal, CompleteUnrolling>
 {
   typedef typename Derived::Scalar Scalar;
-  typedef typename packet_traits<Scalar>::type PacketScalar;
+
+  typedef typename redux_traits<Func, Derived>::PacketType PacketScalar;
   enum {
-    PacketSize = packet_traits<Scalar>::size,
+    PacketSize = redux_traits<Func, Derived>::PacketSize,
     Size = Derived::SizeAtCompileTime,
     VectorizedSize = (Size / PacketSize) * PacketSize
   };
-  static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func)
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func)
   {
     eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix");
-    Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
-    if (VectorizedSize != Size)
-      res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
-    return res;
+    if (VectorizedSize > 0) {
+      Scalar res = func.predux(redux_vec_unroller<Func, Derived, 0, Size / PacketSize>::run(mat,func));
+      if (VectorizedSize != Size)
+        res = func(res,redux_novec_unroller<Func, Derived, VectorizedSize, Size-VectorizedSize>::run(mat,func));
+      return res;
+    }
+    else {
+      return redux_novec_unroller<Func, Derived, 0, Size>::run(mat,func);
+    }
   }
 };
 
+// evaluator adaptor
+template<typename _XprType>
+class redux_evaluator
+{
+public:
+  typedef _XprType XprType;
+  EIGEN_DEVICE_FUNC explicit redux_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
+  
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  typedef typename XprType::PacketScalar PacketScalar;
+  typedef typename XprType::PacketReturnType PacketReturnType;
+  
+  enum {
+    MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = XprType::MaxColsAtCompileTime,
+    // TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator
+    Flags = evaluator<XprType>::Flags & ~DirectAccessBit,
+    IsRowMajor = XprType::IsRowMajor,
+    SizeAtCompileTime = XprType::SizeAtCompileTime,
+    InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime,
+    CoeffReadCost = evaluator<XprType>::CoeffReadCost,
+    Alignment = evaluator<XprType>::Alignment
+  };
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
+  EIGEN_DEVICE_FUNC Index innerSize() const { return m_xpr.innerSize(); }
+  EIGEN_DEVICE_FUNC Index outerSize() const { return m_xpr.outerSize(); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index row, Index col) const
+  { return m_evaluator.coeff(row, col); }
+
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeff(Index index) const
+  { return m_evaluator.coeff(index); }
+
+  template<int LoadMode, typename PacketType>
+  PacketType packet(Index row, Index col) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(row, col); }
+
+  template<int LoadMode, typename PacketType>
+  PacketType packet(Index index) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(index); }
+  
+  EIGEN_DEVICE_FUNC
+  CoeffReturnType coeffByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+  
+  template<int LoadMode, typename PacketType>
+  PacketType packetByOuterInner(Index outer, Index inner) const
+  { return m_evaluator.template packet<LoadMode,PacketType>(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); }
+  
+  const XprType & nestedExpression() const { return m_xpr; }
+  
+protected:
+  internal::evaluator<XprType> m_evaluator;
+  const XprType &m_xpr;
+};
+
 } // end namespace internal
 
 /***************************************************************************
@@ -316,18 +401,21 @@
 /** \returns the result of a full redux operation on the whole matrix or vector using \a func
   *
   * The template parameter \a BinaryOp is the type of the functor \a func which must be
-  * an associative operator. Both current STL and TR1 functor styles are handled.
+  * an associative operator. Both current C++98 and C++11 functor styles are handled.
   *
   * \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise()
   */
 template<typename Derived>
 template<typename Func>
-EIGEN_STRONG_INLINE typename internal::result_of<Func(typename internal::traits<Derived>::Scalar)>::type
+EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::redux(const Func& func) const
 {
-  typedef typename internal::remove_all<typename Derived::Nested>::type ThisNested;
-  return internal::redux_impl<Func, ThisNested>
-            ::run(derived(), func);
+  eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix");
+
+  typedef typename internal::redux_evaluator<Derived> ThisEvaluator;
+  ThisEvaluator thisEval(derived());
+  
+  return internal::redux_impl<Func, ThisEvaluator>::run(thisEval, func);
 }
 
 /** \returns the minimum of all coefficients of \c *this.
@@ -337,7 +425,7 @@
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff() const
 {
-  return this->redux(Eigen::internal::scalar_min_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_min_op<Scalar,Scalar>());
 }
 
 /** \returns the maximum of all coefficients of \c *this.
@@ -347,10 +435,12 @@
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff() const
 {
-  return this->redux(Eigen::internal::scalar_max_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_max_op<Scalar,Scalar>());
 }
 
-/** \returns the sum of all coefficients of *this
+/** \returns the sum of all coefficients of \c *this
+  *
+  * If \c *this is empty, then the value 0 is returned.
   *
   * \sa trace(), prod(), mean()
   */
@@ -360,7 +450,7 @@
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(0);
-  return this->redux(Eigen::internal::scalar_sum_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>());
 }
 
 /** \returns the mean of all coefficients of *this
@@ -371,7 +461,14 @@
 EIGEN_STRONG_INLINE typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::mean() const
 {
-  return Scalar(this->redux(Eigen::internal::scalar_sum_op<Scalar>())) / Scalar(this->size());
+#ifdef __INTEL_COMPILER
+  #pragma warning push
+  #pragma warning ( disable : 2259 )
+#endif
+  return Scalar(derived().redux(Eigen::internal::scalar_sum_op<Scalar,Scalar>())) / Scalar(this->size());
+#ifdef __INTEL_COMPILER
+  #pragma warning pop
+#endif
 }
 
 /** \returns the product of all coefficients of *this
@@ -387,7 +484,7 @@
 {
   if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0))
     return Scalar(1);
-  return this->redux(Eigen::internal::scalar_product_op<Scalar>());
+  return derived().redux(Eigen::internal::scalar_product_op<Scalar>());
 }
 
 /** \returns the trace of \c *this, i.e. the sum of the coefficients on the main diagonal.
diff --git a/Eigen/src/Core/Ref.h b/Eigen/src/Core/Ref.h
index f53674c..9c6e3c5 100644
--- a/Eigen/src/Core/Ref.h
+++ b/Eigen/src/Core/Ref.h
@@ -12,79 +12,6 @@
 
 namespace Eigen { 
 
-template<typename Derived> class RefBase;
-template<typename PlainObjectType, int Options = 0,
-         typename StrideType = typename internal::conditional<PlainObjectType::IsVectorAtCompileTime,InnerStride<1>,OuterStride<> >::type > class Ref;
-
-/** \class Ref
-  * \ingroup Core_Module
-  *
-  * \brief A matrix or vector expression mapping an existing expressions
-  *
-  * \tparam PlainObjectType the equivalent matrix type of the mapped data
-  * \tparam Options specifies whether the pointer is \c #Aligned, or \c #Unaligned.
-  *                The default is \c #Unaligned.
-  * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
-  *                   but accept a variable outer stride (leading dimension).
-  *                   This can be overridden by specifying strides.
-  *                   The type passed here must be a specialization of the Stride template, see examples below.
-  *
-  * This class permits to write non template functions taking Eigen's object as parameters while limiting the number of copies.
-  * A Ref<> object can represent either a const expression or a l-value:
-  * \code
-  * // in-out argument:
-  * void foo1(Ref<VectorXf> x);
-  *
-  * // read-only const argument:
-  * void foo2(const Ref<const VectorXf>& x);
-  * \endcode
-  *
-  * In the in-out case, the input argument must satisfies the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
-  * By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
-  * Likewise, a Ref<MatrixXf> can reference any column major dense matrix expression of float whose column's elements are contiguously stored with
-  * the possibility to have a constant space inbetween each column, i.e.: the inner stride mmust be equal to 1, but the outer-stride (or leading dimension),
-  * can be greater than the number of rows.
-  *
-  * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
-  * Here are some examples:
-  * \code
-  * MatrixXf A;
-  * VectorXf a;
-  * foo1(a.head());             // OK
-  * foo1(A.col());              // OK
-  * foo1(A.row());              // compilation error because here innerstride!=1
-  * foo2(A.row());              // The row is copied into a contiguous temporary
-  * foo2(2*a);                  // The expression is evaluated into a temporary
-  * foo2(A.col().segment(2,4)); // No temporary
-  * \endcode
-  *
-  * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameter.
-  * Here is an example accepting an innerstride!=1:
-  * \code
-  * // in-out argument:
-  * void foo3(Ref<VectorXf,0,InnerStride<> > x);
-  * foo3(A.row());              // OK
-  * \endcode
-  * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involved more
-  * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overloads internally calling a
-  * template function, e.g.:
-  * \code
-  * // in the .h:
-  * void foo(const Ref<MatrixXf>& A);
-  * void foo(const Ref<MatrixXf,0,Stride<> >& A);
-  *
-  * // in the .cpp:
-  * template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
-  *     ... // crazy code goes here
-  * }
-  * void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
-  * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
-  * \endcode
-  *
-  *
-  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
-  */
-
 namespace internal {
 
 template<typename _PlainObjectType, int _Options, typename _StrideType>
@@ -95,7 +22,8 @@
   typedef _StrideType StrideType;
   enum {
     Options = _Options,
-    Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit
+    Flags = traits<Map<_PlainObjectType, _Options, _StrideType> >::Flags | NestByRefBit,
+    Alignment = traits<Map<_PlainObjectType, _Options, _StrideType> >::Alignment
   };
 
   template<typename Derived> struct match {
@@ -107,7 +35,13 @@
                       || (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1),
       OuterStrideMatch = Derived::IsVectorAtCompileTime
                       || int(StrideType::OuterStrideAtCompileTime)==int(Dynamic) || int(StrideType::OuterStrideAtCompileTime)==int(Derived::OuterStrideAtCompileTime),
-      AlignmentMatch = (_Options!=Aligned) || ((PlainObjectType::Flags&AlignedBit)==0) || ((traits<Derived>::Flags&AlignedBit)==AlignedBit),
+      // NOTE, this indirection of evaluator<Derived>::Alignment is needed
+      // to workaround a very strange bug in MSVC related to the instantiation
+      // of has_*ary_operator in evaluator<CwiseNullaryOp>.
+      // This line is surprisingly very sensitive. For instance, simply adding parenthesis
+      // as "DerivedAlignment = (int(evaluator<Derived>::Alignment))," will make MSVC fail...
+      DerivedAlignment = int(evaluator<Derived>::Alignment),
+      AlignmentMatch = (int(traits<PlainObjectType>::Alignment)==int(Unaligned)) || (DerivedAlignment >= int(Alignment)), // FIXME the first condition is not very clear, it should be replaced by the required alignment
       ScalarTypeMatch = internal::is_same<typename PlainObjectType::Scalar, typename Derived::Scalar>::value,
       MatchAtCompileTime = HasDirectAccess && StorageOrderMatch && InnerStrideMatch && OuterStrideMatch && AlignmentMatch && ScalarTypeMatch
     };
@@ -132,12 +66,12 @@
   typedef MapBase<Derived> Base;
   EIGEN_DENSE_PUBLIC_INTERFACE(RefBase)
 
-  inline Index innerStride() const
+  EIGEN_DEVICE_FUNC inline Index innerStride() const
   {
     return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1;
   }
 
-  inline Index outerStride() const
+  EIGEN_DEVICE_FUNC inline Index outerStride() const
   {
     return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer()
          : IsVectorAtCompileTime ? this->size()
@@ -145,7 +79,7 @@
          : this->rows();
   }
 
-  RefBase()
+  EIGEN_DEVICE_FUNC RefBase()
     : Base(0,RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime),
       // Stride<> does not allow default ctor for Dynamic strides, so let' initialize it with dummy values:
       m_stride(StrideType::OuterStrideAtCompileTime==Dynamic?0:StrideType::OuterStrideAtCompileTime,
@@ -159,8 +93,10 @@
   typedef Stride<StrideType::OuterStrideAtCompileTime,StrideType::InnerStrideAtCompileTime> StrideBase;
 
   template<typename Expression>
-  void construct(Expression& expr)
+  EIGEN_DEVICE_FUNC void construct(Expression& expr)
   {
+    EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(PlainObjectType,Expression);
+
     if(PlainObjectType::RowsAtCompileTime==1)
     {
       eigen_assert(expr.rows()==1 || expr.cols()==1);
@@ -184,15 +120,83 @@
   StrideBase m_stride;
 };
 
-
+/** \class Ref
+  * \ingroup Core_Module
+  *
+  * \brief A matrix or vector expression mapping an existing expression
+  *
+  * \tparam PlainObjectType the equivalent matrix type of the mapped data
+  * \tparam Options specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned.
+  *                 The default is \c #Unaligned.
+  * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1),
+  *                   but accepts a variable outer stride (leading dimension).
+  *                   This can be overridden by specifying strides.
+  *                   The type passed here must be a specialization of the Stride template, see examples below.
+  *
+  * This class provides a way to write non-template functions taking Eigen objects as parameters while limiting the number of copies.
+  * A Ref<> object can represent either a const expression or a l-value:
+  * \code
+  * // in-out argument:
+  * void foo1(Ref<VectorXf> x);
+  *
+  * // read-only const argument:
+  * void foo2(const Ref<const VectorXf>& x);
+  * \endcode
+  *
+  * In the in-out case, the input argument must satisfy the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered.
+  * By default, a Ref<VectorXf> can reference any dense vector expression of float having a contiguous memory layout.
+  * Likewise, a Ref<MatrixXf> can reference any column-major dense matrix expression of float whose column's elements are contiguously stored with
+  * the possibility to have a constant space in-between each column, i.e. the inner stride must be equal to 1, but the outer stride (or leading dimension)
+  * can be greater than the number of rows.
+  *
+  * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function.
+  * Here are some examples:
+  * \code
+  * MatrixXf A;
+  * VectorXf a;
+  * foo1(a.head());             // OK
+  * foo1(A.col());              // OK
+  * foo1(A.row());              // Compilation error because here innerstride!=1
+  * foo2(A.row());              // Compilation error because A.row() is a 1xN object while foo2 is expecting a Nx1 object
+  * foo2(A.row().transpose());  // The row is copied into a contiguous temporary
+  * foo2(2*a);                  // The expression is evaluated into a temporary
+  * foo2(A.col().segment(2,4)); // No temporary
+  * \endcode
+  *
+  * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameters.
+  * Here is an example accepting an innerstride!=1:
+  * \code
+  * // in-out argument:
+  * void foo3(Ref<VectorXf,0,InnerStride<> > x);
+  * foo3(A.row());              // OK
+  * \endcode
+  * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involve more
+  * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overload internally calling a
+  * template function, e.g.:
+  * \code
+  * // in the .h:
+  * void foo(const Ref<MatrixXf>& A);
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A);
+  *
+  * // in the .cpp:
+  * template<typename TypeOfA> void foo_impl(const TypeOfA& A) {
+  *     ... // crazy code goes here
+  * }
+  * void foo(const Ref<MatrixXf>& A) { foo_impl(A); }
+  * void foo(const Ref<MatrixXf,0,Stride<> >& A) { foo_impl(A); }
+  * \endcode
+  *
+  *
+  * \sa PlainObjectBase::Map(), \ref TopicStorageOrders
+  */
 template<typename PlainObjectType, int Options, typename StrideType> class Ref
   : public RefBase<Ref<PlainObjectType, Options, StrideType> >
 {
   private:
     typedef internal::traits<Ref> Traits;
     template<typename Derived>
-    inline Ref(const PlainObjectBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
+    EIGEN_DEVICE_FUNC inline Ref(const PlainObjectBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0);
   public:
 
     typedef RefBase<Ref> Base;
@@ -201,23 +205,24 @@
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename Derived>
-    inline Ref(PlainObjectBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    EIGEN_DEVICE_FUNC inline Ref(PlainObjectBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
     {
-      EIGEN_STATIC_ASSERT(static_cast<bool>(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
       Base::construct(expr.derived());
     }
     template<typename Derived>
-    inline Ref(const DenseBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
+    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::MatchAtCompileTime),Derived>::type* = 0)
     #else
+    /** Implicit constructor from any dense expression */
     template<typename Derived>
     inline Ref(DenseBase<Derived>& expr)
     #endif
     {
-      EIGEN_STATIC_ASSERT(static_cast<bool>(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
-      EIGEN_STATIC_ASSERT(static_cast<bool>(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
-      enum { THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY = Derived::ThisConstantIsPrivateInPlainObjectBase};
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      EIGEN_STATIC_ASSERT(!Derived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
       Base::construct(expr.const_cast_derived());
     }
 
@@ -236,8 +241,8 @@
     EIGEN_DENSE_PUBLIC_INTERFACE(Ref)
 
     template<typename Derived>
-    inline Ref(const DenseBase<Derived>& expr,
-               typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
+    EIGEN_DEVICE_FUNC inline Ref(const DenseBase<Derived>& expr,
+                                 typename internal::enable_if<bool(Traits::template match<Derived>::ScalarTypeMatch),Derived>::type* = 0)
     {
 //      std::cout << match_helper<Derived>::HasDirectAccess << "," << match_helper<Derived>::OuterStrideMatch << "," << match_helper<Derived>::InnerStrideMatch << "\n";
 //      std::cout << int(StrideType::OuterStrideAtCompileTime) << " - " << int(Derived::OuterStrideAtCompileTime) << "\n";
@@ -245,18 +250,27 @@
       construct(expr.derived(), typename Traits::template match<Derived>::type());
     }
 
+    EIGEN_DEVICE_FUNC inline Ref(const Ref& other) : Base(other) {
+      // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
+    }
+
+    template<typename OtherRef>
+    EIGEN_DEVICE_FUNC inline Ref(const RefBase<OtherRef>& other) {
+      construct(other.derived(), typename Traits::template match<OtherRef>::type());
+    }
+
   protected:
 
     template<typename Expression>
-    void construct(const Expression& expr,internal::true_type)
+    EIGEN_DEVICE_FUNC void construct(const Expression& expr,internal::true_type)
     {
       Base::construct(expr);
     }
 
     template<typename Expression>
-    void construct(const Expression& expr, internal::false_type)
+    EIGEN_DEVICE_FUNC void construct(const Expression& expr, internal::false_type)
     {
-      m_object.lazyAssign(expr);
+      internal::call_assignment_no_alias(m_object,expr,internal::assign_op<Scalar,Scalar>());
       Base::construct(m_object);
     }
 
diff --git a/Eigen/src/Core/Replicate.h b/Eigen/src/Core/Replicate.h
index ac4537c..9960ef8 100644
--- a/Eigen/src/Core/Replicate.h
+++ b/Eigen/src/Core/Replicate.h
@@ -12,21 +12,6 @@
 
 namespace Eigen { 
 
-/**
-  * \class Replicate
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the multiple replication of a matrix or vector
-  *
-  * \param MatrixType the type of the object we are replicating
-  *
-  * This class represents an expression of the multiple replication of a matrix or vector.
-  * It is the return type of DenseBase::replicate() and most of the time
-  * this is the only way it is used.
-  *
-  * \sa DenseBase::replicate()
-  */
-
 namespace internal {
 template<typename MatrixType,int RowFactor,int ColFactor>
 struct traits<Replicate<MatrixType,RowFactor,ColFactor> >
@@ -35,10 +20,7 @@
   typedef typename MatrixType::Scalar Scalar;
   typedef typename traits<MatrixType>::StorageKind StorageKind;
   typedef typename traits<MatrixType>::XprKind XprKind;
-  enum {
-    Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor
-  };
-  typedef typename nested<MatrixType,Factor>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = RowFactor==Dynamic || int(MatrixType::RowsAtCompileTime)==Dynamic
@@ -53,12 +35,29 @@
     IsRowMajor = MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1 ? 1
                : MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1 ? 0
                : (MatrixType::Flags & RowMajorBit) ? 1 : 0,
-    Flags = (_MatrixTypeNested::Flags & HereditaryBits & ~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0),
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
+    
+    // FIXME enable DirectAccess with negative strides?
+    Flags = IsRowMajor ? RowMajorBit : 0
   };
 };
 }
 
+/**
+  * \class Replicate
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the multiple replication of a matrix or vector
+  *
+  * \tparam MatrixType the type of the object we are replicating
+  * \tparam RowFactor number of repetitions at compile time along the vertical direction, can be Dynamic.
+  * \tparam ColFactor number of repetitions at compile time along the horizontal direction, can be Dynamic.
+  *
+  * This class represents an expression of the multiple replication of a matrix or vector.
+  * It is the return type of DenseBase::replicate() and most of the time
+  * this is the only way it is used.
+  *
+  * \sa DenseBase::replicate()
+  */
 template<typename MatrixType,int RowFactor,int ColFactor> class Replicate
   : public internal::dense_xpr_base< Replicate<MatrixType,RowFactor,ColFactor> >::type
 {
@@ -68,10 +67,12 @@
 
     typedef typename internal::dense_xpr_base<Replicate>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Replicate)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
     template<typename OriginalMatrixType>
-    inline explicit Replicate(const OriginalMatrixType& a_matrix)
-      : m_matrix(a_matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
+    EIGEN_DEVICE_FUNC
+    inline explicit Replicate(const OriginalMatrixType& matrix)
+      : m_matrix(matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor)
     {
       EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
@@ -79,41 +80,20 @@
     }
 
     template<typename OriginalMatrixType>
-    inline Replicate(const OriginalMatrixType& a_matrix, Index rowFactor, Index colFactor)
-      : m_matrix(a_matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
+    EIGEN_DEVICE_FUNC
+    inline Replicate(const OriginalMatrixType& matrix, Index rowFactor, Index colFactor)
+      : m_matrix(matrix), m_rowFactor(rowFactor), m_colFactor(colFactor)
     {
       EIGEN_STATIC_ASSERT((internal::is_same<typename internal::remove_const<MatrixType>::type,OriginalMatrixType>::value),
                           THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE)
     }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); }
 
-    inline Scalar coeff(Index rowId, Index colId) const
-    {
-      // try to avoid using modulo; this is a pure optimization strategy
-      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? rowId
-                            : rowId%m_matrix.rows();
-      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? colId
-                            : colId%m_matrix.cols();
-
-      return m_matrix.coeff(actual_row, actual_col);
-    }
-    template<int LoadMode>
-    inline PacketScalar packet(Index rowId, Index colId) const
-    {
-      const Index actual_row  = internal::traits<MatrixType>::RowsAtCompileTime==1 ? 0
-                            : RowFactor==1 ? rowId
-                            : rowId%m_matrix.rows();
-      const Index actual_col  = internal::traits<MatrixType>::ColsAtCompileTime==1 ? 0
-                            : ColFactor==1 ? colId
-                            : colId%m_matrix.cols();
-
-      return m_matrix.template packet<LoadMode>(actual_row, actual_col);
-    }
-
+    EIGEN_DEVICE_FUNC
     const _MatrixTypeNested& nestedExpression() const
     { 
       return m_matrix; 
@@ -142,21 +122,6 @@
 }
 
 /**
-  * \return an expression of the replication of \c *this
-  *
-  * Example: \include MatrixBase_replicate_int_int.cpp
-  * Output: \verbinclude MatrixBase_replicate_int_int.out
-  *
-  * \sa VectorwiseOp::replicate(), DenseBase::replicate<int,int>(), class Replicate
-  */
-template<typename Derived>
-const typename DenseBase<Derived>::ReplicateReturnType
-DenseBase<Derived>::replicate(Index rowFactor,Index colFactor) const
-{
-  return Replicate<Derived,Dynamic,Dynamic>(derived(),rowFactor,colFactor);
-}
-
-/**
   * \return an expression of the replication of each column (or row) of \c *this
   *
   * Example: \include DirectionWise_replicate_int.cpp
diff --git a/Eigen/src/Core/ReturnByValue.h b/Eigen/src/Core/ReturnByValue.h
index f635598..c44b767 100644
--- a/Eigen/src/Core/ReturnByValue.h
+++ b/Eigen/src/Core/ReturnByValue.h
@@ -13,11 +13,6 @@
 
 namespace Eigen {
 
-/** \class ReturnByValue
-  * \ingroup Core_Module
-  *
-  */
-
 namespace internal {
 
 template<typename Derived>
@@ -38,17 +33,22 @@
  * So internal::nested always gives the plain return matrix type.
  *
  * FIXME: I don't understand why we need this specialization: isn't this taken care of by the EvalBeforeNestingBit ??
+ * Answer: EvalBeforeNestingBit should be deprecated since we have the evaluators
  */
 template<typename Derived,int n,typename PlainObject>
-struct nested<ReturnByValue<Derived>, n, PlainObject>
+struct nested_eval<ReturnByValue<Derived>, n, PlainObject>
 {
   typedef typename traits<Derived>::ReturnType type;
 };
 
 } // end namespace internal
 
+/** \class ReturnByValue
+  * \ingroup Core_Module
+  *
+  */
 template<typename Derived> class ReturnByValue
-  : internal::no_assignment_operator, public internal::dense_xpr_base< ReturnByValue<Derived> >::type
+  : public internal::dense_xpr_base< ReturnByValue<Derived> >::type, internal::no_assignment_operator
 {
   public:
     typedef typename internal::traits<Derived>::ReturnType ReturnType;
@@ -57,10 +57,11 @@
     EIGEN_DENSE_PUBLIC_INTERFACE(ReturnByValue)
 
     template<typename Dest>
+    EIGEN_DEVICE_FUNC
     inline void evalTo(Dest& dst) const
     { static_cast<const Derived*>(this)->evalTo(dst); }
-    inline Index rows() const { return static_cast<const Derived*>(this)->rows(); }
-    inline Index cols() const { return static_cast<const Derived*>(this)->cols(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return static_cast<const Derived*>(this)->rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return static_cast<const Derived*>(this)->cols(); }
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 #define Unusable YOU_ARE_TRYING_TO_ACCESS_A_SINGLE_COEFFICIENT_IN_A_SPECIAL_EXPRESSION_WHERE_THAT_IS_NOT_ALLOWED_BECAUSE_THAT_WOULD_BE_INEFFICIENT
@@ -72,8 +73,7 @@
     const Unusable& coeff(Index,Index) const { return *reinterpret_cast<const Unusable*>(this); }
     Unusable& coeffRef(Index) { return *reinterpret_cast<Unusable*>(this); }
     Unusable& coeffRef(Index,Index) { return *reinterpret_cast<Unusable*>(this); }
-    template<int LoadMode>  Unusable& packet(Index) const;
-    template<int LoadMode>  Unusable& packet(Index, Index) const;
+#undef Unusable
 #endif
 };
 
@@ -85,14 +85,32 @@
   return derived();
 }
 
-template<typename Derived>
-template<typename OtherDerived>
-Derived& DenseBase<Derived>::lazyAssign(const ReturnByValue<OtherDerived>& other)
-{
-  other.evalTo(derived());
-  return derived();
-}
+namespace internal {
 
+// Expression is evaluated in a temporary; default implementation of Assignment is bypassed so that
+// when a ReturnByValue expression is assigned, the evaluator is not constructed.
+// TODO: Finalize port to new regime; ReturnByValue should not exist in the expression world
+  
+template<typename Derived>
+struct evaluator<ReturnByValue<Derived> >
+  : public evaluator<typename internal::traits<Derived>::ReturnType>
+{
+  typedef ReturnByValue<Derived> XprType;
+  typedef typename internal::traits<Derived>::ReturnType PlainObject;
+  typedef evaluator<PlainObject> Base;
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    xpr.evalTo(m_result);
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+} // end namespace internal
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/Core/Reverse.h b/Eigen/src/Core/Reverse.h
index e30ae3d..0640cda 100644
--- a/Eigen/src/Core/Reverse.h
+++ b/Eigen/src/Core/Reverse.h
@@ -14,20 +14,6 @@
 
 namespace Eigen { 
 
-/** \class Reverse
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the reverse of a vector or matrix
-  *
-  * \param MatrixType the type of the object of which we are taking the reverse
-  *
-  * This class represents an expression of the reverse of a vector.
-  * It is the return type of MatrixBase::reverse() and VectorwiseOp::reverse()
-  * and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::reverse(), VectorwiseOp::reverse()
-  */
-
 namespace internal {
 
 template<typename MatrixType, int Direction>
@@ -37,36 +23,43 @@
   typedef typename MatrixType::Scalar Scalar;
   typedef typename traits<MatrixType>::StorageKind StorageKind;
   typedef typename traits<MatrixType>::XprKind XprKind;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = MatrixType::RowsAtCompileTime,
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-
-    // let's enable LinearAccess only with vectorization because of the product overhead
-    LinearAccess = ( (Direction==BothDirections) && (int(_MatrixTypeNested::Flags)&PacketAccessBit) )
-                 ? LinearAccessBit : 0,
-
-    Flags = int(_MatrixTypeNested::Flags) & (HereditaryBits | LvalueBit | PacketAccessBit | LinearAccess),
-
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
+    Flags = _MatrixTypeNested::Flags & (RowMajorBit | LvalueBit)
   };
 };
 
-template<typename PacketScalar, bool ReversePacket> struct reverse_packet_cond
+template<typename PacketType, bool ReversePacket> struct reverse_packet_cond
 {
-  static inline PacketScalar run(const PacketScalar& x) { return preverse(x); }
+  static inline PacketType run(const PacketType& x) { return preverse(x); }
 };
 
-template<typename PacketScalar> struct reverse_packet_cond<PacketScalar,false>
+template<typename PacketType> struct reverse_packet_cond<PacketType,false>
 {
-  static inline PacketScalar run(const PacketScalar& x) { return x; }
+  static inline PacketType run(const PacketType& x) { return x; }
 };
 
 } // end namespace internal 
 
+/** \class Reverse
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the reverse of a vector or matrix
+  *
+  * \tparam MatrixType the type of the object of which we are taking the reverse
+  * \tparam Direction defines the direction of the reverse operation, can be Vertical, Horizontal, or BothDirections
+  *
+  * This class represents an expression of the reverse of a vector.
+  * It is the return type of MatrixBase::reverse() and VectorwiseOp::reverse()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::reverse(), VectorwiseOp::reverse()
+  */
 template<typename MatrixType, int Direction> class Reverse
   : public internal::dense_xpr_base< Reverse<MatrixType, Direction> >::type
 {
@@ -74,12 +67,9 @@
 
     typedef typename internal::dense_xpr_base<Reverse>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Reverse)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
     using Base::IsRowMajor;
 
-    // next line is necessary because otherwise const version of operator()
-    // is hidden by non-const version defined in this file
-    using Base::operator(); 
-
   protected:
     enum {
       PacketSize = internal::packet_traits<Scalar>::size,
@@ -95,82 +85,19 @@
     typedef internal::reverse_packet_cond<PacketScalar,ReversePacket> reverse_packet;
   public:
 
-    inline Reverse(const MatrixType& matrix) : m_matrix(matrix) { }
+    EIGEN_DEVICE_FUNC explicit inline Reverse(const MatrixType& matrix) : m_matrix(matrix) { }
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reverse)
 
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.cols(); }
 
-    inline Index innerStride() const
+    EIGEN_DEVICE_FUNC inline Index innerStride() const
     {
       return -m_matrix.innerStride();
     }
 
-    inline Scalar& operator()(Index row, Index col)
-    {
-      eigen_assert(row >= 0 && row < rows() && col >= 0 && col < cols());
-      return coeffRef(row, col);
-    }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      return m_matrix.const_cast_derived().coeffRef(ReverseRow ? m_matrix.rows() - row - 1 : row,
-                                                    ReverseCol ? m_matrix.cols() - col - 1 : col);
-    }
-
-    inline CoeffReturnType coeff(Index row, Index col) const
-    {
-      return m_matrix.coeff(ReverseRow ? m_matrix.rows() - row - 1 : row,
-                            ReverseCol ? m_matrix.cols() - col - 1 : col);
-    }
-
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return m_matrix.coeff(m_matrix.size() - index - 1);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_matrix.const_cast_derived().coeffRef(m_matrix.size() - index - 1);
-    }
-
-    inline Scalar& operator()(Index index)
-    {
-      eigen_assert(index >= 0 && index < m_matrix.size());
-      return coeffRef(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index row, Index col) const
-    {
-      return reverse_packet::run(m_matrix.template packet<LoadMode>(
-                                    ReverseRow ? m_matrix.rows() - row - OffsetRow : row,
-                                    ReverseCol ? m_matrix.cols() - col - OffsetCol : col));
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index row, Index col, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(
-                                      ReverseRow ? m_matrix.rows() - row - OffsetRow : row,
-                                      ReverseCol ? m_matrix.cols() - col - OffsetCol : col,
-                                      reverse_packet::run(x));
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return internal::preverse(m_matrix.template packet<LoadMode>( m_matrix.size() - index - PacketSize ));
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
-    {
-      m_matrix.const_cast_derived().template writePacket<LoadMode>(m_matrix.size() - index - PacketSize, internal::preverse(x));
-    }
-
-    const typename internal::remove_all<typename MatrixType::Nested>::type& 
+    EIGEN_DEVICE_FUNC const typename internal::remove_all<typename MatrixType::Nested>::type&
     nestedExpression() const 
     {
       return m_matrix;
@@ -190,33 +117,93 @@
 inline typename DenseBase<Derived>::ReverseReturnType
 DenseBase<Derived>::reverse()
 {
-  return derived();
+  return ReverseReturnType(derived());
 }
 
-/** This is the const version of reverse(). */
-template<typename Derived>
-inline const typename DenseBase<Derived>::ConstReverseReturnType
-DenseBase<Derived>::reverse() const
-{
-  return derived();
-}
+
+//reverse const overload moved DenseBase.h due to a CUDA compiler bug
 
 /** This is the "in place" version of reverse: it reverses \c *this.
   *
   * In most cases it is probably better to simply use the reversed expression
   * of a matrix. However, when reversing the matrix data itself is really needed,
   * then this "in-place" version is probably the right choice because it provides
-  * the following additional features:
+  * the following additional benefits:
   *  - less error prone: doing the same operation with .reverse() requires special care:
   *    \code m = m.reverse().eval(); \endcode
-  *  - this API allows to avoid creating a temporary (the current implementation creates a temporary, but that could be avoided using swap)
+  *  - this API enables reverse operations without the need for a temporary
   *  - it allows future optimizations (cache friendliness, etc.)
   *
-  * \sa reverse() */
+  * \sa VectorwiseOp::reverseInPlace(), reverse() */
 template<typename Derived>
 inline void DenseBase<Derived>::reverseInPlace()
 {
-  derived() = derived().reverse().eval();
+  if(cols()>rows())
+  {
+    Index half = cols()/2;
+    leftCols(half).swap(rightCols(half).reverse());
+    if((cols()%2)==1)
+    {
+      Index half2 = rows()/2;
+      col(half).head(half2).swap(col(half).tail(half2).reverse());
+    }
+  }
+  else
+  {
+    Index half = rows()/2;
+    topRows(half).swap(bottomRows(half).reverse());
+    if((rows()%2)==1)
+    {
+      Index half2 = cols()/2;
+      row(half).head(half2).swap(row(half).tail(half2).reverse());
+    }
+  }
+}
+
+namespace internal {
+  
+template<int Direction>
+struct vectorwise_reverse_inplace_impl;
+
+template<>
+struct vectorwise_reverse_inplace_impl<Vertical>
+{
+  template<typename ExpressionType>
+  static void run(ExpressionType &xpr)
+  {
+    Index half = xpr.rows()/2;
+    xpr.topRows(half).swap(xpr.bottomRows(half).colwise().reverse());
+  }
+};
+
+template<>
+struct vectorwise_reverse_inplace_impl<Horizontal>
+{
+  template<typename ExpressionType>
+  static void run(ExpressionType &xpr)
+  {
+    Index half = xpr.cols()/2;
+    xpr.leftCols(half).swap(xpr.rightCols(half).rowwise().reverse());
+  }
+};
+
+} // end namespace internal
+
+/** This is the "in place" version of VectorwiseOp::reverse: it reverses each column or row of \c *this.
+  *
+  * In most cases it is probably better to simply use the reversed expression
+  * of a matrix. However, when reversing the matrix data itself is really needed,
+  * then this "in-place" version is probably the right choice because it provides
+  * the following additional benefits:
+  *  - less error prone: doing the same operation with .reverse() requires special care:
+  *    \code m = m.reverse().eval(); \endcode
+  *  - this API enables reverse operations without the need for a temporary
+  *
+  * \sa DenseBase::reverseInPlace(), reverse() */
+template<typename ExpressionType, int Direction>
+void VectorwiseOp<ExpressionType,Direction>::reverseInPlace()
+{
+  internal::vectorwise_reverse_inplace_impl<Direction>::run(_expression().const_cast_derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/Select.h b/Eigen/src/Core/Select.h
index 87993bb..79eec1b 100644
--- a/Eigen/src/Core/Select.h
+++ b/Eigen/src/Core/Select.h
@@ -43,23 +43,21 @@
     ColsAtCompileTime = ConditionMatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = ConditionMatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = ConditionMatrixType::MaxColsAtCompileTime,
-    Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & HereditaryBits,
-    CoeffReadCost = traits<typename remove_all<ConditionMatrixNested>::type>::CoeffReadCost
-                  + EIGEN_SIZE_MAX(traits<typename remove_all<ThenMatrixNested>::type>::CoeffReadCost,
-                                   traits<typename remove_all<ElseMatrixNested>::type>::CoeffReadCost)
+    Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & RowMajorBit
   };
 };
 }
 
 template<typename ConditionMatrixType, typename ThenMatrixType, typename ElseMatrixType>
-class Select : internal::no_assignment_operator,
-  public internal::dense_xpr_base< Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >::type
+class Select : public internal::dense_xpr_base< Select<ConditionMatrixType, ThenMatrixType, ElseMatrixType> >::type,
+               internal::no_assignment_operator
 {
   public:
 
     typedef typename internal::dense_xpr_base<Select>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Select)
 
+    inline EIGEN_DEVICE_FUNC
     Select(const ConditionMatrixType& a_conditionMatrix,
            const ThenMatrixType& a_thenMatrix,
            const ElseMatrixType& a_elseMatrix)
@@ -69,9 +67,10 @@
       eigen_assert(m_condition.cols() == m_then.cols() && m_condition.cols() == m_else.cols());
     }
 
-    Index rows() const { return m_condition.rows(); }
-    Index cols() const { return m_condition.cols(); }
+    inline EIGEN_DEVICE_FUNC Index rows() const { return m_condition.rows(); }
+    inline EIGEN_DEVICE_FUNC Index cols() const { return m_condition.cols(); }
 
+    inline EIGEN_DEVICE_FUNC
     const Scalar coeff(Index i, Index j) const
     {
       if (m_condition.coeff(i,j))
@@ -80,6 +79,7 @@
         return m_else.coeff(i,j);
     }
 
+    inline EIGEN_DEVICE_FUNC
     const Scalar coeff(Index i) const
     {
       if (m_condition.coeff(i))
@@ -88,17 +88,17 @@
         return m_else.coeff(i);
     }
 
-    const ConditionMatrixType& conditionMatrix() const
+    inline EIGEN_DEVICE_FUNC const ConditionMatrixType& conditionMatrix() const
     {
       return m_condition;
     }
 
-    const ThenMatrixType& thenMatrix() const
+    inline EIGEN_DEVICE_FUNC const ThenMatrixType& thenMatrix() const
     {
       return m_then;
     }
 
-    const ElseMatrixType& elseMatrix() const
+    inline EIGEN_DEVICE_FUNC const ElseMatrixType& elseMatrix() const
     {
       return m_else;
     }
diff --git a/Eigen/src/Core/SelfAdjointView.h b/Eigen/src/Core/SelfAdjointView.h
index 6fa7cd1..b2e51f3 100644
--- a/Eigen/src/Core/SelfAdjointView.h
+++ b/Eigen/src/Core/SelfAdjointView.h
@@ -32,54 +32,62 @@
 template<typename MatrixType, unsigned int UpLo>
 struct traits<SelfAdjointView<MatrixType, UpLo> > : traits<MatrixType>
 {
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::non_const_type MatrixTypeNested;
   typedef typename remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
   typedef MatrixType ExpressionType;
-  typedef typename MatrixType::PlainObject DenseMatrixType;
+  typedef typename MatrixType::PlainObject FullMatrixType;
   enum {
     Mode = UpLo | SelfAdjoint,
-    Flags =  MatrixTypeNestedCleaned::Flags & (HereditaryBits)
-           & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)), // FIXME these flags should be preserved
-    CoeffReadCost = MatrixTypeNestedCleaned::CoeffReadCost
+    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags =  MatrixTypeNestedCleaned::Flags & (HereditaryBits|FlagsLvalueBit)
+           & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)) // FIXME these flags should be preserved
   };
 };
 }
 
-template <typename Lhs, int LhsMode, bool LhsIsVector,
-          typename Rhs, int RhsMode, bool RhsIsVector>
-struct SelfadjointProductMatrix;
 
-// FIXME could also be called SelfAdjointWrapper to be consistent with DiagonalWrapper ??
-template<typename MatrixType, unsigned int UpLo> class SelfAdjointView
-  : public TriangularBase<SelfAdjointView<MatrixType, UpLo> >
+template<typename _MatrixType, unsigned int UpLo> class SelfAdjointView
+  : public TriangularBase<SelfAdjointView<_MatrixType, UpLo> >
 {
   public:
 
+    typedef _MatrixType MatrixType;
     typedef TriangularBase<SelfAdjointView> Base;
     typedef typename internal::traits<SelfAdjointView>::MatrixTypeNested MatrixTypeNested;
     typedef typename internal::traits<SelfAdjointView>::MatrixTypeNestedCleaned MatrixTypeNestedCleaned;
+    typedef MatrixTypeNestedCleaned NestedExpression;
 
     /** \brief The type of coefficients in this matrix */
     typedef typename internal::traits<SelfAdjointView>::Scalar Scalar; 
-
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
 
     enum {
-      Mode = internal::traits<SelfAdjointView>::Mode
+      Mode = internal::traits<SelfAdjointView>::Mode,
+      Flags = internal::traits<SelfAdjointView>::Flags,
+      TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0)
     };
     typedef typename MatrixType::PlainObject PlainObject;
 
-    inline SelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
-    {}
+    EIGEN_DEVICE_FUNC
+    explicit inline SelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
+    {
+      EIGEN_STATIC_ASSERT(UpLo==Lower || UpLo==Upper,SELFADJOINTVIEW_ACCEPTS_UPPER_AND_LOWER_MODE_ONLY);
+    }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return m_matrix.outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return m_matrix.innerStride(); }
 
     /** \sa MatrixBase::coeff()
       * \warning the coordinates must fit into the referenced triangular part
       */
+    EIGEN_DEVICE_FUNC
     inline Scalar coeff(Index row, Index col) const
     {
       Base::check_coordinates_internal(row, col);
@@ -89,36 +97,46 @@
     /** \sa MatrixBase::coeffRef()
       * \warning the coordinates must fit into the referenced triangular part
       */
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index col)
     {
+      EIGEN_STATIC_ASSERT_LVALUE(SelfAdjointView);
       Base::check_coordinates_internal(row, col);
-      return m_matrix.const_cast_derived().coeffRef(row, col);
+      return m_matrix.coeffRef(row, col);
     }
 
     /** \internal */
+    EIGEN_DEVICE_FUNC
     const MatrixTypeNestedCleaned& _expression() const { return m_matrix; }
 
+    EIGEN_DEVICE_FUNC
     const MatrixTypeNestedCleaned& nestedExpression() const { return m_matrix; }
-    MatrixTypeNestedCleaned& nestedExpression() { return *const_cast<MatrixTypeNestedCleaned*>(&m_matrix); }
+    EIGEN_DEVICE_FUNC
+    MatrixTypeNestedCleaned& nestedExpression() { return m_matrix; }
 
-    /** Efficient self-adjoint matrix times vector/matrix product */
+    /** Efficient triangular matrix times vector/matrix product */
     template<typename OtherDerived>
-    SelfadjointProductMatrix<MatrixType,Mode,false,OtherDerived,0,OtherDerived::IsVectorAtCompileTime>
+    EIGEN_DEVICE_FUNC
+    const Product<SelfAdjointView,OtherDerived>
     operator*(const MatrixBase<OtherDerived>& rhs) const
     {
-      return SelfadjointProductMatrix
-              <MatrixType,Mode,false,OtherDerived,0,OtherDerived::IsVectorAtCompileTime>
-              (m_matrix, rhs.derived());
+      return Product<SelfAdjointView,OtherDerived>(*this, rhs.derived());
     }
 
-    /** Efficient vector/matrix times self-adjoint matrix product */
+    /** Efficient vector/matrix times triangular matrix product */
     template<typename OtherDerived> friend
-    SelfadjointProductMatrix<OtherDerived,0,OtherDerived::IsVectorAtCompileTime,MatrixType,Mode,false>
+    EIGEN_DEVICE_FUNC
+    const Product<OtherDerived,SelfAdjointView>
     operator*(const MatrixBase<OtherDerived>& lhs, const SelfAdjointView& rhs)
     {
-      return SelfadjointProductMatrix
-              <OtherDerived,0,OtherDerived::IsVectorAtCompileTime,MatrixType,Mode,false>
-              (lhs.derived(),rhs.m_matrix);
+      return Product<OtherDerived,SelfAdjointView>(lhs.derived(),rhs);
+    }
+    
+    friend EIGEN_DEVICE_FUNC
+    const SelfAdjointView<const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(Scalar,MatrixType,product),UpLo>
+    operator*(const Scalar& s, const SelfAdjointView& mat)
+    {
+      return (s*mat.nestedExpression()).template selfadjointView<UpLo>();
     }
 
     /** Perform a symmetric rank 2 update of the selfadjoint matrix \c *this:
@@ -132,6 +150,7 @@
       * \sa rankUpdate(const MatrixBase<DerivedU>&, Scalar)
       */
     template<typename DerivedU, typename DerivedV>
+    EIGEN_DEVICE_FUNC
     SelfAdjointView& rankUpdate(const MatrixBase<DerivedU>& u, const MatrixBase<DerivedV>& v, const Scalar& alpha = Scalar(1));
 
     /** Perform a symmetric rank K update of the selfadjoint matrix \c *this:
@@ -145,8 +164,74 @@
       * \sa rankUpdate(const MatrixBase<DerivedU>&, const MatrixBase<DerivedV>&, Scalar)
       */
     template<typename DerivedU>
+    EIGEN_DEVICE_FUNC
     SelfAdjointView& rankUpdate(const MatrixBase<DerivedU>& u, const Scalar& alpha = Scalar(1));
 
+    /** \returns an expression of a triangular view extracted from the current selfadjoint view of a given triangular part
+      *
+      * The parameter \a TriMode can have the following values: \c #Upper, \c #StrictlyUpper, \c #UnitUpper,
+      * \c #Lower, \c #StrictlyLower, \c #UnitLower.
+      *
+      * If \c TriMode references the same triangular part than \c *this, then this method simply return a \c TriangularView of the nested expression,
+      * otherwise, the nested expression is first transposed, thus returning a \c TriangularView<Transpose<MatrixType>> object.
+      *
+      * \sa MatrixBase::triangularView(), class TriangularView
+      */
+    template<unsigned int TriMode>
+    EIGEN_DEVICE_FUNC
+    typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)),
+                                   TriangularView<MatrixType,TriMode>,
+                                   TriangularView<typename MatrixType::AdjointReturnType,TriMode> >::type
+    triangularView() const
+    {
+      typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)), MatrixType&, typename MatrixType::ConstTransposeReturnType>::type tmp1(m_matrix);
+      typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)), MatrixType&, typename MatrixType::AdjointReturnType>::type tmp2(tmp1);
+      return typename internal::conditional<(TriMode&(Upper|Lower))==(UpLo&(Upper|Lower)),
+                                   TriangularView<MatrixType,TriMode>,
+                                   TriangularView<typename MatrixType::AdjointReturnType,TriMode> >::type(tmp2);
+    }
+
+    typedef SelfAdjointView<const MatrixConjugateReturnType,UpLo> ConjugateReturnType;
+    /** \sa MatrixBase::conjugate() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConjugateReturnType conjugate() const
+    { return ConjugateReturnType(m_matrix.conjugate()); }
+
+    typedef SelfAdjointView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
+    /** \sa MatrixBase::adjoint() const */
+    EIGEN_DEVICE_FUNC
+    inline const AdjointReturnType adjoint() const
+    { return AdjointReturnType(m_matrix.adjoint()); }
+
+    typedef SelfAdjointView<typename MatrixType::TransposeReturnType,TransposeMode> TransposeReturnType;
+     /** \sa MatrixBase::transpose() */
+    EIGEN_DEVICE_FUNC
+    inline TransposeReturnType transpose()
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
+      typename MatrixType::TransposeReturnType tmp(m_matrix);
+      return TransposeReturnType(tmp);
+    }
+
+    typedef SelfAdjointView<const typename MatrixType::ConstTransposeReturnType,TransposeMode> ConstTransposeReturnType;
+    /** \sa MatrixBase::transpose() const */
+    EIGEN_DEVICE_FUNC
+    inline const ConstTransposeReturnType transpose() const
+    {
+      return ConstTransposeReturnType(m_matrix.transpose());
+    }
+
+    /** \returns a const expression of the main diagonal of the matrix \c *this
+      *
+      * This method simply returns the diagonal of the nested expression, thus by-passing the SelfAdjointView decorator.
+      *
+      * \sa MatrixBase::diagonal(), class Diagonal */
+    EIGEN_DEVICE_FUNC
+    typename MatrixType::ConstDiagonalReturnType diagonal() const
+    {
+      return typename MatrixType::ConstDiagonalReturnType(m_matrix);
+    }
+
 /////////// Cholesky module ///////////
 
     const LLT<PlainObject, UpLo> llt() const;
@@ -159,31 +244,10 @@
     /** Return type of eigenvalues() */
     typedef Matrix<RealScalar, internal::traits<MatrixType>::ColsAtCompileTime, 1> EigenvaluesReturnType;
 
+    EIGEN_DEVICE_FUNC
     EigenvaluesReturnType eigenvalues() const;
+    EIGEN_DEVICE_FUNC
     RealScalar operatorNorm() const;
-    
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    SelfAdjointView& operator=(const MatrixBase<OtherDerived>& other)
-    {
-      enum {
-        OtherPart = UpLo == Upper ? StrictlyLower : StrictlyUpper
-      };
-      m_matrix.const_cast_derived().template triangularView<UpLo>() = other;
-      m_matrix.const_cast_derived().template triangularView<OtherPart>() = other.adjoint();
-      return *this;
-    }
-    template<typename OtherMatrixType, unsigned int OtherMode>
-    SelfAdjointView& operator=(const TriangularView<OtherMatrixType, OtherMode>& other)
-    {
-      enum {
-        OtherPart = UpLo == Upper ? StrictlyLower : StrictlyUpper
-      };
-      m_matrix.const_cast_derived().template triangularView<UpLo>() = other.toDenseMatrix();
-      m_matrix.const_cast_derived().template triangularView<OtherPart>() = other.toDenseMatrix().adjoint();
-      return *this;
-    }
-    #endif
 
   protected:
     MatrixTypeNested m_matrix;
@@ -201,90 +265,54 @@
 
 namespace internal {
 
-template<typename Derived1, typename Derived2, int UnrollCount, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Upper), UnrollCount, ClearOpposite>
+// TODO currently a selfadjoint expression has the form SelfAdjointView<.,.>
+//      in the future selfadjoint-ness should be defined by the expression traits
+//      such that Transpose<SelfAdjointView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
+template<typename MatrixType, unsigned int Mode>
+struct evaluator_traits<SelfAdjointView<MatrixType,Mode> >
 {
-  enum {
-    col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived1::RowsAtCompileTime
-  };
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef SelfAdjointShape Shape;
+};
 
-  static inline void run(Derived1 &dst, const Derived2 &src)
+template<int UpLo, int SetOpposite, typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version>
+class triangular_dense_assignment_kernel<UpLo,SelfAdjoint,SetOpposite,DstEvaluatorTypeT,SrcEvaluatorTypeT,Functor,Version>
+  : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version>
+{
+protected:
+  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version> Base;
+  typedef typename Base::DstXprType DstXprType;
+  typedef typename Base::SrcXprType SrcXprType;
+  using Base::m_dst;
+  using Base::m_src;
+  using Base::m_functor;
+public:
+  
+  typedef typename Base::DstEvaluatorType DstEvaluatorType;
+  typedef typename Base::SrcEvaluatorType SrcEvaluatorType;
+  typedef typename Base::Scalar Scalar;
+  typedef typename Base::AssignmentTraits AssignmentTraits;
+  
+  
+  EIGEN_DEVICE_FUNC triangular_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
+    : Base(dst, src, func, dstExpr)
+  {}
+  
+  EIGEN_DEVICE_FUNC void assignCoeff(Index row, Index col)
   {
-    triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Upper), UnrollCount-1, ClearOpposite>::run(dst, src);
-
-    if(row == col)
-      dst.coeffRef(row, col) = numext::real(src.coeff(row, col));
-    else if(row < col)
-      dst.coeffRef(col, row) = numext::conj(dst.coeffRef(row, col) = src.coeff(row, col));
+    eigen_internal_assert(row!=col);
+    Scalar tmp = m_src.coeff(row,col);
+    m_functor.assignCoeff(m_dst.coeffRef(row,col), tmp);
+    m_functor.assignCoeff(m_dst.coeffRef(col,row), numext::conj(tmp));
   }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Upper, 0, ClearOpposite>
-{
-  static inline void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, int UnrollCount, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Lower), UnrollCount, ClearOpposite>
-{
-  enum {
-    col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived1::RowsAtCompileTime
-  };
-
-  static inline void run(Derived1 &dst, const Derived2 &src)
+  
+  EIGEN_DEVICE_FUNC void assignDiagonalCoeff(Index id)
   {
-    triangular_assignment_selector<Derived1, Derived2, (SelfAdjoint|Lower), UnrollCount-1, ClearOpposite>::run(dst, src);
-
-    if(row == col)
-      dst.coeffRef(row, col) = numext::real(src.coeff(row, col));
-    else if(row > col)
-      dst.coeffRef(col, row) = numext::conj(dst.coeffRef(row, col) = src.coeff(row, col));
+    Base::assignCoeff(id,id);
   }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Lower, 0, ClearOpposite>
-{
-  static inline void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Upper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      for(Index i = 0; i < j; ++i)
-      {
-        dst.copyCoeff(i, j, src);
-        dst.coeffRef(j,i) = numext::conj(dst.coeff(i,j));
-      }
-      dst.copyCoeff(j, j, src);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, SelfAdjoint|Lower, Dynamic, ClearOpposite>
-{
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-  typedef typename Derived1::Index Index;
-    for(Index i = 0; i < dst.rows(); ++i)
-    {
-      for(Index j = 0; j < i; ++j)
-      {
-        dst.copyCoeff(i, j, src);
-        dst.coeffRef(j,i) = numext::conj(dst.coeff(i,j));
-      }
-      dst.copyCoeff(i, i, src);
-    }
-  }
+  
+  EIGEN_DEVICE_FUNC void assignOppositeCoeff(Index, Index)
+  { eigen_internal_assert(false && "should never be called"); }
 };
 
 } // end namespace internal
@@ -293,20 +321,30 @@
 * Implementation of MatrixBase methods
 ***************************************************************************/
 
+/** This is the const version of MatrixBase::selfadjointView() */
 template<typename Derived>
 template<unsigned int UpLo>
 typename MatrixBase<Derived>::template ConstSelfAdjointViewReturnType<UpLo>::Type
 MatrixBase<Derived>::selfadjointView() const
 {
-  return derived();
+  return typename ConstSelfAdjointViewReturnType<UpLo>::Type(derived());
 }
 
+/** \returns an expression of a symmetric/self-adjoint view extracted from the upper or lower triangular part of the current matrix
+  *
+  * The parameter \a UpLo can be either \c #Upper or \c #Lower
+  *
+  * Example: \include MatrixBase_selfadjointView.cpp
+  * Output: \verbinclude MatrixBase_selfadjointView.out
+  *
+  * \sa class SelfAdjointView
+  */
 template<typename Derived>
 template<unsigned int UpLo>
 typename MatrixBase<Derived>::template SelfAdjointViewReturnType<UpLo>::Type
 MatrixBase<Derived>::selfadjointView()
 {
-  return derived();
+  return typename SelfAdjointViewReturnType<UpLo>::Type(derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/SelfCwiseBinaryOp.h b/Eigen/src/Core/SelfCwiseBinaryOp.h
index 22f3047..7c89c2e 100644
--- a/Eigen/src/Core/SelfCwiseBinaryOp.h
+++ b/Eigen/src/Core/SelfCwiseBinaryOp.h
@@ -12,183 +12,33 @@
 
 namespace Eigen { 
 
-/** \class SelfCwiseBinaryOp
-  * \ingroup Core_Module
-  *
-  * \internal
-  *
-  * \brief Internal helper class for optimizing operators like +=, -=
-  *
-  * This is a pseudo expression class re-implementing the copyCoeff/copyPacket
-  * method to directly performs a +=/-= operations in an optimal way. In particular,
-  * this allows to make sure that the input/output data are loaded only once using
-  * aligned packet loads.
-  *
-  * \sa class SwapWrapper for a similar trick.
-  */
-
-namespace internal {
-template<typename BinaryOp, typename Lhs, typename Rhs>
-struct traits<SelfCwiseBinaryOp<BinaryOp,Lhs,Rhs> >
-  : traits<CwiseBinaryOp<BinaryOp,Lhs,Rhs> >
-{
-  enum {
-    // Note that it is still a good idea to preserve the DirectAccessBit
-    // so that assign can correctly align the data.
-    Flags = traits<CwiseBinaryOp<BinaryOp,Lhs,Rhs> >::Flags | (Lhs::Flags&DirectAccessBit) | (Lhs::Flags&LvalueBit),
-    OuterStrideAtCompileTime = Lhs::OuterStrideAtCompileTime,
-    InnerStrideAtCompileTime = Lhs::InnerStrideAtCompileTime
-  };
-};
-}
-
-template<typename BinaryOp, typename Lhs, typename Rhs> class SelfCwiseBinaryOp
-  : public internal::dense_xpr_base< SelfCwiseBinaryOp<BinaryOp, Lhs, Rhs> >::type
-{
-  public:
-
-    typedef typename internal::dense_xpr_base<SelfCwiseBinaryOp>::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SelfCwiseBinaryOp)
-
-    typedef typename internal::packet_traits<Scalar>::type Packet;
-
-    inline SelfCwiseBinaryOp(Lhs& xpr, const BinaryOp& func = BinaryOp()) : m_matrix(xpr), m_functor(func) {}
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-    inline Index outerStride() const { return m_matrix.outerStride(); }
-    inline Index innerStride() const { return m_matrix.innerStride(); }
-    inline const Scalar* data() const { return m_matrix.data(); }
-
-    // note that this function is needed by assign to correctly align loads/stores
-    // TODO make Assign use .data()
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(Lhs)
-      return m_matrix.const_cast_derived().coeffRef(row, col);
-    }
-    inline const Scalar& coeffRef(Index row, Index col) const
-    {
-      return m_matrix.coeffRef(row, col);
-    }
-
-    // note that this function is needed by assign to correctly align loads/stores
-    // TODO make Assign use .data()
-    inline Scalar& coeffRef(Index index)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(Lhs)
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-    inline const Scalar& coeffRef(Index index) const
-    {
-      return m_matrix.const_cast_derived().coeffRef(index);
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index row, Index col, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(row >= 0 && row < rows()
-                         && col >= 0 && col < cols());
-      Scalar& tmp = m_matrix.coeffRef(row,col);
-      tmp = m_functor(tmp, _other.coeff(row,col));
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_matrix.size());
-      Scalar& tmp = m_matrix.coeffRef(index);
-      tmp = m_functor(tmp, _other.coeff(index));
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index row, Index col, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(row >= 0 && row < rows()
-                        && col >= 0 && col < cols());
-      m_matrix.template writePacket<StoreMode>(row, col,
-        m_functor.packetOp(m_matrix.template packet<StoreMode>(row, col),_other.template packet<LoadMode>(row, col)) );
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_matrix.size());
-      m_matrix.template writePacket<StoreMode>(index,
-        m_functor.packetOp(m_matrix.template packet<StoreMode>(index),_other.template packet<LoadMode>(index)) );
-    }
-
-    // reimplement lazyAssign to handle complex *= real
-    // see CwiseBinaryOp ctor for details
-    template<typename RhsDerived>
-    EIGEN_STRONG_INLINE SelfCwiseBinaryOp& lazyAssign(const DenseBase<RhsDerived>& rhs)
-    {
-      EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Lhs,RhsDerived)
-      EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename RhsDerived::Scalar);
-      
-    #ifdef EIGEN_DEBUG_ASSIGN
-      internal::assign_traits<SelfCwiseBinaryOp, RhsDerived>::debug();
-    #endif
-      eigen_assert(rows() == rhs.rows() && cols() == rhs.cols());
-      internal::assign_impl<SelfCwiseBinaryOp, RhsDerived>::run(*this,rhs.derived());
-    #ifndef EIGEN_NO_DEBUG
-      this->checkTransposeAliasing(rhs.derived());
-    #endif
-      return *this;
-    }
-    
-    // overloaded to honor evaluation of special matrices
-    // maybe another solution would be to not use SelfCwiseBinaryOp
-    // at first...
-    SelfCwiseBinaryOp& operator=(const Rhs& _rhs)
-    {
-      typename internal::nested<Rhs>::type rhs(_rhs);
-      return Base::operator=(rhs);
-    }
-
-    Lhs& expression() const 
-    { 
-      return m_matrix;
-    }
-
-    const BinaryOp& functor() const 
-    { 
-      return m_functor;
-    }
-
-  protected:
-    Lhs& m_matrix;
-    const BinaryOp& m_functor;
-
-  private:
-    SelfCwiseBinaryOp& operator=(const SelfCwiseBinaryOp&);
-};
+// TODO generalize the scalar type of 'other'
 
 template<typename Derived>
-inline Derived& DenseBase<Derived>::operator*=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator*=(const Scalar& other)
 {
-  typedef typename Derived::PlainObject PlainObject;
-  SelfCwiseBinaryOp<internal::scalar_product_op<Scalar>, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
-  tmp = PlainObject::Constant(rows(),cols(),other);
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::mul_assign_op<Scalar,Scalar>());
   return derived();
 }
 
 template<typename Derived>
-inline Derived& DenseBase<Derived>::operator/=(const Scalar& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator+=(const Scalar& other)
 {
-  typedef typename internal::conditional<NumTraits<Scalar>::IsInteger,
-                                        internal::scalar_quotient_op<Scalar>,
-                                        internal::scalar_product_op<Scalar> >::type BinOp;
-  typedef typename Derived::PlainObject PlainObject;
-  SelfCwiseBinaryOp<BinOp, Derived, typename PlainObject::ConstantReturnType> tmp(derived());
-  Scalar actual_other;
-  if(NumTraits<Scalar>::IsInteger)  actual_other = other;
-  else                              actual_other = Scalar(1)/other;
-  tmp = PlainObject::Constant(rows(),cols(), actual_other);
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::add_assign_op<Scalar,Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& ArrayBase<Derived>::operator-=(const Scalar& other)
+{
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::sub_assign_op<Scalar,Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase<Derived>::operator/=(const Scalar& other)
+{
+  internal::call_assignment(this->derived(), PlainObject::Constant(rows(),cols(),other), internal::div_assign_op<Scalar,Scalar>());
   return derived();
 }
 
diff --git a/Eigen/src/Core/Solve.h b/Eigen/src/Core/Solve.h
new file mode 100644
index 0000000..a8daea5
--- /dev/null
+++ b/Eigen/src/Core/Solve.h
@@ -0,0 +1,188 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SOLVE_H
+#define EIGEN_SOLVE_H
+
+namespace Eigen {
+
+template<typename Decomposition, typename RhsType, typename StorageKind> class SolveImpl;
+  
+/** \class Solve
+  * \ingroup Core_Module
+  *
+  * \brief Pseudo expression representing a solving operation
+  *
+  * \tparam Decomposition the type of the matrix or decomposion object
+  * \tparam Rhstype the type of the right-hand side
+  *
+  * This class represents an expression of A.solve(B)
+  * and most of the time this is the only way it is used.
+  *
+  */
+namespace internal {
+
+// this solve_traits class permits to determine the evaluation type with respect to storage kind (Dense vs Sparse)
+template<typename Decomposition, typename RhsType,typename StorageKind> struct solve_traits;
+
+template<typename Decomposition, typename RhsType>
+struct solve_traits<Decomposition,RhsType,Dense>
+{
+  typedef typename make_proper_matrix_type<typename RhsType::Scalar,
+                 Decomposition::ColsAtCompileTime,
+                 RhsType::ColsAtCompileTime,
+                 RhsType::PlainObject::Options,
+                 Decomposition::MaxColsAtCompileTime,
+                 RhsType::MaxColsAtCompileTime>::type PlainObject;
+};
+
+template<typename Decomposition, typename RhsType>
+struct traits<Solve<Decomposition, RhsType> >
+  : traits<typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject>
+{
+  typedef typename solve_traits<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>::PlainObject PlainObject;
+  typedef typename promote_index_type<typename Decomposition::StorageIndex, typename RhsType::StorageIndex>::type StorageIndex;
+  typedef traits<PlainObject> BaseTraits;
+  enum {
+    Flags = BaseTraits::Flags & RowMajorBit,
+    CoeffReadCost = HugeCost
+  };
+};
+
+}
+
+
+template<typename Decomposition, typename RhsType>
+class Solve : public SolveImpl<Decomposition,RhsType,typename internal::traits<RhsType>::StorageKind>
+{
+public:
+  typedef typename internal::traits<Solve>::PlainObject PlainObject;
+  typedef typename internal::traits<Solve>::StorageIndex StorageIndex;
+  
+  Solve(const Decomposition &dec, const RhsType &rhs)
+    : m_dec(dec), m_rhs(rhs)
+  {}
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_dec.cols(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_rhs.cols(); }
+
+  EIGEN_DEVICE_FUNC const Decomposition& dec() const { return m_dec; }
+  EIGEN_DEVICE_FUNC const RhsType&       rhs() const { return m_rhs; }
+
+protected:
+  const Decomposition &m_dec;
+  const RhsType       &m_rhs;
+};
+
+
+// Specialization of the Solve expression for dense results
+template<typename Decomposition, typename RhsType>
+class SolveImpl<Decomposition,RhsType,Dense>
+  : public MatrixBase<Solve<Decomposition,RhsType> >
+{
+  typedef Solve<Decomposition,RhsType> Derived;
+  
+public:
+  
+  typedef MatrixBase<Solve<Decomposition,RhsType> > Base;
+  EIGEN_DENSE_PUBLIC_INTERFACE(Derived)
+
+private:
+  
+  Scalar coeff(Index row, Index col) const;
+  Scalar coeff(Index i) const;
+};
+
+// Generic API dispatcher
+template<typename Decomposition, typename RhsType, typename StorageKind>
+class SolveImpl : public internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type
+{
+  public:
+    typedef typename internal::generic_xpr_base<Solve<Decomposition,RhsType>, MatrixXpr, StorageKind>::type Base;
+};
+
+namespace internal {
+
+// Evaluator of Solve -> eval into a temporary
+template<typename Decomposition, typename RhsType>
+struct evaluator<Solve<Decomposition,RhsType> >
+  : public evaluator<typename Solve<Decomposition,RhsType>::PlainObject>
+{
+  typedef Solve<Decomposition,RhsType> SolveType;
+  typedef typename SolveType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  enum { Flags = Base::Flags | EvalBeforeNestingBit };
+  
+  EIGEN_DEVICE_FUNC explicit evaluator(const SolveType& solve)
+    : m_result(solve.rows(), solve.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    solve.dec()._solve_impl(solve.rhs(), m_result);
+  }
+  
+protected:  
+  PlainObject m_result;
+};
+
+// Specialization for "dst = dec.solve(rhs)"
+// NOTE we need to specialize it for Dense2Dense to avoid ambiguous specialization error and a Sparse2Sparse specialization must exist somewhere
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<DecType,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef Solve<DecType,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    src.dec()._solve_impl(src.rhs(), dst);
+  }
+};
+
+// Specialization for "dst = dec.transpose().solve(rhs)"
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<Transpose<const DecType>,RhsType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef Solve<Transpose<const DecType>,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    src.dec().nestedExpression().template _solve_impl_transposed<false>(src.rhs(), dst);
+  }
+};
+
+// Specialization for "dst = dec.adjoint().solve(rhs)"
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<CwiseUnaryOp<internal::scalar_conjugate_op<typename DecType::Scalar>, const Transpose<const DecType> >,RhsType>,
+                  internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef Solve<CwiseUnaryOp<internal::scalar_conjugate_op<typename DecType::Scalar>, const Transpose<const DecType> >,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    src.dec().nestedExpression().nestedExpression().template _solve_impl_transposed<true>(src.rhs(), dst);
+  }
+};
+
+} // end namepsace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SOLVE_H
diff --git a/Eigen/src/Core/SolveTriangular.h b/Eigen/src/Core/SolveTriangular.h
index ef17f28..4652e2e 100644
--- a/Eigen/src/Core/SolveTriangular.h
+++ b/Eigen/src/Core/SolveTriangular.h
@@ -68,7 +68,7 @@
     if(!useRhsDirectly)
       MappedRhs(actualRhs,rhs.size()) = rhs;
 
-    triangular_solve_vector<LhsScalar, RhsScalar, typename Lhs::Index, Side, Mode, LhsProductTraits::NeedToConjugate,
+    triangular_solve_vector<LhsScalar, RhsScalar, Index, Side, Mode, LhsProductTraits::NeedToConjugate,
                             (int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor>
       ::run(actualLhs.cols(), actualLhs.data(), actualLhs.outerStride(), actualRhs);
 
@@ -82,7 +82,6 @@
 struct triangular_solver_selector<Lhs,Rhs,Side,Mode,NoUnrolling,Dynamic>
 {
   typedef typename Rhs::Scalar Scalar;
-  typedef typename Rhs::Index Index;
   typedef blas_traits<Lhs> LhsProductTraits;
   typedef typename LhsProductTraits::DirectLinearAccessType ActualLhsType;
 
@@ -96,7 +95,7 @@
     typedef internal::gemm_blocking_space<(Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
               Rhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxRowsAtCompileTime,4> BlockingType;
 
-    BlockingType blocking(rhs.rows(), rhs.cols(), size);
+    BlockingType blocking(rhs.rows(), rhs.cols(), size, 1, false);
 
     triangular_solve_matrix<Scalar,Index,Side,Mode,LhsProductTraits::NeedToConjugate,(int(Lhs::Flags) & RowMajorBit) ? RowMajor : ColMajor,
                                (Rhs::Flags&RowMajorBit) ? RowMajor : ColMajor>
@@ -108,32 +107,32 @@
 * meta-unrolling implementation
 ***************************************************************************/
 
-template<typename Lhs, typename Rhs, int Mode, int Index, int Size,
-         bool Stop = Index==Size>
+template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size,
+         bool Stop = LoopIndex==Size>
 struct triangular_solver_unroller;
 
-template<typename Lhs, typename Rhs, int Mode, int Index, int Size>
-struct triangular_solver_unroller<Lhs,Rhs,Mode,Index,Size,false> {
+template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size>
+struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,false> {
   enum {
     IsLower = ((Mode&Lower)==Lower),
-    I = IsLower ? Index : Size - Index - 1,
-    S = IsLower ? 0     : I+1
+    DiagIndex  = IsLower ? LoopIndex : Size - LoopIndex - 1,
+    StartIndex = IsLower ? 0         : DiagIndex+1
   };
   static void run(const Lhs& lhs, Rhs& rhs)
   {
-    if (Index>0)
-      rhs.coeffRef(I) -= lhs.row(I).template segment<Index>(S).transpose()
-                         .cwiseProduct(rhs.template segment<Index>(S)).sum();
+    if (LoopIndex>0)
+      rhs.coeffRef(DiagIndex) -= lhs.row(DiagIndex).template segment<LoopIndex>(StartIndex).transpose()
+                                .cwiseProduct(rhs.template segment<LoopIndex>(StartIndex)).sum();
 
     if(!(Mode & UnitDiag))
-      rhs.coeffRef(I) /= lhs.coeff(I,I);
+      rhs.coeffRef(DiagIndex) /= lhs.coeff(DiagIndex,DiagIndex);
 
-    triangular_solver_unroller<Lhs,Rhs,Mode,Index+1,Size>::run(lhs,rhs);
+    triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex+1,Size>::run(lhs,rhs);
   }
 };
 
-template<typename Lhs, typename Rhs, int Mode, int Index, int Size>
-struct triangular_solver_unroller<Lhs,Rhs,Mode,Index,Size,true> {
+template<typename Lhs, typename Rhs, int Mode, int LoopIndex, int Size>
+struct triangular_solver_unroller<Lhs,Rhs,Mode,LoopIndex,Size,true> {
   static void run(const Lhs&, Rhs&) {}
 };
 
@@ -162,61 +161,38 @@
 * TriangularView methods
 ***************************************************************************/
 
-/** "in-place" version of TriangularView::solve() where the result is written in \a other
-  *
-  * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
-  * This function will const_cast it, so constness isn't honored here.
-  *
-  * See TriangularView:solve() for the details.
-  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename MatrixType, unsigned int Mode>
 template<int Side, typename OtherDerived>
-void TriangularView<MatrixType,Mode>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
+void TriangularViewImpl<MatrixType,Mode,Dense>::solveInPlace(const MatrixBase<OtherDerived>& _other) const
 {
   OtherDerived& other = _other.const_cast_derived();
-  eigen_assert( cols() == rows() && ((Side==OnTheLeft && cols() == other.rows()) || (Side==OnTheRight && cols() == other.cols())) );
+  eigen_assert( derived().cols() == derived().rows() && ((Side==OnTheLeft && derived().cols() == other.rows()) || (Side==OnTheRight && derived().cols() == other.cols())) );
   eigen_assert((!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
+  // If solving for a 0x0 matrix, nothing to do, simply return.
+  if (derived().cols() == 0)
+    return;
 
-  enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit  && OtherDerived::IsVectorAtCompileTime };
+  enum { copy = (internal::traits<OtherDerived>::Flags & RowMajorBit)  && OtherDerived::IsVectorAtCompileTime && OtherDerived::SizeAtCompileTime!=1};
   typedef typename internal::conditional<copy,
     typename internal::plain_matrix_type_column_major<OtherDerived>::type, OtherDerived&>::type OtherCopy;
   OtherCopy otherCopy(other);
 
   internal::triangular_solver_selector<MatrixType, typename internal::remove_reference<OtherCopy>::type,
-    Side, Mode>::run(nestedExpression(), otherCopy);
+    Side, Mode>::run(derived().nestedExpression(), otherCopy);
 
   if (copy)
     other = otherCopy;
 }
 
-/** \returns the product of the inverse of \c *this with \a other, \a *this being triangular.
-  *
-  * This function computes the inverse-matrix matrix product inverse(\c *this) * \a other if
-  * \a Side==OnTheLeft (the default), or the right-inverse-multiply  \a other * inverse(\c *this) if
-  * \a Side==OnTheRight.
-  *
-  * The matrix \c *this must be triangular and invertible (i.e., all the coefficients of the
-  * diagonal must be non zero). It works as a forward (resp. backward) substitution if \c *this
-  * is an upper (resp. lower) triangular matrix.
-  *
-  * Example: \include MatrixBase_marked.cpp
-  * Output: \verbinclude MatrixBase_marked.out
-  *
-  * This function returns an expression of the inverse-multiply and can works in-place if it is assigned
-  * to the same matrix or vector \a other.
-  *
-  * For users coming from BLAS, this function (and more specifically solveInPlace()) offer
-  * all the operations supported by the \c *TRSV and \c *TRSM BLAS routines.
-  *
-  * \sa TriangularView::solveInPlace()
-  */
 template<typename Derived, unsigned int Mode>
 template<int Side, typename Other>
 const internal::triangular_solve_retval<Side,TriangularView<Derived,Mode>,Other>
-TriangularView<Derived,Mode>::solve(const MatrixBase<Other>& other) const
+TriangularViewImpl<Derived,Mode,Dense>::solve(const MatrixBase<Other>& other) const
 {
-  return internal::triangular_solve_retval<Side,TriangularView,Other>(*this, other.derived());
+  return internal::triangular_solve_retval<Side,TriangularViewType,Other>(derived(), other.derived());
 }
+#endif
 
 namespace internal {
 
@@ -232,7 +208,6 @@
 {
   typedef typename remove_all<typename Rhs::Nested>::type RhsNestedCleaned;
   typedef ReturnByValue<triangular_solve_retval> Base;
-  typedef typename Base::Index Index;
 
   triangular_solve_retval(const TriangularType& tri, const Rhs& rhs)
     : m_triangularMatrix(tri), m_rhs(rhs)
@@ -243,7 +218,7 @@
 
   template<typename Dest> inline void evalTo(Dest& dst) const
   {
-    if(!(is_same<RhsNestedCleaned,Dest>::value && extract_data(dst) == extract_data(m_rhs)))
+    if(!is_same_dense(dst,m_rhs))
       dst = m_rhs;
     m_triangularMatrix.template solveInPlace<Side>(dst);
   }
diff --git a/Eigen/src/Core/SolverBase.h b/Eigen/src/Core/SolverBase.h
new file mode 100644
index 0000000..8a4adc2
--- /dev/null
+++ b/Eigen/src/Core/SolverBase.h
@@ -0,0 +1,130 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SOLVERBASE_H
+#define EIGEN_SOLVERBASE_H
+
+namespace Eigen {
+
+namespace internal {
+
+
+
+} // end namespace internal
+
+/** \class SolverBase
+  * \brief A base class for matrix decomposition and solvers
+  *
+  * \tparam Derived the actual type of the decomposition/solver.
+  *
+  * Any matrix decomposition inheriting this base class provide the following API:
+  *
+  * \code
+  * MatrixType A, b, x;
+  * DecompositionType dec(A);
+  * x = dec.solve(b);             // solve A   * x = b
+  * x = dec.transpose().solve(b); // solve A^T * x = b
+  * x = dec.adjoint().solve(b);   // solve A'  * x = b
+  * \endcode
+  *
+  * \warning Currently, any other usage of transpose() and adjoint() are not supported and will produce compilation errors.
+  *
+  * \sa class PartialPivLU, class FullPivLU
+  */
+template<typename Derived>
+class SolverBase : public EigenBase<Derived>
+{
+  public:
+
+    typedef EigenBase<Derived> Base;
+    typedef typename internal::traits<Derived>::Scalar Scalar;
+    typedef Scalar CoeffReturnType;
+
+    enum {
+      RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
+      SizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::RowsAtCompileTime,
+                                                          internal::traits<Derived>::ColsAtCompileTime>::ret),
+      MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
+      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
+      MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
+                                                             internal::traits<Derived>::MaxColsAtCompileTime>::ret),
+      IsVectorAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime == 1
+                           || internal::traits<Derived>::MaxColsAtCompileTime == 1
+    };
+
+    /** Default constructor */
+    SolverBase()
+    {}
+
+    ~SolverBase()
+    {}
+
+    using Base::derived;
+
+    /** \returns an expression of the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      */
+    template<typename Rhs>
+    inline const Solve<Derived, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<Derived, Rhs>(derived(), b.derived());
+    }
+
+    /** \internal the return type of transpose() */
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
+    /** \returns an expression of the transposed of the factored matrix.
+      *
+      * A typical usage is to solve for the transposed problem A^T x = b:
+      * \code x = dec.transpose().solve(b); \endcode
+      *
+      * \sa adjoint(), solve()
+      */
+    inline ConstTransposeReturnType transpose() const
+    {
+      return ConstTransposeReturnType(derived());
+    }
+
+    /** \internal the return type of adjoint() */
+    typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
+                        CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, ConstTransposeReturnType>,
+                        ConstTransposeReturnType
+                     >::type AdjointReturnType;
+    /** \returns an expression of the adjoint of the factored matrix
+      *
+      * A typical usage is to solve for the adjoint problem A' x = b:
+      * \code x = dec.adjoint().solve(b); \endcode
+      *
+      * For real scalar types, this function is equivalent to transpose().
+      *
+      * \sa transpose(), solve()
+      */
+    inline AdjointReturnType adjoint() const
+    {
+      return AdjointReturnType(derived().transpose());
+    }
+
+  protected:
+};
+
+namespace internal {
+
+template<typename Derived>
+struct generic_xpr_base<Derived, MatrixXpr, SolverStorage>
+{
+  typedef SolverBase<Derived> type;
+
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SOLVERBASE_H
diff --git a/Eigen/src/Core/StableNorm.h b/Eigen/src/Core/StableNorm.h
index 389d942..88c8d98 100644
--- a/Eigen/src/Core/StableNorm.h
+++ b/Eigen/src/Core/StableNorm.h
@@ -17,10 +17,9 @@
 template<typename ExpressionType, typename Scalar>
 inline void stable_norm_kernel(const ExpressionType& bl, Scalar& ssq, Scalar& scale, Scalar& invScale)
 {
-  using std::max;
   Scalar maxCoeff = bl.cwiseAbs().maxCoeff();
   
-  if (maxCoeff>scale)
+  if(maxCoeff>scale)
   {
     ssq = ssq * numext::abs2(scale/maxCoeff);
     Scalar tmp = Scalar(1)/maxCoeff;
@@ -29,12 +28,21 @@
       invScale = NumTraits<Scalar>::highest();
       scale = Scalar(1)/invScale;
     }
+    else if(maxCoeff>NumTraits<Scalar>::highest()) // we got a INF
+    {
+      invScale = Scalar(1);
+      scale = maxCoeff;
+    }
     else
     {
       scale = maxCoeff;
       invScale = tmp;
     }
   }
+  else if(maxCoeff!=maxCoeff) // we got a NaN
+  {
+    scale = maxCoeff;
+  }
   
   // TODO if the maxCoeff is much much smaller than the current scale,
   // then we can neglect this sub vector
@@ -47,15 +55,12 @@
 blueNorm_impl(const EigenBase<Derived>& _vec)
 {
   typedef typename Derived::RealScalar RealScalar;  
-  typedef typename Derived::Index Index;
   using std::pow;
-  using std::min;
-  using std::max;
   using std::sqrt;
   using std::abs;
   const Derived& vec(_vec.derived());
   static bool initialized = false;
-  static RealScalar b1, b2, s1m, s2m, overfl, rbig, relerr;
+  static RealScalar b1, b2, s1m, s2m, rbig, relerr;
   if(!initialized)
   {
     int ibeta, it, iemin, iemax, iexp;
@@ -84,7 +89,6 @@
     iexp  = - ((iemax+it)/2);
     s2m   = RealScalar(pow(RealScalar(ibeta),RealScalar(iexp)));    // scaling factor for upper range
 
-    overfl  = rbig*s2m;                                             // overflow boundary for abig
     eps     = RealScalar(pow(double(ibeta), 1-it));
     relerr  = sqrt(eps);                                            // tolerance for neglecting asml
     initialized = true;
@@ -101,13 +105,13 @@
     else if(ax < b1) asml += numext::abs2(ax*s1m);
     else             amed += numext::abs2(ax);
   }
+  if(amed!=amed)
+    return amed;  // we got a NaN
   if(abig > RealScalar(0))
   {
     abig = sqrt(abig);
-    if(abig > overfl)
-    {
-      return rbig;
-    }
+    if(abig > rbig) // overflow, or *this contains INF values
+      return abig;  // return INF
     if(amed > RealScalar(0))
     {
       abig = abig/s2m;
@@ -128,8 +132,8 @@
   }
   else
     return sqrt(amed);
-  asml = (min)(abig, amed);
-  abig = (max)(abig, amed);
+  asml = numext::mini(abig, amed);
+  abig = numext::maxi(abig, amed);
   if(asml <= abig*relerr)
     return abig;
   else
@@ -152,21 +156,35 @@
 inline typename NumTraits<typename internal::traits<Derived>::Scalar>::Real
 MatrixBase<Derived>::stableNorm() const
 {
-  using std::min;
   using std::sqrt;
+  using std::abs;
   const Index blockSize = 4096;
   RealScalar scale(0);
   RealScalar invScale(1);
   RealScalar ssq(0); // sum of square
+  
+  typedef typename internal::nested_eval<Derived,2>::type DerivedCopy;
+  typedef typename internal::remove_all<DerivedCopy>::type DerivedCopyClean;
+  const DerivedCopy copy(derived());
+  
   enum {
-    Alignment = (int(Flags)&DirectAccessBit) || (int(Flags)&AlignedBit) ? 1 : 0
+    CanAlign = (   (int(DerivedCopyClean::Flags)&DirectAccessBit)
+                || (int(internal::evaluator<DerivedCopyClean>::Alignment)>0) // FIXME Alignment)>0 might not be enough
+               ) && (blockSize*sizeof(Scalar)*2<EIGEN_STACK_ALLOCATION_LIMIT)
+                 && (EIGEN_MAX_STATIC_ALIGN_BYTES>0) // if we cannot allocate on the stack, then let's not bother about this optimization
   };
+  typedef typename internal::conditional<CanAlign, Ref<const Matrix<Scalar,Dynamic,1,0,blockSize,1>, internal::evaluator<DerivedCopyClean>::Alignment>,
+                                                   typename DerivedCopyClean::ConstSegmentReturnType>::type SegmentWrapper;
   Index n = size();
-  Index bi = internal::first_aligned(derived());
+  
+  if(n==1)
+    return abs(this->coeff(0));
+  
+  Index bi = internal::first_default_aligned(copy);
   if (bi>0)
-    internal::stable_norm_kernel(this->head(bi), ssq, scale, invScale);
+    internal::stable_norm_kernel(copy.head(bi), ssq, scale, invScale);
   for (; bi<n; bi+=blockSize)
-    internal::stable_norm_kernel(this->segment(bi,(min)(blockSize, n - bi)).template forceAlignedAccessIf<Alignment>(), ssq, scale, invScale);
+    internal::stable_norm_kernel(SegmentWrapper(copy.segment(bi,numext::mini(blockSize, n - bi))), ssq, scale, invScale);
   return scale * sqrt(ssq);
 }
 
diff --git a/Eigen/src/Core/Stride.h b/Eigen/src/Core/Stride.h
index 1e3f5fe..513742f 100644
--- a/Eigen/src/Core/Stride.h
+++ b/Eigen/src/Core/Stride.h
@@ -31,8 +31,8 @@
   * arguments to the constructor.
   *
   * Indeed, this class takes two template parameters:
-  *  \param _OuterStrideAtCompileTime the outer stride, or Dynamic if you want to specify it at runtime.
-  *  \param _InnerStrideAtCompileTime the inner stride, or Dynamic if you want to specify it at runtime.
+  *  \tparam _OuterStrideAtCompileTime the outer stride, or Dynamic if you want to specify it at runtime.
+  *  \tparam _InnerStrideAtCompileTime the inner stride, or Dynamic if you want to specify it at runtime.
   *
   * Here is an example:
   * \include Map_general_stride.cpp
@@ -44,13 +44,14 @@
 class Stride
 {
   public:
-    typedef DenseIndex Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
     enum {
       InnerStrideAtCompileTime = _InnerStrideAtCompileTime,
       OuterStrideAtCompileTime = _OuterStrideAtCompileTime
     };
 
     /** Default constructor, for use when strides are fixed at compile time */
+    EIGEN_DEVICE_FUNC
     Stride()
       : m_outer(OuterStrideAtCompileTime), m_inner(InnerStrideAtCompileTime)
     {
@@ -58,6 +59,7 @@
     }
 
     /** Constructor allowing to pass the strides at runtime */
+    EIGEN_DEVICE_FUNC
     Stride(Index outerStride, Index innerStride)
       : m_outer(outerStride), m_inner(innerStride)
     {
@@ -65,13 +67,16 @@
     }
 
     /** Copy constructor */
+    EIGEN_DEVICE_FUNC
     Stride(const Stride& other)
       : m_outer(other.outer()), m_inner(other.inner())
     {}
 
     /** \returns the outer stride */
+    EIGEN_DEVICE_FUNC
     inline Index outer() const { return m_outer.value(); }
     /** \returns the inner stride */
+    EIGEN_DEVICE_FUNC
     inline Index inner() const { return m_inner.value(); }
 
   protected:
@@ -81,26 +86,24 @@
 
 /** \brief Convenience specialization of Stride to specify only an inner stride
   * See class Map for some examples */
-template<int Value = Dynamic>
+template<int Value>
 class InnerStride : public Stride<0, Value>
 {
     typedef Stride<0, Value> Base;
   public:
-    typedef DenseIndex Index;
-    InnerStride() : Base() {}
-    InnerStride(Index v) : Base(0, v) {}
+    EIGEN_DEVICE_FUNC InnerStride() : Base() {}
+    EIGEN_DEVICE_FUNC InnerStride(Index v) : Base(0, v) {} // FIXME making this explicit could break valid code
 };
 
 /** \brief Convenience specialization of Stride to specify only an outer stride
   * See class Map for some examples */
-template<int Value = Dynamic>
+template<int Value>
 class OuterStride : public Stride<Value, 0>
 {
     typedef Stride<Value, 0> Base;
   public:
-    typedef DenseIndex Index;
-    OuterStride() : Base() {}
-    OuterStride(Index v) : Base(v,0) {}
+    EIGEN_DEVICE_FUNC OuterStride() : Base() {}
+    EIGEN_DEVICE_FUNC OuterStride(Index v) : Base(v,0) {} // FIXME making this explicit could break valid code
 };
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/Swap.h b/Eigen/src/Core/Swap.h
index bf58bd5..d702009 100644
--- a/Eigen/src/Core/Swap.h
+++ b/Eigen/src/Core/Swap.h
@@ -12,115 +12,56 @@
 
 namespace Eigen { 
 
-/** \class SwapWrapper
-  * \ingroup Core_Module
-  *
-  * \internal
-  *
-  * \brief Internal helper class for swapping two expressions
-  */
 namespace internal {
-template<typename ExpressionType>
-struct traits<SwapWrapper<ExpressionType> > : traits<ExpressionType> {};
-}
 
-template<typename ExpressionType> class SwapWrapper
-  : public internal::dense_xpr_base<SwapWrapper<ExpressionType> >::type
+// Overload default assignPacket behavior for swapping them
+template<typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT>
+class generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, Specialized>
+ : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn>
 {
-  public:
-
-    typedef typename internal::dense_xpr_base<SwapWrapper>::type Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SwapWrapper)
-    typedef typename internal::packet_traits<Scalar>::type Packet;
-
-    inline SwapWrapper(ExpressionType& xpr) : m_expression(xpr) {}
-
-    inline Index rows() const { return m_expression.rows(); }
-    inline Index cols() const { return m_expression.cols(); }
-    inline Index outerStride() const { return m_expression.outerStride(); }
-    inline Index innerStride() const { return m_expression.innerStride(); }
-    
-    typedef typename internal::conditional<
-                       internal::is_lvalue<ExpressionType>::value,
-                       Scalar,
-                       const Scalar
-                     >::type ScalarWithConstIfNotLvalue;
-                     
-    inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); }
-    inline const Scalar* data() const { return m_expression.data(); }
-
-    inline Scalar& coeffRef(Index rowId, Index colId)
-    {
-      return m_expression.const_cast_derived().coeffRef(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index index)
-    {
-      return m_expression.const_cast_derived().coeffRef(index);
-    }
-
-    inline Scalar& coeffRef(Index rowId, Index colId) const
-    {
-      return m_expression.coeffRef(rowId, colId);
-    }
-
-    inline Scalar& coeffRef(Index index) const
-    {
-      return m_expression.coeffRef(index);
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index rowId, Index colId, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(rowId >= 0 && rowId < rows()
-                         && colId >= 0 && colId < cols());
-      Scalar tmp = m_expression.coeff(rowId, colId);
-      m_expression.coeffRef(rowId, colId) = _other.coeff(rowId, colId);
-      _other.coeffRef(rowId, colId) = tmp;
-    }
-
-    template<typename OtherDerived>
-    void copyCoeff(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_expression.size());
-      Scalar tmp = m_expression.coeff(index);
-      m_expression.coeffRef(index) = _other.coeff(index);
-      _other.coeffRef(index) = tmp;
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index rowId, Index colId, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(rowId >= 0 && rowId < rows()
-                        && colId >= 0 && colId < cols());
-      Packet tmp = m_expression.template packet<StoreMode>(rowId, colId);
-      m_expression.template writePacket<StoreMode>(rowId, colId,
-        _other.template packet<LoadMode>(rowId, colId)
-      );
-      _other.template writePacket<LoadMode>(rowId, colId, tmp);
-    }
-
-    template<typename OtherDerived, int StoreMode, int LoadMode>
-    void copyPacket(Index index, const DenseBase<OtherDerived>& other)
-    {
-      OtherDerived& _other = other.const_cast_derived();
-      eigen_internal_assert(index >= 0 && index < m_expression.size());
-      Packet tmp = m_expression.template packet<StoreMode>(index);
-      m_expression.template writePacket<StoreMode>(index,
-        _other.template packet<LoadMode>(index)
-      );
-      _other.template writePacket<LoadMode>(index, tmp);
-    }
-
-    ExpressionType& expression() const { return m_expression; }
-
-  protected:
-    ExpressionType& m_expression;
+protected:
+  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, swap_assign_op<typename DstEvaluatorTypeT::Scalar>, BuiltIn> Base;
+  using Base::m_dst;
+  using Base::m_src;
+  using Base::m_functor;
+  
+public:
+  typedef typename Base::Scalar Scalar;
+  typedef typename Base::DstXprType DstXprType;
+  typedef swap_assign_op<Scalar> Functor;
+  
+  EIGEN_DEVICE_FUNC generic_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr)
+    : Base(dst, src, func, dstExpr)
+  {}
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  void assignPacket(Index row, Index col)
+  {
+    PacketType tmp = m_src.template packet<LoadMode,PacketType>(row,col);
+    const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(row,col, m_dst.template packet<StoreMode,PacketType>(row,col));
+    m_dst.template writePacket<StoreMode>(row,col,tmp);
+  }
+  
+  template<int StoreMode, int LoadMode, typename PacketType>
+  void assignPacket(Index index)
+  {
+    PacketType tmp = m_src.template packet<LoadMode,PacketType>(index);
+    const_cast<SrcEvaluatorTypeT&>(m_src).template writePacket<LoadMode>(index, m_dst.template packet<StoreMode,PacketType>(index));
+    m_dst.template writePacket<StoreMode>(index,tmp);
+  }
+  
+  // TODO find a simple way not to have to copy/paste this function from generic_dense_assignment_kernel, by simple I mean no CRTP (Gael)
+  template<int StoreMode, int LoadMode, typename PacketType>
+  void assignPacketByOuterInner(Index outer, Index inner)
+  {
+    Index row = Base::rowIndexByOuterInner(outer, inner); 
+    Index col = Base::colIndexByOuterInner(outer, inner);
+    assignPacket<StoreMode,LoadMode,PacketType>(row, col);
+  }
 };
 
+} // namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SWAP_H
diff --git a/Eigen/src/Core/Transpose.h b/Eigen/src/Core/Transpose.h
index 22096ea..79b767b 100644
--- a/Eigen/src/Core/Transpose.h
+++ b/Eigen/src/Core/Transpose.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2009-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -13,39 +13,21 @@
 
 namespace Eigen { 
 
-/** \class Transpose
-  * \ingroup Core_Module
-  *
-  * \brief Expression of the transpose of a matrix
-  *
-  * \param MatrixType the type of the object of which we are taking the transpose
-  *
-  * This class represents an expression of the transpose of a matrix.
-  * It is the return type of MatrixBase::transpose() and MatrixBase::adjoint()
-  * and most of the time this is the only way it is used.
-  *
-  * \sa MatrixBase::transpose(), MatrixBase::adjoint()
-  */
-
 namespace internal {
 template<typename MatrixType>
-struct traits<Transpose<MatrixType> > : traits<MatrixType>
+struct traits<Transpose<MatrixType> > : public traits<MatrixType>
 {
-  typedef typename MatrixType::Scalar Scalar;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type MatrixTypeNestedPlain;
-  typedef typename traits<MatrixType>::StorageKind StorageKind;
-  typedef typename traits<MatrixType>::XprKind XprKind;
   enum {
     RowsAtCompileTime = MatrixType::ColsAtCompileTime,
     ColsAtCompileTime = MatrixType::RowsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxColsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
-    Flags0 = MatrixTypeNestedPlain::Flags & ~(LvalueBit | NestByRefBit),
+    Flags0 = traits<MatrixTypeNestedPlain>::Flags & ~(LvalueBit | NestByRefBit),
     Flags1 = Flags0 | FlagsLvalueBit,
     Flags = Flags1 ^ RowMajorBit,
-    CoeffReadCost = MatrixTypeNestedPlain::CoeffReadCost,
     InnerStrideAtCompileTime = inner_stride_at_compile_time<MatrixType>::ret,
     OuterStrideAtCompileTime = outer_stride_at_compile_time<MatrixType>::ret
   };
@@ -54,31 +36,55 @@
 
 template<typename MatrixType, typename StorageKind> class TransposeImpl;
 
+/** \class Transpose
+  * \ingroup Core_Module
+  *
+  * \brief Expression of the transpose of a matrix
+  *
+  * \tparam MatrixType the type of the object of which we are taking the transpose
+  *
+  * This class represents an expression of the transpose of a matrix.
+  * It is the return type of MatrixBase::transpose() and MatrixBase::adjoint()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::transpose(), MatrixBase::adjoint()
+  */
 template<typename MatrixType> class Transpose
   : public TransposeImpl<MatrixType,typename internal::traits<MatrixType>::StorageKind>
 {
   public:
 
+    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
+
     typedef typename TransposeImpl<MatrixType,typename internal::traits<MatrixType>::StorageKind>::Base Base;
     EIGEN_GENERIC_PUBLIC_INTERFACE(Transpose)
+    typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-    inline Transpose(MatrixType& a_matrix) : m_matrix(a_matrix) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline Transpose(MatrixType& matrix) : m_matrix(matrix) {}
 
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Transpose)
 
-    inline Index rows() const { return m_matrix.cols(); }
-    inline Index cols() const { return m_matrix.rows(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.cols(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.rows(); }
 
     /** \returns the nested expression */
-    const typename internal::remove_all<typename MatrixType::Nested>::type&
+    EIGEN_DEVICE_FUNC
+    const typename internal::remove_all<MatrixTypeNested>::type&
     nestedExpression() const { return m_matrix; }
 
     /** \returns the nested expression */
-    typename internal::remove_all<typename MatrixType::Nested>::type&
-    nestedExpression() { return m_matrix.const_cast_derived(); }
+    EIGEN_DEVICE_FUNC
+    typename internal::remove_reference<MatrixTypeNested>::type&
+    nestedExpression() { return m_matrix; }
+
+    /** \internal */
+    void resize(Index nrows, Index ncols) {
+      m_matrix.resize(ncols,nrows);
+    }
 
   protected:
-    typename MatrixType::Nested m_matrix;
+    typename internal::ref_selector<MatrixType>::non_const_type m_matrix;
 };
 
 namespace internal {
@@ -97,17 +103,27 @@
 
 } // end namespace internal
 
+// Generic API dispatcher
+template<typename XprType, typename StorageKind>
+class TransposeImpl
+  : public internal::generic_xpr_base<Transpose<XprType> >::type
+{
+public:
+  typedef typename internal::generic_xpr_base<Transpose<XprType> >::type Base;
+};
+
 template<typename MatrixType> class TransposeImpl<MatrixType,Dense>
   : public internal::TransposeImpl_base<MatrixType>::type
 {
   public:
 
     typedef typename internal::TransposeImpl_base<MatrixType>::type Base;
+    using Base::coeffRef;
     EIGEN_DENSE_PUBLIC_INTERFACE(Transpose<MatrixType>)
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(TransposeImpl)
 
-    inline Index innerStride() const { return derived().nestedExpression().innerStride(); }
-    inline Index outerStride() const { return derived().nestedExpression().outerStride(); }
+    EIGEN_DEVICE_FUNC inline Index innerStride() const { return derived().nestedExpression().innerStride(); }
+    EIGEN_DEVICE_FUNC inline Index outerStride() const { return derived().nestedExpression().outerStride(); }
 
     typedef typename internal::conditional<
                        internal::is_lvalue<MatrixType>::value,
@@ -115,64 +131,21 @@
                        const Scalar
                      >::type ScalarWithConstIfNotLvalue;
 
-    inline ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
-    inline const Scalar* data() const { return derived().nestedExpression().data(); }
+    EIGEN_DEVICE_FUNC inline ScalarWithConstIfNotLvalue* data() { return derived().nestedExpression().data(); }
+    EIGEN_DEVICE_FUNC inline const Scalar* data() const { return derived().nestedExpression().data(); }
 
-    inline ScalarWithConstIfNotLvalue& coeffRef(Index rowId, Index colId)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return derived().nestedExpression().const_cast_derived().coeffRef(colId, rowId);
-    }
-
-    inline ScalarWithConstIfNotLvalue& coeffRef(Index index)
-    {
-      EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return derived().nestedExpression().const_cast_derived().coeffRef(index);
-    }
-
+    // FIXME: shall we keep the const version of coeffRef?
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index rowId, Index colId) const
     {
       return derived().nestedExpression().coeffRef(colId, rowId);
     }
 
+    EIGEN_DEVICE_FUNC
     inline const Scalar& coeffRef(Index index) const
     {
       return derived().nestedExpression().coeffRef(index);
     }
-
-    inline CoeffReturnType coeff(Index rowId, Index colId) const
-    {
-      return derived().nestedExpression().coeff(colId, rowId);
-    }
-
-    inline CoeffReturnType coeff(Index index) const
-    {
-      return derived().nestedExpression().coeff(index);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index rowId, Index colId) const
-    {
-      return derived().nestedExpression().template packet<LoadMode>(colId, rowId);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index rowId, Index colId, const PacketScalar& x)
-    {
-      derived().nestedExpression().const_cast_derived().template writePacket<LoadMode>(colId, rowId, x);
-    }
-
-    template<int LoadMode>
-    inline const PacketScalar packet(Index index) const
-    {
-      return derived().nestedExpression().template packet<LoadMode>(index);
-    }
-
-    template<int LoadMode>
-    inline void writePacket(Index index, const PacketScalar& x)
-    {
-      derived().nestedExpression().const_cast_derived().template writePacket<LoadMode>(index, x);
-    }
 };
 
 /** \returns an expression of the transpose of *this.
@@ -198,7 +171,7 @@
 inline Transpose<Derived>
 DenseBase<Derived>::transpose()
 {
-  return derived();
+  return TransposeReturnType(derived());
 }
 
 /** This is the const version of transpose().
@@ -236,8 +209,7 @@
 inline const typename MatrixBase<Derived>::AdjointReturnType
 MatrixBase<Derived>::adjoint() const
 {
-  return this->transpose(); // in the complex case, the .conjugate() is be implicit here
-                            // due to implicit conversion to return type
+  return AdjointReturnType(this->transpose());
 }
 
 /***************************************************************************
@@ -247,18 +219,38 @@
 namespace internal {
 
 template<typename MatrixType,
-  bool IsSquare = (MatrixType::RowsAtCompileTime == MatrixType::ColsAtCompileTime) && MatrixType::RowsAtCompileTime!=Dynamic>
+  bool IsSquare = (MatrixType::RowsAtCompileTime == MatrixType::ColsAtCompileTime) && MatrixType::RowsAtCompileTime!=Dynamic,
+  bool MatchPacketSize =
+        (int(MatrixType::RowsAtCompileTime) == int(internal::packet_traits<typename MatrixType::Scalar>::size))
+    &&  (internal::evaluator<MatrixType>::Flags&PacketAccessBit) >
 struct inplace_transpose_selector;
 
 template<typename MatrixType>
-struct inplace_transpose_selector<MatrixType,true> { // square matrix
+struct inplace_transpose_selector<MatrixType,true,false> { // square matrix
   static void run(MatrixType& m) {
     m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose());
   }
 };
 
+// TODO: vectorized path is currently limited to LargestPacketSize x LargestPacketSize cases only.
 template<typename MatrixType>
-struct inplace_transpose_selector<MatrixType,false> { // non square matrix
+struct inplace_transpose_selector<MatrixType,true,true> { // PacketSize x PacketSize
+  static void run(MatrixType& m) {
+    typedef typename MatrixType::Scalar Scalar;
+    typedef typename internal::packet_traits<typename MatrixType::Scalar>::type Packet;
+    const Index PacketSize = internal::packet_traits<Scalar>::size;
+    const Index Alignment = internal::evaluator<MatrixType>::Alignment;
+    PacketBlock<Packet> A;
+    for (Index i=0; i<PacketSize; ++i)
+      A.packet[i] = m.template packetByOuterInner<Alignment>(i,0);
+    internal::ptranspose(A);
+    for (Index i=0; i<PacketSize; ++i)
+      m.template writePacket<Alignment>(m.rowIndexByOuterInner(i,0), m.colIndexByOuterInner(i,0), A.packet[i]);
+  }
+};
+
+template<typename MatrixType,bool MatchPacketSize>
+struct inplace_transpose_selector<MatrixType,false,MatchPacketSize> { // non square matrix
   static void run(MatrixType& m) {
     if (m.rows()==m.cols())
       m.matrix().template triangularView<StrictlyUpper>().swap(m.matrix().transpose());
@@ -331,14 +323,6 @@
 
 namespace internal {
 
-template<typename BinOp,typename NestedXpr,typename Rhs>
-struct blas_traits<SelfCwiseBinaryOp<BinOp,NestedXpr,Rhs> >
- : blas_traits<NestedXpr>
-{
-  typedef SelfCwiseBinaryOp<BinOp,NestedXpr,Rhs> XprType;
-  static inline const XprType extract(const XprType& x) { return x; }
-};
-
 template<bool DestIsTransposed, typename OtherDerived>
 struct check_transpose_aliasing_compile_time_selector
 {
@@ -404,15 +388,15 @@
     }
 };
 
+template<typename Dst, typename Src>
+void check_for_aliasing(const Dst &dst, const Src &src)
+{
+  internal::checkTransposeAliasing_impl<Dst, Src>::run(dst, src);
+}
+
 } // end namespace internal
 
-template<typename Derived>
-template<typename OtherDerived>
-void DenseBase<Derived>::checkTransposeAliasing(const OtherDerived& other) const
-{
-    internal::checkTransposeAliasing_impl<Derived, OtherDerived>::run(derived(), other);
-}
-#endif
+#endif // EIGEN_NO_DEBUG
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/Core/Transpositions.h b/Eigen/src/Core/Transpositions.h
index e4ba075..86da5af 100644
--- a/Eigen/src/Core/Transpositions.h
+++ b/Eigen/src/Core/Transpositions.h
@@ -12,39 +12,6 @@
 
 namespace Eigen { 
 
-/** \class Transpositions
-  * \ingroup Core_Module
-  *
-  * \brief Represents a sequence of transpositions (row/column interchange)
-  *
-  * \param SizeAtCompileTime the number of transpositions, or Dynamic
-  * \param MaxSizeAtCompileTime the maximum number of transpositions, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
-  *
-  * This class represents a permutation transformation as a sequence of \em n transpositions
-  * \f$[T_{n-1} \ldots T_{i} \ldots T_{0}]\f$. It is internally stored as a vector of integers \c indices.
-  * Each transposition \f$ T_{i} \f$ applied on the left of a matrix (\f$ T_{i} M\f$) interchanges
-  * the rows \c i and \c indices[i] of the matrix \c M.
-  * A transposition applied on the right (e.g., \f$ M T_{i}\f$) yields a column interchange.
-  *
-  * Compared to the class PermutationMatrix, such a sequence of transpositions is what is
-  * computed during a decomposition with pivoting, and it is faster when applying the permutation in-place.
-  * 
-  * To apply a sequence of transpositions to a matrix, simply use the operator * as in the following example:
-  * \code
-  * Transpositions tr;
-  * MatrixXf mat;
-  * mat = tr * mat;
-  * \endcode
-  * In this example, we detect that the matrix appears on both side, and so the transpositions
-  * are applied in-place without any temporary or extra copy.
-  *
-  * \sa class PermutationMatrix
-  */
-
-namespace internal {
-template<typename TranspositionType, typename MatrixType, int Side, bool Transposed=false> struct transposition_matrix_product_retval;
-}
-
 template<typename Derived>
 class TranspositionsBase
 {
@@ -53,7 +20,8 @@
   public:
 
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     Derived& derived() { return *static_cast<Derived*>(this); }
     const Derived& derived() const { return *static_cast<const Derived*>(this); }
@@ -65,7 +33,7 @@
       indices() = other.indices();
       return derived();
     }
-
+    
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** This is a special case of the templated operator=. Its purpose is to
       * prevent a default operator= from hiding the templated operator=.
@@ -78,20 +46,24 @@
     #endif
 
     /** \returns the number of transpositions */
-    inline Index size() const { return indices().size(); }
+    Index size() const { return indices().size(); }
+    /** \returns the number of rows of the equivalent permutation matrix */
+    Index rows() const { return indices().size(); }
+    /** \returns the number of columns of the equivalent permutation matrix */
+    Index cols() const { return indices().size(); }
 
     /** Direct access to the underlying index vector */
-    inline const Index& coeff(Index i) const { return indices().coeff(i); }
+    inline const StorageIndex& coeff(Index i) const { return indices().coeff(i); }
     /** Direct access to the underlying index vector */
-    inline Index& coeffRef(Index i) { return indices().coeffRef(i); }
+    inline StorageIndex& coeffRef(Index i) { return indices().coeffRef(i); }
     /** Direct access to the underlying index vector */
-    inline const Index& operator()(Index i) const { return indices()(i); }
+    inline const StorageIndex& operator()(Index i) const { return indices()(i); }
     /** Direct access to the underlying index vector */
-    inline Index& operator()(Index i) { return indices()(i); }
+    inline StorageIndex& operator()(Index i) { return indices()(i); }
     /** Direct access to the underlying index vector */
-    inline const Index& operator[](Index i) const { return indices()(i); }
+    inline const StorageIndex& operator[](Index i) const { return indices()(i); }
     /** Direct access to the underlying index vector */
-    inline Index& operator[](Index i) { return indices()(i); }
+    inline StorageIndex& operator[](Index i) { return indices()(i); }
 
     /** const version of indices(). */
     const IndicesType& indices() const { return derived().indices(); }
@@ -99,7 +71,7 @@
     IndicesType& indices() { return derived().indices(); }
 
     /** Resizes to given size. */
-    inline void resize(int newSize)
+    inline void resize(Index newSize)
     {
       indices().resize(newSize);
     }
@@ -107,7 +79,7 @@
     /** Sets \c *this to represents an identity transformation */
     void setIdentity()
     {
-      for(int i = 0; i < indices().size(); ++i)
+      for(StorageIndex i = 0; i < indices().size(); ++i)
         coeffRef(i) = i;
     }
 
@@ -144,23 +116,53 @@
 };
 
 namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-struct traits<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+struct traits<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
+ : traits<PermutationMatrix<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
 {
-  typedef IndexType Index;
-  typedef Matrix<Index, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+  typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType;
+  typedef TranspositionsStorage StorageKind;
 };
 }
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType>
-class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType> >
+/** \class Transpositions
+  * \ingroup Core_Module
+  *
+  * \brief Represents a sequence of transpositions (row/column interchange)
+  *
+  * \tparam SizeAtCompileTime the number of transpositions, or Dynamic
+  * \tparam MaxSizeAtCompileTime the maximum number of transpositions, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it.
+  *
+  * This class represents a permutation transformation as a sequence of \em n transpositions
+  * \f$[T_{n-1} \ldots T_{i} \ldots T_{0}]\f$. It is internally stored as a vector of integers \c indices.
+  * Each transposition \f$ T_{i} \f$ applied on the left of a matrix (\f$ T_{i} M\f$) interchanges
+  * the rows \c i and \c indices[i] of the matrix \c M.
+  * A transposition applied on the right (e.g., \f$ M T_{i}\f$) yields a column interchange.
+  *
+  * Compared to the class PermutationMatrix, such a sequence of transpositions is what is
+  * computed during a decomposition with pivoting, and it is faster when applying the permutation in-place.
+  *
+  * To apply a sequence of transpositions to a matrix, simply use the operator * as in the following example:
+  * \code
+  * Transpositions tr;
+  * MatrixXf mat;
+  * mat = tr * mat;
+  * \endcode
+  * In this example, we detect that the matrix appears on both side, and so the transpositions
+  * are applied in-place without any temporary or extra copy.
+  *
+  * \sa class PermutationMatrix
+  */
+
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex>
+class Transpositions : public TranspositionsBase<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
 {
     typedef internal::traits<Transpositions> Traits;
   public:
 
     typedef TranspositionsBase<Transpositions> Base;
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
 
     inline Transpositions() {}
 
@@ -177,7 +179,7 @@
 
     /** Generic constructor from expression of the transposition indices. */
     template<typename Other>
-    explicit inline Transpositions(const MatrixBase<Other>& a_indices) : m_indices(a_indices)
+    explicit inline Transpositions(const MatrixBase<Other>& indices) : m_indices(indices)
     {}
 
     /** Copies the \a other transpositions into \c *this */
@@ -215,30 +217,32 @@
 
 
 namespace internal {
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int _PacketAccess>
-struct traits<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType>,_PacketAccess> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int _PacketAccess>
+struct traits<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,_PacketAccess> >
+ : traits<PermutationMatrix<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex> >
 {
-  typedef IndexType Index;
-  typedef Map<const Matrix<Index,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1>, _PacketAccess> IndicesType;
+  typedef Map<const Matrix<_StorageIndex,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1>, _PacketAccess> IndicesType;
+  typedef _StorageIndex StorageIndex;
+  typedef TranspositionsStorage StorageKind;
 };
 }
 
-template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename IndexType, int PacketAccess>
-class Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType>,PacketAccess>
- : public TranspositionsBase<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,IndexType>,PacketAccess> >
+template<int SizeAtCompileTime, int MaxSizeAtCompileTime, typename _StorageIndex, int PacketAccess>
+class Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,PacketAccess>
+ : public TranspositionsBase<Map<Transpositions<SizeAtCompileTime,MaxSizeAtCompileTime,_StorageIndex>,PacketAccess> >
 {
     typedef internal::traits<Map> Traits;
   public:
 
     typedef TranspositionsBase<Map> Base;
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
 
-    inline Map(const Index* indicesPtr)
+    explicit inline Map(const StorageIndex* indicesPtr)
       : m_indices(indicesPtr)
     {}
 
-    inline Map(const Index* indicesPtr, Index size)
+    inline Map(const StorageIndex* indicesPtr, Index size)
       : m_indices(indicesPtr,size)
     {}
 
@@ -274,9 +278,9 @@
 namespace internal {
 template<typename _IndicesType>
 struct traits<TranspositionsWrapper<_IndicesType> >
+ : traits<PermutationWrapper<_IndicesType> >
 {
-  typedef typename _IndicesType::Scalar Index;
-  typedef _IndicesType IndicesType;
+  typedef TranspositionsStorage StorageKind;
 };
 }
 
@@ -289,10 +293,10 @@
 
     typedef TranspositionsBase<TranspositionsWrapper> Base;
     typedef typename Traits::IndicesType IndicesType;
-    typedef typename IndicesType::Scalar Index;
+    typedef typename IndicesType::Scalar StorageIndex;
 
-    inline TranspositionsWrapper(IndicesType& a_indices)
-      : m_indices(a_indices)
+    explicit inline TranspositionsWrapper(IndicesType& indices)
+      : m_indices(indices)
     {}
 
     /** Copies the \a other transpositions into \c *this */
@@ -321,83 +325,46 @@
 
   protected:
 
-    const typename IndicesType::Nested m_indices;
+    typename IndicesType::Nested m_indices;
 };
 
+
+
 /** \returns the \a matrix with the \a transpositions applied to the columns.
   */
-template<typename Derived, typename TranspositionsDerived>
-inline const internal::transposition_matrix_product_retval<TranspositionsDerived, Derived, OnTheRight>
-operator*(const MatrixBase<Derived>& matrix,
-          const TranspositionsBase<TranspositionsDerived> &transpositions)
+template<typename MatrixDerived, typename TranspositionsDerived>
+EIGEN_DEVICE_FUNC
+const Product<MatrixDerived, TranspositionsDerived, AliasFreeProduct>
+operator*(const MatrixBase<MatrixDerived> &matrix,
+          const TranspositionsBase<TranspositionsDerived>& transpositions)
 {
-  return internal::transposition_matrix_product_retval
-           <TranspositionsDerived, Derived, OnTheRight>
-           (transpositions.derived(), matrix.derived());
+  return Product<MatrixDerived, TranspositionsDerived, AliasFreeProduct>
+            (matrix.derived(), transpositions.derived());
 }
 
 /** \returns the \a matrix with the \a transpositions applied to the rows.
   */
-template<typename Derived, typename TranspositionDerived>
-inline const internal::transposition_matrix_product_retval
-               <TranspositionDerived, Derived, OnTheLeft>
-operator*(const TranspositionsBase<TranspositionDerived> &transpositions,
-          const MatrixBase<Derived>& matrix)
+template<typename TranspositionsDerived, typename MatrixDerived>
+EIGEN_DEVICE_FUNC
+const Product<TranspositionsDerived, MatrixDerived, AliasFreeProduct>
+operator*(const TranspositionsBase<TranspositionsDerived> &transpositions,
+          const MatrixBase<MatrixDerived>& matrix)
 {
-  return internal::transposition_matrix_product_retval
-           <TranspositionDerived, Derived, OnTheLeft>
-           (transpositions.derived(), matrix.derived());
+  return Product<TranspositionsDerived, MatrixDerived, AliasFreeProduct>
+            (transpositions.derived(), matrix.derived());
 }
 
+// Template partial specialization for transposed/inverse transpositions
+
 namespace internal {
 
-template<typename TranspositionType, typename MatrixType, int Side, bool Transposed>
-struct traits<transposition_matrix_product_retval<TranspositionType, MatrixType, Side, Transposed> >
-{
-  typedef typename MatrixType::PlainObject ReturnType;
-};
-
-template<typename TranspositionType, typename MatrixType, int Side, bool Transposed>
-struct transposition_matrix_product_retval
- : public ReturnByValue<transposition_matrix_product_retval<TranspositionType, MatrixType, Side, Transposed> >
-{
-    typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-    typedef typename TranspositionType::Index Index;
-
-    transposition_matrix_product_retval(const TranspositionType& tr, const MatrixType& matrix)
-      : m_transpositions(tr), m_matrix(matrix)
-    {}
-
-    inline int rows() const { return m_matrix.rows(); }
-    inline int cols() const { return m_matrix.cols(); }
-
-    template<typename Dest> inline void evalTo(Dest& dst) const
-    {
-      const int size = m_transpositions.size();
-      Index j = 0;
-
-      if(!(is_same<MatrixTypeNestedCleaned,Dest>::value && extract_data(dst) == extract_data(m_matrix)))
-        dst = m_matrix;
-
-      for(int k=(Transposed?size-1:0) ; Transposed?k>=0:k<size ; Transposed?--k:++k)
-        if((j=m_transpositions.coeff(k))!=k)
-        {
-          if(Side==OnTheLeft)
-            dst.row(k).swap(dst.row(j));
-          else if(Side==OnTheRight)
-            dst.col(k).swap(dst.col(j));
-        }
-    }
-
-  protected:
-    const TranspositionType& m_transpositions;
-    typename MatrixType::Nested m_matrix;
-};
+template<typename Derived>
+struct traits<Transpose<TranspositionsBase<Derived> > >
+ : traits<Derived>
+{};
 
 } // end namespace internal
 
-/* Template partial specialization for transposed/inverse transpositions */
-
 template<typename TranspositionsDerived>
 class Transpose<TranspositionsBase<TranspositionsDerived> >
 {
@@ -405,27 +372,31 @@
     typedef typename TranspositionType::IndicesType IndicesType;
   public:
 
-    Transpose(const TranspositionType& t) : m_transpositions(t) {}
+    explicit Transpose(const TranspositionType& t) : m_transpositions(t) {}
 
-    inline int size() const { return m_transpositions.size(); }
+    Index size() const { return m_transpositions.size(); }
+    Index rows() const { return m_transpositions.size(); }
+    Index cols() const { return m_transpositions.size(); }
 
     /** \returns the \a matrix with the inverse transpositions applied to the columns.
       */
-    template<typename Derived> friend
-    inline const internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheRight, true>
-    operator*(const MatrixBase<Derived>& matrix, const Transpose& trt)
+    template<typename OtherDerived> friend
+    const Product<OtherDerived, Transpose, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix, const Transpose& trt)
     {
-      return internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheRight, true>(trt.m_transpositions, matrix.derived());
+      return Product<OtherDerived, Transpose, AliasFreeProduct>(matrix.derived(), trt);
     }
 
     /** \returns the \a matrix with the inverse transpositions applied to the rows.
       */
-    template<typename Derived>
-    inline const internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheLeft, true>
-    operator*(const MatrixBase<Derived>& matrix) const
+    template<typename OtherDerived>
+    const Product<Transpose, OtherDerived, AliasFreeProduct>
+    operator*(const MatrixBase<OtherDerived>& matrix) const
     {
-      return internal::transposition_matrix_product_retval<TranspositionType, Derived, OnTheLeft, true>(m_transpositions, matrix.derived());
+      return Product<Transpose, OtherDerived, AliasFreeProduct>(*this, matrix.derived());
     }
+    
+    const TranspositionType& nestedExpression() const { return m_transpositions; }
 
   protected:
     const TranspositionType& m_transpositions;
diff --git a/Eigen/src/Core/TriangularMatrix.h b/Eigen/src/Core/TriangularMatrix.h
index 4d65392..667ef09 100644
--- a/Eigen/src/Core/TriangularMatrix.h
+++ b/Eigen/src/Core/TriangularMatrix.h
@@ -19,9 +19,7 @@
   
 }
 
-/** \internal
-  *
-  * \class TriangularBase
+/** \class TriangularBase
   * \ingroup Core_Module
   *
   * \brief Base class for triangular part in a matrix
@@ -32,41 +30,69 @@
 
     enum {
       Mode = internal::traits<Derived>::Mode,
-      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
       RowsAtCompileTime = internal::traits<Derived>::RowsAtCompileTime,
       ColsAtCompileTime = internal::traits<Derived>::ColsAtCompileTime,
       MaxRowsAtCompileTime = internal::traits<Derived>::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime
+      MaxColsAtCompileTime = internal::traits<Derived>::MaxColsAtCompileTime,
+      
+      SizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::RowsAtCompileTime,
+                                                   internal::traits<Derived>::ColsAtCompileTime>::ret),
+      /**< This is equal to the number of coefficients, i.e. the number of
+          * rows times the number of columns, or to \a Dynamic if this is not
+          * known at compile-time. \sa RowsAtCompileTime, ColsAtCompileTime */
+      
+      MaxSizeAtCompileTime = (internal::size_at_compile_time<internal::traits<Derived>::MaxRowsAtCompileTime,
+                                                   internal::traits<Derived>::MaxColsAtCompileTime>::ret)
+        
     };
     typedef typename internal::traits<Derived>::Scalar Scalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
-    typedef typename internal::traits<Derived>::DenseMatrixType DenseMatrixType;
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+    typedef typename internal::traits<Derived>::FullMatrixType DenseMatrixType;
     typedef DenseMatrixType DenseType;
+    typedef Derived const& Nested;
 
+    EIGEN_DEVICE_FUNC
     inline TriangularBase() { eigen_assert(!((Mode&UnitDiag) && (Mode&ZeroDiag))); }
 
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return derived().rows(); }
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return derived().cols(); }
+    EIGEN_DEVICE_FUNC
     inline Index outerStride() const { return derived().outerStride(); }
+    EIGEN_DEVICE_FUNC
     inline Index innerStride() const { return derived().innerStride(); }
+    
+    // dummy resize function
+    void resize(Index rows, Index cols)
+    {
+      EIGEN_UNUSED_VARIABLE(rows);
+      EIGEN_UNUSED_VARIABLE(cols);
+      eigen_assert(rows==this->rows() && cols==this->cols());
+    }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar coeff(Index row, Index col) const  { return derived().coeff(row,col); }
+    EIGEN_DEVICE_FUNC
     inline Scalar& coeffRef(Index row, Index col) { return derived().coeffRef(row,col); }
 
     /** \see MatrixBase::copyCoeff(row,col)
       */
     template<typename Other>
+    EIGEN_DEVICE_FUNC
     EIGEN_STRONG_INLINE void copyCoeff(Index row, Index col, Other& other)
     {
       derived().coeffRef(row, col) = other.coeff(row, col);
     }
 
+    EIGEN_DEVICE_FUNC
     inline Scalar operator()(Index row, Index col) const
     {
       check_coordinates(row, col);
       return coeff(row,col);
     }
+    EIGEN_DEVICE_FUNC
     inline Scalar& operator()(Index row, Index col)
     {
       check_coordinates(row, col);
@@ -74,15 +100,20 @@
     }
 
     #ifndef EIGEN_PARSED_BY_DOXYGEN
+    EIGEN_DEVICE_FUNC
     inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    EIGEN_DEVICE_FUNC
     inline Derived& derived() { return *static_cast<Derived*>(this); }
     #endif // not EIGEN_PARSED_BY_DOXYGEN
 
     template<typename DenseDerived>
+    EIGEN_DEVICE_FUNC
     void evalTo(MatrixBase<DenseDerived> &other) const;
     template<typename DenseDerived>
+    EIGEN_DEVICE_FUNC
     void evalToLazy(MatrixBase<DenseDerived> &other) const;
 
+    EIGEN_DEVICE_FUNC
     DenseMatrixType toDenseMatrix() const
     {
       DenseMatrixType res(rows(), cols());
@@ -119,17 +150,17 @@
 /** \class TriangularView
   * \ingroup Core_Module
   *
-  * \brief Base class for triangular part in a matrix
+  * \brief Expression of a triangular part in a matrix
   *
   * \param MatrixType the type of the object in which we are taking the triangular part
   * \param Mode the kind of triangular matrix expression to construct. Can be #Upper,
   *             #Lower, #UnitUpper, #UnitLower, #StrictlyUpper, or #StrictlyLower.
   *             This is in fact a bit field; it must have either #Upper or #Lower, 
-  *             and additionnaly it may have #UnitDiag or #ZeroDiag or neither.
+  *             and additionally it may have #UnitDiag or #ZeroDiag or neither.
   *
   * This class represents a triangular part of a matrix, not necessarily square. Strictly speaking, for rectangular
   * matrices one should speak of "trapezoid" parts. This class is the return type
-  * of MatrixBase::triangularView() and most of the time this is the only way it is used.
+  * of MatrixBase::triangularView() and SparseMatrixBase::triangularView(), and most of the time this is the only way it is used.
   *
   * \sa MatrixBase::triangularView()
   */
@@ -137,234 +168,144 @@
 template<typename MatrixType, unsigned int _Mode>
 struct traits<TriangularView<MatrixType, _Mode> > : traits<MatrixType>
 {
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::non_const_type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type MatrixTypeNestedNonRef;
   typedef typename remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
+  typedef typename MatrixType::PlainObject FullMatrixType;
   typedef MatrixType ExpressionType;
-  typedef typename MatrixType::PlainObject DenseMatrixType;
   enum {
     Mode = _Mode,
-    Flags = (MatrixTypeNestedCleaned::Flags & (HereditaryBits) & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit))) | Mode,
-    CoeffReadCost = MatrixTypeNestedCleaned::CoeffReadCost
+    FlagsLvalueBit = is_lvalue<MatrixType>::value ? LvalueBit : 0,
+    Flags = (MatrixTypeNestedCleaned::Flags & (HereditaryBits | FlagsLvalueBit) & (~(PacketAccessBit | DirectAccessBit | LinearAccessBit)))
   };
 };
 }
 
-template<int Mode, bool LhsIsTriangular,
-         typename Lhs, bool LhsIsVector,
-         typename Rhs, bool RhsIsVector>
-struct TriangularProduct;
+template<typename _MatrixType, unsigned int _Mode, typename StorageKind> class TriangularViewImpl;
 
 template<typename _MatrixType, unsigned int _Mode> class TriangularView
-  : public TriangularBase<TriangularView<_MatrixType, _Mode> >
+  : public TriangularViewImpl<_MatrixType, _Mode, typename internal::traits<_MatrixType>::StorageKind >
 {
   public:
 
-    typedef TriangularBase<TriangularView> Base;
+    typedef TriangularViewImpl<_MatrixType, _Mode, typename internal::traits<_MatrixType>::StorageKind > Base;
     typedef typename internal::traits<TriangularView>::Scalar Scalar;
-
     typedef _MatrixType MatrixType;
-    typedef typename internal::traits<TriangularView>::DenseMatrixType DenseMatrixType;
-    typedef DenseMatrixType PlainObject;
 
   protected:
     typedef typename internal::traits<TriangularView>::MatrixTypeNested MatrixTypeNested;
     typedef typename internal::traits<TriangularView>::MatrixTypeNestedNonRef MatrixTypeNestedNonRef;
-    typedef typename internal::traits<TriangularView>::MatrixTypeNestedCleaned MatrixTypeNestedCleaned;
 
     typedef typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type MatrixConjugateReturnType;
     
   public:
-    using Base::evalToLazy;
-  
 
     typedef typename internal::traits<TriangularView>::StorageKind StorageKind;
-    typedef typename internal::traits<TriangularView>::Index Index;
+    typedef typename internal::traits<TriangularView>::MatrixTypeNestedCleaned NestedExpression;
 
     enum {
       Mode = _Mode,
+      Flags = internal::traits<TriangularView>::Flags,
       TransposeMode = (Mode & Upper ? Lower : 0)
                     | (Mode & Lower ? Upper : 0)
                     | (Mode & (UnitDiag))
-                    | (Mode & (ZeroDiag))
+                    | (Mode & (ZeroDiag)),
+      IsVectorAtCompileTime = false
     };
 
-    inline TriangularView(const MatrixType& matrix) : m_matrix(matrix)
+    EIGEN_DEVICE_FUNC
+    explicit inline TriangularView(MatrixType& matrix) : m_matrix(matrix)
     {}
+    
+    using Base::operator=;
+    TriangularView& operator=(const TriangularView &other)
+    { return Base::operator=(other); }
 
+    /** \copydoc EigenBase::rows() */
+    EIGEN_DEVICE_FUNC
     inline Index rows() const { return m_matrix.rows(); }
+    /** \copydoc EigenBase::cols() */
+    EIGEN_DEVICE_FUNC
     inline Index cols() const { return m_matrix.cols(); }
-    inline Index outerStride() const { return m_matrix.outerStride(); }
-    inline Index innerStride() const { return m_matrix.innerStride(); }
 
-    /** \sa MatrixBase::operator+=() */
-    template<typename Other> TriangularView&  operator+=(const DenseBase<Other>& other) { return *this = m_matrix + other.derived(); }
-    /** \sa MatrixBase::operator-=() */
-    template<typename Other> TriangularView&  operator-=(const DenseBase<Other>& other) { return *this = m_matrix - other.derived(); }
-    /** \sa MatrixBase::operator*=() */
-    TriangularView&  operator*=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = m_matrix * other; }
-    /** \sa MatrixBase::operator/=() */
-    TriangularView&  operator/=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = m_matrix / other; }
+    /** \returns a const reference to the nested expression */
+    EIGEN_DEVICE_FUNC
+    const NestedExpression& nestedExpression() const { return m_matrix; }
 
-    /** \sa MatrixBase::fill() */
-    void fill(const Scalar& value) { setConstant(value); }
-    /** \sa MatrixBase::setConstant() */
-    TriangularView& setConstant(const Scalar& value)
-    { return *this = MatrixType::Constant(rows(), cols(), value); }
-    /** \sa MatrixBase::setZero() */
-    TriangularView& setZero() { return setConstant(Scalar(0)); }
-    /** \sa MatrixBase::setOnes() */
-    TriangularView& setOnes() { return setConstant(Scalar(1)); }
-
-    /** \sa MatrixBase::coeff()
-      * \warning the coordinates must fit into the referenced triangular part
-      */
-    inline Scalar coeff(Index row, Index col) const
-    {
-      Base::check_coordinates_internal(row, col);
-      return m_matrix.coeff(row, col);
-    }
-
-    /** \sa MatrixBase::coeffRef()
-      * \warning the coordinates must fit into the referenced triangular part
-      */
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      Base::check_coordinates_internal(row, col);
-      return m_matrix.const_cast_derived().coeffRef(row, col);
-    }
-
-    const MatrixTypeNestedCleaned& nestedExpression() const { return m_matrix; }
-    MatrixTypeNestedCleaned& nestedExpression() { return *const_cast<MatrixTypeNestedCleaned*>(&m_matrix); }
-
-    /** Assigns a triangular matrix to a triangular part of a dense matrix */
-    template<typename OtherDerived>
-    TriangularView& operator=(const TriangularBase<OtherDerived>& other);
-
-    template<typename OtherDerived>
-    TriangularView& operator=(const MatrixBase<OtherDerived>& other);
-
-    TriangularView& operator=(const TriangularView& other)
-    { return *this = other.nestedExpression(); }
-
-    template<typename OtherDerived>
-    void lazyAssign(const TriangularBase<OtherDerived>& other);
-
-    template<typename OtherDerived>
-    void lazyAssign(const MatrixBase<OtherDerived>& other);
-
-    /** \sa MatrixBase::conjugate() */
-    inline TriangularView<MatrixConjugateReturnType,Mode> conjugate()
-    { return m_matrix.conjugate(); }
+    /** \returns a reference to the nested expression */
+    EIGEN_DEVICE_FUNC
+    NestedExpression& nestedExpression() { return m_matrix; }
+    
+    typedef TriangularView<const MatrixConjugateReturnType,Mode> ConjugateReturnType;
     /** \sa MatrixBase::conjugate() const */
-    inline const TriangularView<MatrixConjugateReturnType,Mode> conjugate() const
-    { return m_matrix.conjugate(); }
+    EIGEN_DEVICE_FUNC
+    inline const ConjugateReturnType conjugate() const
+    { return ConjugateReturnType(m_matrix.conjugate()); }
 
+    typedef TriangularView<const typename MatrixType::AdjointReturnType,TransposeMode> AdjointReturnType;
     /** \sa MatrixBase::adjoint() const */
-    inline const TriangularView<const typename MatrixType::AdjointReturnType,TransposeMode> adjoint() const
-    { return m_matrix.adjoint(); }
+    EIGEN_DEVICE_FUNC
+    inline const AdjointReturnType adjoint() const
+    { return AdjointReturnType(m_matrix.adjoint()); }
 
-    /** \sa MatrixBase::transpose() */
-    inline TriangularView<Transpose<MatrixType>,TransposeMode> transpose()
+    typedef TriangularView<typename MatrixType::TransposeReturnType,TransposeMode> TransposeReturnType;
+     /** \sa MatrixBase::transpose() */
+    EIGEN_DEVICE_FUNC
+    inline TransposeReturnType transpose()
     {
       EIGEN_STATIC_ASSERT_LVALUE(MatrixType)
-      return m_matrix.const_cast_derived().transpose();
+      typename MatrixType::TransposeReturnType tmp(m_matrix);
+      return TransposeReturnType(tmp);
     }
+    
+    typedef TriangularView<const typename MatrixType::ConstTransposeReturnType,TransposeMode> ConstTransposeReturnType;
     /** \sa MatrixBase::transpose() const */
-    inline const TriangularView<Transpose<MatrixType>,TransposeMode> transpose() const
+    EIGEN_DEVICE_FUNC
+    inline const ConstTransposeReturnType transpose() const
     {
-      return m_matrix.transpose();
+      return ConstTransposeReturnType(m_matrix.transpose());
     }
 
-    /** Efficient triangular matrix times vector/matrix product */
-    template<typename OtherDerived>
-    TriangularProduct<Mode, true, MatrixType, false, OtherDerived, OtherDerived::ColsAtCompileTime==1>
-    operator*(const MatrixBase<OtherDerived>& rhs) const
-    {
-      return TriangularProduct
-              <Mode, true, MatrixType, false, OtherDerived, OtherDerived::ColsAtCompileTime==1>
-              (m_matrix, rhs.derived());
-    }
-
-    /** Efficient vector/matrix times triangular matrix product */
-    template<typename OtherDerived> friend
-    TriangularProduct<Mode, false, OtherDerived, OtherDerived::RowsAtCompileTime==1, MatrixType, false>
-    operator*(const MatrixBase<OtherDerived>& lhs, const TriangularView& rhs)
-    {
-      return TriangularProduct
-              <Mode, false, OtherDerived, OtherDerived::RowsAtCompileTime==1, MatrixType, false>
-              (lhs.derived(),rhs.m_matrix);
-    }
-
-    #ifdef EIGEN2_SUPPORT
-    template<typename OtherDerived>
-    struct eigen2_product_return_type
-    {
-      typedef typename TriangularView<MatrixType,Mode>::DenseMatrixType DenseMatrixType;
-      typedef typename OtherDerived::PlainObject::DenseType OtherPlainObject;
-      typedef typename ProductReturnType<DenseMatrixType, OtherPlainObject>::Type ProdRetType;
-      typedef typename ProdRetType::PlainObject type;
-    };
-    template<typename OtherDerived>
-    const typename eigen2_product_return_type<OtherDerived>::type
-    operator*(const EigenBase<OtherDerived>& rhs) const
-    {
-      typename OtherDerived::PlainObject::DenseType rhsPlainObject;
-      rhs.evalTo(rhsPlainObject);
-      return this->toDenseMatrix() * rhsPlainObject;
-    }
-    template<typename OtherMatrixType>
-    bool isApprox(const TriangularView<OtherMatrixType, Mode>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return this->toDenseMatrix().isApprox(other.toDenseMatrix(), precision);
-    }
-    template<typename OtherDerived>
-    bool isApprox(const MatrixBase<OtherDerived>& other, typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision()) const
-    {
-      return this->toDenseMatrix().isApprox(other, precision);
-    }
-    #endif // EIGEN2_SUPPORT
-
-    template<int Side, typename Other>
-    inline const internal::triangular_solve_retval<Side,TriangularView, Other>
-    solve(const MatrixBase<Other>& other) const;
-
-    template<int Side, typename OtherDerived>
-    void solveInPlace(const MatrixBase<OtherDerived>& other) const;
-
     template<typename Other>
-    inline const internal::triangular_solve_retval<OnTheLeft,TriangularView, Other> 
+    EIGEN_DEVICE_FUNC
+    inline const Solve<TriangularView, Other> 
     solve(const MatrixBase<Other>& other) const
-    { return solve<OnTheLeft>(other); }
+    { return Solve<TriangularView, Other>(*this, other.derived()); }
+    
+  // workaround MSVC ICE
+  #if EIGEN_COMP_MSVC
+    template<int Side, typename Other>
+    EIGEN_DEVICE_FUNC
+    inline const internal::triangular_solve_retval<Side,TriangularView, Other>
+    solve(const MatrixBase<Other>& other) const
+    { return Base::template solve<Side>(other); }
+  #else
+    using Base::solve;
+  #endif
 
-    template<typename OtherDerived>
-    void solveInPlace(const MatrixBase<OtherDerived>& other) const
-    { return solveInPlace<OnTheLeft>(other); }
-
-    const SelfAdjointView<MatrixTypeNestedNonRef,Mode> selfadjointView() const
-    {
-      EIGEN_STATIC_ASSERT((Mode&UnitDiag)==0,PROGRAMMING_ERROR);
-      return SelfAdjointView<MatrixTypeNestedNonRef,Mode>(m_matrix);
-    }
+    /** \returns a selfadjoint view of the referenced triangular part which must be either \c #Upper or \c #Lower.
+      *
+      * This is a shortcut for \code this->nestedExpression().selfadjointView<(*this)::Mode>() \endcode
+      * \sa MatrixBase::selfadjointView() */
+    EIGEN_DEVICE_FUNC
     SelfAdjointView<MatrixTypeNestedNonRef,Mode> selfadjointView()
     {
-      EIGEN_STATIC_ASSERT((Mode&UnitDiag)==0,PROGRAMMING_ERROR);
+      EIGEN_STATIC_ASSERT((Mode&(UnitDiag|ZeroDiag))==0,PROGRAMMING_ERROR);
       return SelfAdjointView<MatrixTypeNestedNonRef,Mode>(m_matrix);
     }
 
-    template<typename OtherDerived>
-    void swap(TriangularBase<OtherDerived> const & other)
+    /** This is the const version of selfadjointView() */
+    EIGEN_DEVICE_FUNC
+    const SelfAdjointView<MatrixTypeNestedNonRef,Mode> selfadjointView() const
     {
-      TriangularView<SwapWrapper<MatrixType>,Mode>(const_cast<MatrixType&>(m_matrix)).lazyAssign(other.derived());
+      EIGEN_STATIC_ASSERT((Mode&(UnitDiag|ZeroDiag))==0,PROGRAMMING_ERROR);
+      return SelfAdjointView<MatrixTypeNestedNonRef,Mode>(m_matrix);
     }
 
-    template<typename OtherDerived>
-    void swap(MatrixBase<OtherDerived> const & other)
-    {
-      SwapWrapper<MatrixType> swaper(const_cast<MatrixType&>(m_matrix));
-      TriangularView<SwapWrapper<MatrixType>,Mode>(swaper).lazyAssign(other.derived());
-    }
 
+    /** \returns the determinant of the triangular matrix
+      * \sa MatrixBase::determinant() */
+    EIGEN_DEVICE_FUNC
     Scalar determinant() const
     {
       if (Mode & UnitDiag)
@@ -374,262 +315,258 @@
       else
         return m_matrix.diagonal().prod();
     }
-    
-    // TODO simplify the following:
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& operator=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    {
-      setZero();
-      return assignProduct(other.derived(),1);
-    }
-    
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& operator+=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    {
-      return assignProduct(other.derived(),1);
-    }
-    
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& operator-=(const ProductBase<ProductDerived, Lhs,Rhs>& other)
-    {
-      return assignProduct(other.derived(),-1);
-    }
-    
-    
-    template<typename ProductDerived>
-    EIGEN_STRONG_INLINE TriangularView& operator=(const ScaledProduct<ProductDerived>& other)
-    {
-      setZero();
-      return assignProduct(other.derived(),other.alpha());
-    }
-    
-    template<typename ProductDerived>
-    EIGEN_STRONG_INLINE TriangularView& operator+=(const ScaledProduct<ProductDerived>& other)
-    {
-      return assignProduct(other.derived(),other.alpha());
-    }
-    
-    template<typename ProductDerived>
-    EIGEN_STRONG_INLINE TriangularView& operator-=(const ScaledProduct<ProductDerived>& other)
-    {
-      return assignProduct(other.derived(),-other.alpha());
-    }
-    
+      
   protected:
-    
-    template<typename ProductDerived, typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE TriangularView& assignProduct(const ProductBase<ProductDerived, Lhs,Rhs>& prod, const Scalar& alpha);
-    
-    template<int Mode, bool LhsIsTriangular,
-         typename Lhs, bool LhsIsVector,
-         typename Rhs, bool RhsIsVector>
-    EIGEN_STRONG_INLINE TriangularView& assignProduct(const TriangularProduct<Mode, LhsIsTriangular, Lhs, LhsIsVector, Rhs, RhsIsVector>& prod, const Scalar& alpha)
-    {
-      lazyAssign(alpha*prod.eval());
-      return *this;
-    }
 
     MatrixTypeNested m_matrix;
 };
 
+/** \ingroup Core_Module
+  *
+  * \brief Base class for a triangular part in a \b dense matrix
+  *
+  * This class is an abstract base class of class TriangularView, and objects of type TriangularViewImpl cannot be instantiated.
+  * It extends class TriangularView with additional methods which available for dense expressions only.
+  *
+  * \sa class TriangularView, MatrixBase::triangularView()
+  */
+template<typename _MatrixType, unsigned int _Mode> class TriangularViewImpl<_MatrixType,_Mode,Dense>
+  : public TriangularBase<TriangularView<_MatrixType, _Mode> >
+{
+  public:
+
+    typedef TriangularView<_MatrixType, _Mode> TriangularViewType;
+    typedef TriangularBase<TriangularViewType> Base;
+    typedef typename internal::traits<TriangularViewType>::Scalar Scalar;
+
+    typedef _MatrixType MatrixType;
+    typedef typename MatrixType::PlainObject DenseMatrixType;
+    typedef DenseMatrixType PlainObject;
+
+  public:
+    using Base::evalToLazy;
+    using Base::derived;
+
+    typedef typename internal::traits<TriangularViewType>::StorageKind StorageKind;
+
+    enum {
+      Mode = _Mode,
+      Flags = internal::traits<TriangularViewType>::Flags
+    };
+
+    /** \returns the outer-stride of the underlying dense matrix
+      * \sa DenseCoeffsBase::outerStride() */
+    EIGEN_DEVICE_FUNC
+    inline Index outerStride() const { return derived().nestedExpression().outerStride(); }
+    /** \returns the inner-stride of the underlying dense matrix
+      * \sa DenseCoeffsBase::innerStride() */
+    EIGEN_DEVICE_FUNC
+    inline Index innerStride() const { return derived().nestedExpression().innerStride(); }
+
+    /** \sa MatrixBase::operator+=() */
+    template<typename Other>
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator+=(const DenseBase<Other>& other) {
+      internal::call_assignment_no_alias(derived(), other.derived(), internal::add_assign_op<Scalar,typename Other::Scalar>());
+      return derived();
+    }
+    /** \sa MatrixBase::operator-=() */
+    template<typename Other>
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator-=(const DenseBase<Other>& other) {
+      internal::call_assignment_no_alias(derived(), other.derived(), internal::sub_assign_op<Scalar,typename Other::Scalar>());
+      return derived();
+    }
+    
+    /** \sa MatrixBase::operator*=() */
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator*=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = derived().nestedExpression() * other; }
+    /** \sa DenseBase::operator/=() */
+    EIGEN_DEVICE_FUNC
+    TriangularViewType&  operator/=(const typename internal::traits<MatrixType>::Scalar& other) { return *this = derived().nestedExpression() / other; }
+
+    /** \sa MatrixBase::fill() */
+    EIGEN_DEVICE_FUNC
+    void fill(const Scalar& value) { setConstant(value); }
+    /** \sa MatrixBase::setConstant() */
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& setConstant(const Scalar& value)
+    { return *this = MatrixType::Constant(derived().rows(), derived().cols(), value); }
+    /** \sa MatrixBase::setZero() */
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& setZero() { return setConstant(Scalar(0)); }
+    /** \sa MatrixBase::setOnes() */
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& setOnes() { return setConstant(Scalar(1)); }
+
+    /** \sa MatrixBase::coeff()
+      * \warning the coordinates must fit into the referenced triangular part
+      */
+    EIGEN_DEVICE_FUNC
+    inline Scalar coeff(Index row, Index col) const
+    {
+      Base::check_coordinates_internal(row, col);
+      return derived().nestedExpression().coeff(row, col);
+    }
+
+    /** \sa MatrixBase::coeffRef()
+      * \warning the coordinates must fit into the referenced triangular part
+      */
+    EIGEN_DEVICE_FUNC
+    inline Scalar& coeffRef(Index row, Index col)
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(TriangularViewType);
+      Base::check_coordinates_internal(row, col);
+      return derived().nestedExpression().coeffRef(row, col);
+    }
+
+    /** Assigns a triangular matrix to a triangular part of a dense matrix */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& operator=(const TriangularBase<OtherDerived>& other);
+
+    /** Shortcut for\code *this = other.other.triangularView<(*this)::Mode>() \endcode */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& operator=(const MatrixBase<OtherDerived>& other);
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    EIGEN_DEVICE_FUNC
+    TriangularViewType& operator=(const TriangularViewImpl& other)
+    { return *this = other.derived().nestedExpression(); }
+
+    /** \deprecated */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void lazyAssign(const TriangularBase<OtherDerived>& other);
+
+    /** \deprecated */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void lazyAssign(const MatrixBase<OtherDerived>& other);
+#endif
+
+    /** Efficient triangular matrix times vector/matrix product */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    const Product<TriangularViewType,OtherDerived>
+    operator*(const MatrixBase<OtherDerived>& rhs) const
+    {
+      return Product<TriangularViewType,OtherDerived>(derived(), rhs.derived());
+    }
+
+    /** Efficient vector/matrix times triangular matrix product */
+    template<typename OtherDerived> friend
+    EIGEN_DEVICE_FUNC
+    const Product<OtherDerived,TriangularViewType>
+    operator*(const MatrixBase<OtherDerived>& lhs, const TriangularViewImpl& rhs)
+    {
+      return Product<OtherDerived,TriangularViewType>(lhs.derived(),rhs.derived());
+    }
+
+    /** \returns the product of the inverse of \c *this with \a other, \a *this being triangular.
+      *
+      * This function computes the inverse-matrix matrix product inverse(\c *this) * \a other if
+      * \a Side==OnTheLeft (the default), or the right-inverse-multiply  \a other * inverse(\c *this) if
+      * \a Side==OnTheRight.
+      *
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
+      *
+      * The matrix \c *this must be triangular and invertible (i.e., all the coefficients of the
+      * diagonal must be non zero). It works as a forward (resp. backward) substitution if \c *this
+      * is an upper (resp. lower) triangular matrix.
+      *
+      * Example: \include Triangular_solve.cpp
+      * Output: \verbinclude Triangular_solve.out
+      *
+      * This function returns an expression of the inverse-multiply and can works in-place if it is assigned
+      * to the same matrix or vector \a other.
+      *
+      * For users coming from BLAS, this function (and more specifically solveInPlace()) offer
+      * all the operations supported by the \c *TRSV and \c *TRSM BLAS routines.
+      *
+      * \sa TriangularView::solveInPlace()
+      */
+    template<int Side, typename Other>
+    EIGEN_DEVICE_FUNC
+    inline const internal::triangular_solve_retval<Side,TriangularViewType, Other>
+    solve(const MatrixBase<Other>& other) const;
+
+    /** "in-place" version of TriangularView::solve() where the result is written in \a other
+      *
+      * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here.
+      * This function will const_cast it, so constness isn't honored here.
+      *
+      * Note that the template parameter \c Side can be ommitted, in which case \c Side==OnTheLeft
+      *
+      * See TriangularView:solve() for the details.
+      */
+    template<int Side, typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void solveInPlace(const MatrixBase<OtherDerived>& other) const;
+
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void solveInPlace(const MatrixBase<OtherDerived>& other) const
+    { return solveInPlace<OnTheLeft>(other); }
+
+    /** Swaps the coefficients of the common triangular parts of two matrices */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+    void swap(TriangularBase<OtherDerived> &other)
+#else
+    void swap(TriangularBase<OtherDerived> const & other)
+#endif
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(OtherDerived);
+      call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
+    }
+
+    /** \deprecated
+      * Shortcut for \code (*this).swap(other.triangularView<(*this)::Mode>()) \endcode */
+    template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
+    void swap(MatrixBase<OtherDerived> const & other)
+    {
+      EIGEN_STATIC_ASSERT_LVALUE(OtherDerived);
+      call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op<Scalar>());
+    }
+
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _solve_impl(const RhsType &rhs, DstType &dst) const {
+      if(!internal::is_same_dense(dst,rhs))
+        dst = rhs;
+      this->solveInPlace(dst);
+    }
+
+    template<typename ProductType>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE TriangularViewType& _assignProduct(const ProductType& prod, const Scalar& alpha, bool beta);
+};
+
 /***************************************************************************
 * Implementation of triangular evaluation/assignment
 ***************************************************************************/
 
-namespace internal {
-
-template<typename Derived1, typename Derived2, unsigned int Mode, int UnrollCount, bool ClearOpposite>
-struct triangular_assignment_selector
-{
-  enum {
-    col = (UnrollCount-1) / Derived1::RowsAtCompileTime,
-    row = (UnrollCount-1) % Derived1::RowsAtCompileTime
-  };
-  
-  typedef typename Derived1::Scalar Scalar;
-
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    triangular_assignment_selector<Derived1, Derived2, Mode, UnrollCount-1, ClearOpposite>::run(dst, src);
-
-    eigen_assert( Mode == Upper || Mode == Lower
-            || Mode == StrictlyUpper || Mode == StrictlyLower
-            || Mode == UnitUpper || Mode == UnitLower);
-    if((Mode == Upper && row <= col)
-    || (Mode == Lower && row >= col)
-    || (Mode == StrictlyUpper && row < col)
-    || (Mode == StrictlyLower && row > col)
-    || (Mode == UnitUpper && row < col)
-    || (Mode == UnitLower && row > col))
-      dst.copyCoeff(row, col, src);
-    else if(ClearOpposite)
-    {
-      if (Mode&UnitDiag && row==col)
-        dst.coeffRef(row, col) = Scalar(1);
-      else
-        dst.coeffRef(row, col) = Scalar(0);
-    }
-  }
-};
-
-// prevent buggy user code from causing an infinite recursion
-template<typename Derived1, typename Derived2, unsigned int Mode, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, Mode, 0, ClearOpposite>
-{
-  static inline void run(Derived1 &, const Derived2 &) {}
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, Upper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  typedef typename Derived1::Scalar Scalar;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows()-1);
-      for(Index i = 0; i <= maxi; ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-        for(Index i = maxi+1; i < dst.rows(); ++i)
-          dst.coeffRef(i, j) = Scalar(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, Lower, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      for(Index i = j; i < dst.rows(); ++i)
-        dst.copyCoeff(i, j, src);
-      Index maxi = (std::min)(j, dst.rows());
-      if (ClearOpposite)
-        for(Index i = 0; i < maxi; ++i)
-          dst.coeffRef(i, j) = static_cast<typename Derived1::Scalar>(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, StrictlyUpper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  typedef typename Derived1::Scalar Scalar;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows());
-      for(Index i = 0; i < maxi; ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-        for(Index i = maxi; i < dst.rows(); ++i)
-          dst.coeffRef(i, j) = Scalar(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, StrictlyLower, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      for(Index i = j+1; i < dst.rows(); ++i)
-        dst.copyCoeff(i, j, src);
-      Index maxi = (std::min)(j, dst.rows()-1);
-      if (ClearOpposite)
-        for(Index i = 0; i <= maxi; ++i)
-          dst.coeffRef(i, j) = static_cast<typename Derived1::Scalar>(0);
-    }
-  }
-};
-
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, UnitUpper, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows());
-      for(Index i = 0; i < maxi; ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-      {
-        for(Index i = maxi+1; i < dst.rows(); ++i)
-          dst.coeffRef(i, j) = 0;
-      }
-    }
-    dst.diagonal().setOnes();
-  }
-};
-template<typename Derived1, typename Derived2, bool ClearOpposite>
-struct triangular_assignment_selector<Derived1, Derived2, UnitLower, Dynamic, ClearOpposite>
-{
-  typedef typename Derived1::Index Index;
-  static inline void run(Derived1 &dst, const Derived2 &src)
-  {
-    for(Index j = 0; j < dst.cols(); ++j)
-    {
-      Index maxi = (std::min)(j, dst.rows());
-      for(Index i = maxi+1; i < dst.rows(); ++i)
-        dst.copyCoeff(i, j, src);
-      if (ClearOpposite)
-      {
-        for(Index i = 0; i < maxi; ++i)
-          dst.coeffRef(i, j) = 0;
-      }
-    }
-    dst.diagonal().setOnes();
-  }
-};
-
-} // end namespace internal
-
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 // FIXME should we keep that possibility
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
 inline TriangularView<MatrixType, Mode>&
-TriangularView<MatrixType, Mode>::operator=(const MatrixBase<OtherDerived>& other)
+TriangularViewImpl<MatrixType, Mode, Dense>::operator=(const MatrixBase<OtherDerived>& other)
 {
-  if(OtherDerived::Flags & EvalBeforeAssigningBit)
-  {
-    typename internal::plain_matrix_type<OtherDerived>::type other_evaluated(other.rows(), other.cols());
-    other_evaluated.template triangularView<Mode>().lazyAssign(other.derived());
-    lazyAssign(other_evaluated);
-  }
-  else
-    lazyAssign(other.derived());
-  return *this;
+  internal::call_assignment_no_alias(derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
 }
 
 // FIXME should we keep that possibility
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
-void TriangularView<MatrixType, Mode>::lazyAssign(const MatrixBase<OtherDerived>& other)
+void TriangularViewImpl<MatrixType, Mode, Dense>::lazyAssign(const MatrixBase<OtherDerived>& other)
 {
-  enum {
-    unroll = MatrixType::SizeAtCompileTime != Dynamic
-          && internal::traits<OtherDerived>::CoeffReadCost != Dynamic
-          && MatrixType::SizeAtCompileTime*internal::traits<OtherDerived>::CoeffReadCost/2 <= EIGEN_UNROLLING_LIMIT
-  };
-  eigen_assert(m_matrix.rows() == other.rows() && m_matrix.cols() == other.cols());
-
-  internal::triangular_assignment_selector
-    <MatrixType, OtherDerived, int(Mode),
-    unroll ? int(MatrixType::SizeAtCompileTime) : Dynamic,
-    false // do not change the opposite triangular part
-    >::run(m_matrix.const_cast_derived(), other.derived());
+  internal::call_assignment_no_alias(derived(), other.template triangularView<Mode>());
 }
 
 
@@ -637,38 +574,21 @@
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
 inline TriangularView<MatrixType, Mode>&
-TriangularView<MatrixType, Mode>::operator=(const TriangularBase<OtherDerived>& other)
+TriangularViewImpl<MatrixType, Mode, Dense>::operator=(const TriangularBase<OtherDerived>& other)
 {
   eigen_assert(Mode == int(OtherDerived::Mode));
-  if(internal::traits<OtherDerived>::Flags & EvalBeforeAssigningBit)
-  {
-    typename OtherDerived::DenseMatrixType other_evaluated(other.rows(), other.cols());
-    other_evaluated.template triangularView<Mode>().lazyAssign(other.derived().nestedExpression());
-    lazyAssign(other_evaluated);
-  }
-  else
-    lazyAssign(other.derived().nestedExpression());
-  return *this;
+  internal::call_assignment(derived(), other.derived());
+  return derived();
 }
 
 template<typename MatrixType, unsigned int Mode>
 template<typename OtherDerived>
-void TriangularView<MatrixType, Mode>::lazyAssign(const TriangularBase<OtherDerived>& other)
+void TriangularViewImpl<MatrixType, Mode, Dense>::lazyAssign(const TriangularBase<OtherDerived>& other)
 {
-  enum {
-    unroll = MatrixType::SizeAtCompileTime != Dynamic
-                   && internal::traits<OtherDerived>::CoeffReadCost != Dynamic
-                   && MatrixType::SizeAtCompileTime * internal::traits<OtherDerived>::CoeffReadCost / 2
-                        <= EIGEN_UNROLLING_LIMIT
-  };
-  eigen_assert(m_matrix.rows() == other.rows() && m_matrix.cols() == other.cols());
-
-  internal::triangular_assignment_selector
-    <MatrixType, OtherDerived, int(Mode),
-    unroll ? int(MatrixType::SizeAtCompileTime) : Dynamic,
-    false // preserve the opposite triangular part
-    >::run(m_matrix.const_cast_derived(), other.derived().nestedExpression());
+  eigen_assert(Mode == int(OtherDerived::Mode));
+  internal::call_assignment_no_alias(derived(), other.derived());
 }
+#endif
 
 /***************************************************************************
 * Implementation of TriangularBase methods
@@ -680,35 +600,7 @@
 template<typename DenseDerived>
 void TriangularBase<Derived>::evalTo(MatrixBase<DenseDerived> &other) const
 {
-  if(internal::traits<Derived>::Flags & EvalBeforeAssigningBit)
-  {
-    typename internal::plain_matrix_type<Derived>::type other_evaluated(rows(), cols());
-    evalToLazy(other_evaluated);
-    other.derived().swap(other_evaluated);
-  }
-  else
-    evalToLazy(other.derived());
-}
-
-/** Assigns a triangular or selfadjoint matrix to a dense matrix.
-  * If the matrix is triangular, the opposite part is set to zero. */
-template<typename Derived>
-template<typename DenseDerived>
-void TriangularBase<Derived>::evalToLazy(MatrixBase<DenseDerived> &other) const
-{
-  enum {
-    unroll = DenseDerived::SizeAtCompileTime != Dynamic
-                   && internal::traits<Derived>::CoeffReadCost != Dynamic
-                   && DenseDerived::SizeAtCompileTime * internal::traits<Derived>::CoeffReadCost / 2
-                        <= EIGEN_UNROLLING_LIMIT
-  };
-  other.derived().resize(this->rows(), this->cols());
-
-  internal::triangular_assignment_selector
-    <DenseDerived, typename internal::traits<Derived>::MatrixTypeNestedCleaned, Derived::Mode,
-    unroll ? int(DenseDerived::SizeAtCompileTime) : Dynamic,
-    true // clear the opposite triangular part
-    >::run(other.derived(), derived().nestedExpression());
+  evalToLazy(other.derived());
 }
 
 /***************************************************************************
@@ -719,49 +611,14 @@
 * Implementation of MatrixBase methods
 ***************************************************************************/
 
-#ifdef EIGEN2_SUPPORT
-
-// implementation of part<>(), including the SelfAdjoint case.
-
-namespace internal {
-template<typename MatrixType, unsigned int Mode>
-struct eigen2_part_return_type
-{
-  typedef TriangularView<MatrixType, Mode> type;
-};
-
-template<typename MatrixType>
-struct eigen2_part_return_type<MatrixType, SelfAdjoint>
-{
-  typedef SelfAdjointView<MatrixType, Upper> type;
-};
-}
-
-/** \deprecated use MatrixBase::triangularView() */
-template<typename Derived>
-template<unsigned int Mode>
-const typename internal::eigen2_part_return_type<Derived, Mode>::type MatrixBase<Derived>::part() const
-{
-  return derived();
-}
-
-/** \deprecated use MatrixBase::triangularView() */
-template<typename Derived>
-template<unsigned int Mode>
-typename internal::eigen2_part_return_type<Derived, Mode>::type MatrixBase<Derived>::part()
-{
-  return derived();
-}
-#endif
-
 /**
   * \returns an expression of a triangular view extracted from the current matrix
   *
   * The parameter \a Mode can have the following values: \c #Upper, \c #StrictlyUpper, \c #UnitUpper,
   * \c #Lower, \c #StrictlyLower, \c #UnitLower.
   *
-  * Example: \include MatrixBase_extract.cpp
-  * Output: \verbinclude MatrixBase_extract.out
+  * Example: \include MatrixBase_triangularView.cpp
+  * Output: \verbinclude MatrixBase_triangularView.out
   *
   * \sa class TriangularView
   */
@@ -770,7 +627,7 @@
 typename MatrixBase<Derived>::template TriangularViewReturnType<Mode>::Type
 MatrixBase<Derived>::triangularView()
 {
-  return derived();
+  return typename TriangularViewReturnType<Mode>::Type(derived());
 }
 
 /** This is the const version of MatrixBase::triangularView() */
@@ -779,7 +636,7 @@
 typename MatrixBase<Derived>::template ConstTriangularViewReturnType<Mode>::Type
 MatrixBase<Derived>::triangularView() const
 {
-  return derived();
+  return typename ConstTriangularViewReturnType<Mode>::Type(derived());
 }
 
 /** \returns true if *this is approximately equal to an upper triangular matrix,
@@ -790,21 +647,20 @@
 template<typename Derived>
 bool MatrixBase<Derived>::isUpperTriangular(const RealScalar& prec) const
 {
-  using std::abs;
   RealScalar maxAbsOnUpperPart = static_cast<RealScalar>(-1);
   for(Index j = 0; j < cols(); ++j)
   {
-    Index maxi = (std::min)(j, rows()-1);
+    Index maxi = numext::mini(j, rows()-1);
     for(Index i = 0; i <= maxi; ++i)
     {
-      RealScalar absValue = abs(coeff(i,j));
+      RealScalar absValue = numext::abs(coeff(i,j));
       if(absValue > maxAbsOnUpperPart) maxAbsOnUpperPart = absValue;
     }
   }
   RealScalar threshold = maxAbsOnUpperPart * prec;
   for(Index j = 0; j < cols(); ++j)
     for(Index i = j+1; i < rows(); ++i)
-      if(abs(coeff(i, j)) > threshold) return false;
+      if(numext::abs(coeff(i, j)) > threshold) return false;
   return true;
 }
 
@@ -816,24 +672,312 @@
 template<typename Derived>
 bool MatrixBase<Derived>::isLowerTriangular(const RealScalar& prec) const
 {
-  using std::abs;
   RealScalar maxAbsOnLowerPart = static_cast<RealScalar>(-1);
   for(Index j = 0; j < cols(); ++j)
     for(Index i = j; i < rows(); ++i)
     {
-      RealScalar absValue = abs(coeff(i,j));
+      RealScalar absValue = numext::abs(coeff(i,j));
       if(absValue > maxAbsOnLowerPart) maxAbsOnLowerPart = absValue;
     }
   RealScalar threshold = maxAbsOnLowerPart * prec;
   for(Index j = 1; j < cols(); ++j)
   {
-    Index maxi = (std::min)(j, rows()-1);
+    Index maxi = numext::mini(j, rows()-1);
     for(Index i = 0; i < maxi; ++i)
-      if(abs(coeff(i, j)) > threshold) return false;
+      if(numext::abs(coeff(i, j)) > threshold) return false;
   }
   return true;
 }
 
+
+/***************************************************************************
+****************************************************************************
+* Evaluators and Assignment of triangular expressions
+***************************************************************************
+***************************************************************************/
+
+namespace internal {
+
+  
+// TODO currently a triangular expression has the form TriangularView<.,.>
+//      in the future triangular-ness should be defined by the expression traits
+//      such that Transpose<TriangularView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
+template<typename MatrixType, unsigned int Mode>
+struct evaluator_traits<TriangularView<MatrixType,Mode> >
+{
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef typename glue_shapes<typename evaluator_traits<MatrixType>::Shape, TriangularShape>::type Shape;
+};
+
+template<typename MatrixType, unsigned int Mode>
+struct unary_evaluator<TriangularView<MatrixType,Mode>, IndexBased>
+ : evaluator<typename internal::remove_all<MatrixType>::type>
+{
+  typedef TriangularView<MatrixType,Mode> XprType;
+  typedef evaluator<typename internal::remove_all<MatrixType>::type> Base;
+  unary_evaluator(const XprType &xpr) : Base(xpr.nestedExpression()) {}
+};
+
+// Additional assignment kinds:
+struct Triangular2Triangular    {};
+struct Triangular2Dense         {};
+struct Dense2Triangular         {};
+
+
+template<typename Kernel, unsigned int Mode, int UnrollCount, bool ClearOpposite> struct triangular_assignment_loop;
+
+ 
+/** \internal Specialization of the dense assignment kernel for triangular matrices.
+  * The main difference is that the triangular, diagonal, and opposite parts are processed through three different functions.
+  * \tparam UpLo must be either Lower or Upper
+  * \tparam Mode must be either 0, UnitDiag, ZeroDiag, or SelfAdjoint
+  */
+template<int UpLo, int Mode, int SetOpposite, typename DstEvaluatorTypeT, typename SrcEvaluatorTypeT, typename Functor, int Version = Specialized>
+class triangular_dense_assignment_kernel : public generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version>
+{
+protected:
+  typedef generic_dense_assignment_kernel<DstEvaluatorTypeT, SrcEvaluatorTypeT, Functor, Version> Base;
+  typedef typename Base::DstXprType DstXprType;
+  typedef typename Base::SrcXprType SrcXprType;
+  using Base::m_dst;
+  using Base::m_src;
+  using Base::m_functor;
+public:
+  
+  typedef typename Base::DstEvaluatorType DstEvaluatorType;
+  typedef typename Base::SrcEvaluatorType SrcEvaluatorType;
+  typedef typename Base::Scalar Scalar;
+  typedef typename Base::AssignmentTraits AssignmentTraits;
+  
+  
+  EIGEN_DEVICE_FUNC triangular_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr)
+    : Base(dst, src, func, dstExpr)
+  {}
+  
+#ifdef EIGEN_INTERNAL_DEBUGGING
+  EIGEN_DEVICE_FUNC void assignCoeff(Index row, Index col)
+  {
+    eigen_internal_assert(row!=col);
+    Base::assignCoeff(row,col);
+  }
+#else
+  using Base::assignCoeff;
+#endif
+  
+  EIGEN_DEVICE_FUNC void assignDiagonalCoeff(Index id)
+  {
+         if(Mode==UnitDiag && SetOpposite) m_functor.assignCoeff(m_dst.coeffRef(id,id), Scalar(1));
+    else if(Mode==ZeroDiag && SetOpposite) m_functor.assignCoeff(m_dst.coeffRef(id,id), Scalar(0));
+    else if(Mode==0)                       Base::assignCoeff(id,id);
+  }
+  
+  EIGEN_DEVICE_FUNC void assignOppositeCoeff(Index row, Index col)
+  { 
+    eigen_internal_assert(row!=col);
+    if(SetOpposite)
+      m_functor.assignCoeff(m_dst.coeffRef(row,col), Scalar(0));
+  }
+};
+
+template<int Mode, bool SetOpposite, typename DstXprType, typename SrcXprType, typename Functor>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_triangular_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func)
+{
+  typedef evaluator<DstXprType> DstEvaluatorType;
+  typedef evaluator<SrcXprType> SrcEvaluatorType;
+
+  SrcEvaluatorType srcEvaluator(src);
+
+  Index dstRows = src.rows();
+  Index dstCols = src.cols();
+  if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+    dst.resize(dstRows, dstCols);
+  DstEvaluatorType dstEvaluator(dst);
+    
+  typedef triangular_dense_assignment_kernel< Mode&(Lower|Upper),Mode&(UnitDiag|ZeroDiag|SelfAdjoint),SetOpposite,
+                                              DstEvaluatorType,SrcEvaluatorType,Functor> Kernel;
+  Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived());
+  
+  enum {
+      unroll = DstXprType::SizeAtCompileTime != Dynamic
+            && SrcEvaluatorType::CoeffReadCost < HugeCost
+            && DstXprType::SizeAtCompileTime * (DstEvaluatorType::CoeffReadCost+SrcEvaluatorType::CoeffReadCost) / 2 <= EIGEN_UNROLLING_LIMIT
+    };
+  
+  triangular_assignment_loop<Kernel, Mode, unroll ? int(DstXprType::SizeAtCompileTime) : Dynamic, SetOpposite>::run(kernel);
+}
+
+template<int Mode, bool SetOpposite, typename DstXprType, typename SrcXprType>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+void call_triangular_assignment_loop(DstXprType& dst, const SrcXprType& src)
+{
+  call_triangular_assignment_loop<Mode,SetOpposite>(dst, src, internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>());
+}
+
+template<> struct AssignmentKind<TriangularShape,TriangularShape> { typedef Triangular2Triangular Kind; };
+template<> struct AssignmentKind<DenseShape,TriangularShape>      { typedef Triangular2Dense      Kind; };
+template<> struct AssignmentKind<TriangularShape,DenseShape>      { typedef Dense2Triangular      Kind; };
+
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Triangular2Triangular>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    eigen_assert(int(DstXprType::Mode) == int(SrcXprType::Mode));
+    
+    call_triangular_assignment_loop<DstXprType::Mode, false>(dst, src, func);  
+  }
+};
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Triangular2Dense>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    call_triangular_assignment_loop<SrcXprType::Mode, (SrcXprType::Mode&SelfAdjoint)==0>(dst, src, func);  
+  }
+};
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Dense2Triangular>
+{
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    call_triangular_assignment_loop<DstXprType::Mode, false>(dst, src, func);  
+  }
+};
+
+
+template<typename Kernel, unsigned int Mode, int UnrollCount, bool SetOpposite>
+struct triangular_assignment_loop
+{
+  // FIXME: this is not very clean, perhaps this information should be provided by the kernel?
+  typedef typename Kernel::DstEvaluatorType DstEvaluatorType;
+  typedef typename DstEvaluatorType::XprType DstXprType;
+  
+  enum {
+    col = (UnrollCount-1) / DstXprType::RowsAtCompileTime,
+    row = (UnrollCount-1) % DstXprType::RowsAtCompileTime
+  };
+  
+  typedef typename Kernel::Scalar Scalar;
+
+  EIGEN_DEVICE_FUNC
+  static inline void run(Kernel &kernel)
+  {
+    triangular_assignment_loop<Kernel, Mode, UnrollCount-1, SetOpposite>::run(kernel);
+    
+    if(row==col)
+      kernel.assignDiagonalCoeff(row);
+    else if( ((Mode&Lower) && row>col) || ((Mode&Upper) && row<col) )
+      kernel.assignCoeff(row,col);
+    else if(SetOpposite)
+      kernel.assignOppositeCoeff(row,col);
+  }
+};
+
+// prevent buggy user code from causing an infinite recursion
+template<typename Kernel, unsigned int Mode, bool SetOpposite>
+struct triangular_assignment_loop<Kernel, Mode, 0, SetOpposite>
+{
+  EIGEN_DEVICE_FUNC
+  static inline void run(Kernel &) {}
+};
+
+
+
+// TODO: experiment with a recursive assignment procedure splitting the current
+//       triangular part into one rectangular and two triangular parts.
+
+
+template<typename Kernel, unsigned int Mode, bool SetOpposite>
+struct triangular_assignment_loop<Kernel, Mode, Dynamic, SetOpposite>
+{
+  typedef typename Kernel::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
+  static inline void run(Kernel &kernel)
+  {
+    for(Index j = 0; j < kernel.cols(); ++j)
+    {
+      Index maxi = numext::mini(j, kernel.rows());
+      Index i = 0;
+      if (((Mode&Lower) && SetOpposite) || (Mode&Upper))
+      {
+        for(; i < maxi; ++i)
+          if(Mode&Upper) kernel.assignCoeff(i, j);
+          else           kernel.assignOppositeCoeff(i, j);
+      }
+      else
+        i = maxi;
+      
+      if(i<kernel.rows()) // then i==j
+        kernel.assignDiagonalCoeff(i++);
+      
+      if (((Mode&Upper) && SetOpposite) || (Mode&Lower))
+      {
+        for(; i < kernel.rows(); ++i)
+          if(Mode&Lower) kernel.assignCoeff(i, j);
+          else           kernel.assignOppositeCoeff(i, j);
+      }
+    }
+  }
+};
+
+} // end namespace internal
+
+/** Assigns a triangular or selfadjoint matrix to a dense matrix.
+  * If the matrix is triangular, the opposite part is set to zero. */
+template<typename Derived>
+template<typename DenseDerived>
+void TriangularBase<Derived>::evalToLazy(MatrixBase<DenseDerived> &other) const
+{
+  other.derived().resize(this->rows(), this->cols());
+  internal::call_triangular_assignment_loop<Derived::Mode,(Derived::Mode&SelfAdjoint)==0 /* SetOpposite */>(other.derived(), derived().nestedExpression());
+}
+
+namespace internal {
+  
+// Triangular = Product
+template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
+{
+  typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename SrcXprType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst._assignProduct(src, 1, 0);
+  }
+};
+
+// Triangular += Product
+template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::add_assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
+{
+  typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<Scalar,typename SrcXprType::Scalar> &)
+  {
+    dst._assignProduct(src, 1, 1);
+  }
+};
+
+// Triangular -= Product
+template< typename DstXprType, typename Lhs, typename Rhs, typename Scalar>
+struct Assignment<DstXprType, Product<Lhs,Rhs,DefaultProduct>, internal::sub_assign_op<Scalar,typename Product<Lhs,Rhs,DefaultProduct>::Scalar>, Dense2Triangular>
+{
+  typedef Product<Lhs,Rhs,DefaultProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<Scalar,typename SrcXprType::Scalar> &)
+  {
+    dst._assignProduct(src, -1, 1);
+  }
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_TRIANGULARMATRIX_H
diff --git a/Eigen/src/Core/VectorBlock.h b/Eigen/src/Core/VectorBlock.h
index 1a7330f..d72fbf7 100644
--- a/Eigen/src/Core/VectorBlock.h
+++ b/Eigen/src/Core/VectorBlock.h
@@ -13,13 +13,23 @@
 
 namespace Eigen { 
 
+namespace internal {
+template<typename VectorType, int Size>
+struct traits<VectorBlock<VectorType, Size> >
+  : public traits<Block<VectorType,
+                     traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
+                     traits<VectorType>::Flags & RowMajorBit ? Size : 1> >
+{
+};
+}
+
 /** \class VectorBlock
   * \ingroup Core_Module
   *
   * \brief Expression of a fixed-size or dynamic-size sub-vector
   *
-  * \param VectorType the type of the object in which we are taking a sub-vector
-  * \param Size size of the sub-vector we are taking at compile time (optional)
+  * \tparam VectorType the type of the object in which we are taking a sub-vector
+  * \tparam Size size of the sub-vector we are taking at compile time (optional)
   *
   * This class represents an expression of either a fixed-size or dynamic-size sub-vector.
   * It is the return type of DenseBase::segment(Index,Index) and DenseBase::segment<int>(Index) and
@@ -43,17 +53,6 @@
   *
   * \sa class Block, DenseBase::segment(Index,Index,Index,Index), DenseBase::segment(Index,Index)
   */
-
-namespace internal {
-template<typename VectorType, int Size>
-struct traits<VectorBlock<VectorType, Size> >
-  : public traits<Block<VectorType,
-                     traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
-                     traits<VectorType>::Flags & RowMajorBit ? Size : 1> >
-{
-};
-}
-
 template<typename VectorType, int Size> class VectorBlock
   : public Block<VectorType,
                      internal::traits<VectorType>::Flags & RowMajorBit ? 1 : Size,
@@ -72,6 +71,7 @@
 
     /** Dynamic-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline VectorBlock(VectorType& vector, Index start, Index size)
       : Base(vector,
              IsColVector ? start : 0, IsColVector ? 0 : start,
@@ -82,6 +82,7 @@
 
     /** Fixed-size constructor
       */
+    EIGEN_DEVICE_FUNC
     inline VectorBlock(VectorType& vector, Index start)
       : Base(vector, IsColVector ? start : 0, IsColVector ? 0 : start)
     {
diff --git a/Eigen/src/Core/VectorwiseOp.h b/Eigen/src/Core/VectorwiseOp.h
index d5ab036..4fe267e 100644
--- a/Eigen/src/Core/VectorwiseOp.h
+++ b/Eigen/src/Core/VectorwiseOp.h
@@ -11,7 +11,7 @@
 #ifndef EIGEN_PARTIAL_REDUX_H
 #define EIGEN_PARTIAL_REDUX_H
 
-namespace Eigen { 
+namespace Eigen {
 
 /** \class PartialReduxExpr
   * \ingroup Core_Module
@@ -41,64 +41,43 @@
   typedef typename traits<MatrixType>::StorageKind StorageKind;
   typedef typename traits<MatrixType>::XprKind XprKind;
   typedef typename MatrixType::Scalar InputScalar;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
-  typedef typename remove_all<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsAtCompileTime = Direction==Vertical   ? 1 : MatrixType::RowsAtCompileTime,
     ColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = Direction==Vertical   ? 1 : MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = Direction==Horizontal ? 1 : MatrixType::MaxColsAtCompileTime,
-    Flags0 = (unsigned int)_MatrixTypeNested::Flags & HereditaryBits,
-    Flags = (Flags0 & ~RowMajorBit) | (RowsAtCompileTime == 1 ? RowMajorBit : 0),
+    Flags = RowsAtCompileTime == 1 ? RowMajorBit : 0,
     TraversalSize = Direction==Vertical ? MatrixType::RowsAtCompileTime :  MatrixType::ColsAtCompileTime
   };
-  #if EIGEN_GNUC_AT_LEAST(3,4)
-  typedef typename MemberOp::template Cost<InputScalar,int(TraversalSize)> CostOpType;
-  #else
-  typedef typename MemberOp::template Cost<InputScalar,TraversalSize> CostOpType;
-  #endif
-  enum {
-    CoeffReadCost = TraversalSize==Dynamic ? Dynamic
-                  : TraversalSize * traits<_MatrixTypeNested>::CoeffReadCost + int(CostOpType::value)
-  };
 };
 }
 
 template< typename MatrixType, typename MemberOp, int Direction>
-class PartialReduxExpr : internal::no_assignment_operator,
-  public internal::dense_xpr_base< PartialReduxExpr<MatrixType, MemberOp, Direction> >::type
+class PartialReduxExpr : public internal::dense_xpr_base< PartialReduxExpr<MatrixType, MemberOp, Direction> >::type,
+                         internal::no_assignment_operator
 {
   public:
 
     typedef typename internal::dense_xpr_base<PartialReduxExpr>::type Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(PartialReduxExpr)
-    typedef typename internal::traits<PartialReduxExpr>::MatrixTypeNested MatrixTypeNested;
-    typedef typename internal::traits<PartialReduxExpr>::_MatrixTypeNested _MatrixTypeNested;
 
-    PartialReduxExpr(const MatrixType& mat, const MemberOp& func = MemberOp())
+    EIGEN_DEVICE_FUNC
+    explicit PartialReduxExpr(const MatrixType& mat, const MemberOp& func = MemberOp())
       : m_matrix(mat), m_functor(func) {}
 
+    EIGEN_DEVICE_FUNC
     Index rows() const { return (Direction==Vertical   ? 1 : m_matrix.rows()); }
+    EIGEN_DEVICE_FUNC
     Index cols() const { return (Direction==Horizontal ? 1 : m_matrix.cols()); }
 
-    EIGEN_STRONG_INLINE const Scalar coeff(Index i, Index j) const
-    {
-      if (Direction==Vertical)
-        return m_functor(m_matrix.col(j));
-      else
-        return m_functor(m_matrix.row(i));
-    }
+    EIGEN_DEVICE_FUNC
+    typename MatrixType::Nested nestedExpression() const { return m_matrix; }
 
-    const Scalar coeff(Index index) const
-    {
-      if (Direction==Vertical)
-        return m_functor(m_matrix.col(index));
-      else
-        return m_functor(m_matrix.row(index));
-    }
+    EIGEN_DEVICE_FUNC
+    const MemberOp& functor() const { return m_functor; }
 
   protected:
-    MatrixTypeNested m_matrix;
+    typename MatrixType::Nested m_matrix;
     const MemberOp m_functor;
 };
 
@@ -110,7 +89,8 @@
     template<typename Scalar, int Size> struct Cost                     \
     { enum { value = COST }; };                                         \
     template<typename XprType>                                          \
-    EIGEN_STRONG_INLINE ResultType operator()(const XprType& mat) const \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE                               \
+    ResultType operator()(const XprType& mat) const                     \
     { return mat.MEMBER(); } \
   }
 
@@ -130,17 +110,27 @@
 EIGEN_MEMBER_FUNCTOR(count, (Size-1)*NumTraits<Scalar>::AddCost);
 EIGEN_MEMBER_FUNCTOR(prod, (Size-1)*NumTraits<Scalar>::MulCost);
 
+template <int p, typename ResultType>
+struct member_lpnorm {
+  typedef ResultType result_type;
+  template<typename Scalar, int Size> struct Cost
+  { enum { value = (Size+5) * NumTraits<Scalar>::MulCost + (Size-1)*NumTraits<Scalar>::AddCost }; };
+  EIGEN_DEVICE_FUNC member_lpnorm() {}
+  template<typename XprType>
+  EIGEN_DEVICE_FUNC inline ResultType operator()(const XprType& mat) const
+  { return mat.template lpNorm<p>(); }
+};
 
 template <typename BinaryOp, typename Scalar>
 struct member_redux {
   typedef typename result_of<
-                     BinaryOp(Scalar)
+                     BinaryOp(const Scalar&,const Scalar&)
                    >::type  result_type;
   template<typename _Scalar, int Size> struct Cost
   { enum { value = (Size-1) * functor_traits<BinaryOp>::Cost }; };
-  member_redux(const BinaryOp func) : m_functor(func) {}
+  EIGEN_DEVICE_FUNC explicit member_redux(const BinaryOp func) : m_functor(func) {}
   template<typename Derived>
-  inline result_type operator()(const DenseBase<Derived>& mat) const
+  EIGEN_DEVICE_FUNC inline result_type operator()(const DenseBase<Derived>& mat) const
   { return mat.redux(m_functor); }
   const BinaryOp m_functor;
 };
@@ -151,8 +141,8 @@
   *
   * \brief Pseudo expression providing partial reduction operations
   *
-  * \param ExpressionType the type of the object on which to do partial reductions
-  * \param Direction indicates the direction of the redux (#Vertical or #Horizontal)
+  * \tparam ExpressionType the type of the object on which to do partial reductions
+  * \tparam Direction indicates the direction of the redux (#Vertical or #Horizontal)
   *
   * This class represents a pseudo expression with partial reduction features.
   * It is the return type of DenseBase::colwise() and DenseBase::rowwise()
@@ -169,16 +159,15 @@
 
     typedef typename ExpressionType::Scalar Scalar;
     typedef typename ExpressionType::RealScalar RealScalar;
-    typedef typename ExpressionType::Index Index;
-    typedef typename internal::conditional<internal::must_nest_by_value<ExpressionType>::ret,
-        ExpressionType, ExpressionType&>::type ExpressionTypeNested;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    typedef typename internal::ref_selector<ExpressionType>::non_const_type ExpressionTypeNested;
     typedef typename internal::remove_all<ExpressionTypeNested>::type ExpressionTypeNestedCleaned;
 
     template<template<typename _Scalar> class Functor,
-                      typename Scalar=typename internal::traits<ExpressionType>::Scalar> struct ReturnType
+                      typename Scalar_=Scalar> struct ReturnType
     {
       typedef PartialReduxExpr<ExpressionType,
-                               Functor<Scalar>,
+                               Functor<Scalar_>,
                                Direction
                               > Type;
     };
@@ -186,23 +175,24 @@
     template<typename BinaryOp> struct ReduxReturnType
     {
       typedef PartialReduxExpr<ExpressionType,
-                               internal::member_redux<BinaryOp,typename internal::traits<ExpressionType>::Scalar>,
+                               internal::member_redux<BinaryOp,Scalar>,
                                Direction
                               > Type;
     };
 
     enum {
-      IsVertical   = (Direction==Vertical) ? 1 : 0,
-      IsHorizontal = (Direction==Horizontal) ? 1 : 0
+      isVertical   = (Direction==Vertical) ? 1 : 0,
+      isHorizontal = (Direction==Horizontal) ? 1 : 0
     };
 
   protected:
 
-    /** \internal
-      * \returns the i-th subvector according to the \c Direction */
-    typedef typename internal::conditional<Direction==Vertical,
+    typedef typename internal::conditional<isVertical,
                                typename ExpressionType::ColXpr,
                                typename ExpressionType::RowXpr>::type SubVector;
+    /** \internal
+      * \returns the i-th subvector according to the \c Direction */
+    EIGEN_DEVICE_FUNC
     SubVector subVector(Index i)
     {
       return SubVector(m_matrix.derived(),i);
@@ -210,58 +200,62 @@
 
     /** \internal
       * \returns the number of subvectors in the direction \c Direction */
+    EIGEN_DEVICE_FUNC
     Index subVectors() const
-    { return Direction==Vertical?m_matrix.cols():m_matrix.rows(); }
+    { return isVertical?m_matrix.cols():m_matrix.rows(); }
 
     template<typename OtherDerived> struct ExtendedType {
       typedef Replicate<OtherDerived,
-                        Direction==Vertical   ? 1 : ExpressionType::RowsAtCompileTime,
-                        Direction==Horizontal ? 1 : ExpressionType::ColsAtCompileTime> Type;
+                        isVertical   ? 1 : ExpressionType::RowsAtCompileTime,
+                        isHorizontal ? 1 : ExpressionType::ColsAtCompileTime> Type;
     };
 
     /** \internal
       * Replicates a vector to match the size of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     typename ExtendedType<OtherDerived>::Type
     extendedTo(const DenseBase<OtherDerived>& other) const
     {
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Vertical, OtherDerived::MaxColsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isVertical, OtherDerived::MaxColsAtCompileTime==1),
                           YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED)
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Horizontal, OtherDerived::MaxRowsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isHorizontal, OtherDerived::MaxRowsAtCompileTime==1),
                           YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED)
       return typename ExtendedType<OtherDerived>::Type
                       (other.derived(),
-                       Direction==Vertical   ? 1 : m_matrix.rows(),
-                       Direction==Horizontal ? 1 : m_matrix.cols());
+                       isVertical   ? 1 : m_matrix.rows(),
+                       isHorizontal ? 1 : m_matrix.cols());
     }
-    
+
     template<typename OtherDerived> struct OppositeExtendedType {
       typedef Replicate<OtherDerived,
-                        Direction==Horizontal ? 1 : ExpressionType::RowsAtCompileTime,
-                        Direction==Vertical   ? 1 : ExpressionType::ColsAtCompileTime> Type;
+                        isHorizontal ? 1 : ExpressionType::RowsAtCompileTime,
+                        isVertical   ? 1 : ExpressionType::ColsAtCompileTime> Type;
     };
 
     /** \internal
       * Replicates a vector in the opposite direction to match the size of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     typename OppositeExtendedType<OtherDerived>::Type
     extendedToOpposite(const DenseBase<OtherDerived>& other) const
     {
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Horizontal, OtherDerived::MaxColsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isHorizontal, OtherDerived::MaxColsAtCompileTime==1),
                           YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED)
-      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(Direction==Vertical, OtherDerived::MaxRowsAtCompileTime==1),
+      EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(isVertical, OtherDerived::MaxRowsAtCompileTime==1),
                           YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED)
       return typename OppositeExtendedType<OtherDerived>::Type
                       (other.derived(),
-                       Direction==Horizontal  ? 1 : m_matrix.rows(),
-                       Direction==Vertical    ? 1 : m_matrix.cols());
+                       isHorizontal  ? 1 : m_matrix.rows(),
+                       isVertical    ? 1 : m_matrix.cols());
     }
 
   public:
-
-    inline VectorwiseOp(ExpressionType& matrix) : m_matrix(matrix) {}
+    EIGEN_DEVICE_FUNC
+    explicit inline VectorwiseOp(ExpressionType& matrix) : m_matrix(matrix) {}
 
     /** \internal */
+    EIGEN_DEVICE_FUNC
     inline const ExpressionType& _expression() const { return m_matrix; }
 
     /** \returns a row or column vector expression of \c *this reduxed by \a func
@@ -272,80 +266,126 @@
       * \sa class VectorwiseOp, DenseBase::colwise(), DenseBase::rowwise()
       */
     template<typename BinaryOp>
+    EIGEN_DEVICE_FUNC
     const typename ReduxReturnType<BinaryOp>::Type
     redux(const BinaryOp& func = BinaryOp()) const
-    { return typename ReduxReturnType<BinaryOp>::Type(_expression(), func); }
+    { return typename ReduxReturnType<BinaryOp>::Type(_expression(), internal::member_redux<BinaryOp,Scalar>(func)); }
+
+    typedef typename ReturnType<internal::member_minCoeff>::Type MinCoeffReturnType;
+    typedef typename ReturnType<internal::member_maxCoeff>::Type MaxCoeffReturnType;
+    typedef typename ReturnType<internal::member_squaredNorm,RealScalar>::Type SquaredNormReturnType;
+    typedef typename ReturnType<internal::member_norm,RealScalar>::Type NormReturnType;
+    typedef typename ReturnType<internal::member_blueNorm,RealScalar>::Type BlueNormReturnType;
+    typedef typename ReturnType<internal::member_stableNorm,RealScalar>::Type StableNormReturnType;
+    typedef typename ReturnType<internal::member_hypotNorm,RealScalar>::Type HypotNormReturnType;
+    typedef typename ReturnType<internal::member_sum>::Type SumReturnType;
+    typedef typename ReturnType<internal::member_mean>::Type MeanReturnType;
+    typedef typename ReturnType<internal::member_all>::Type AllReturnType;
+    typedef typename ReturnType<internal::member_any>::Type AnyReturnType;
+    typedef PartialReduxExpr<ExpressionType, internal::member_count<Index>, Direction> CountReturnType;
+    typedef typename ReturnType<internal::member_prod>::Type ProdReturnType;
+    typedef Reverse<const ExpressionType, Direction> ConstReverseReturnType;
+    typedef Reverse<ExpressionType, Direction> ReverseReturnType;
+
+    template<int p> struct LpNormReturnType {
+      typedef PartialReduxExpr<ExpressionType, internal::member_lpnorm<p,RealScalar>,Direction> Type;
+    };
 
     /** \returns a row (or column) vector expression of the smallest coefficient
       * of each column (or row) of the referenced expression.
-      * 
+      *
       * \warning the result is undefined if \c *this contains NaN.
       *
       * Example: \include PartialRedux_minCoeff.cpp
       * Output: \verbinclude PartialRedux_minCoeff.out
       *
       * \sa DenseBase::minCoeff() */
-    const typename ReturnType<internal::member_minCoeff>::Type minCoeff() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const MinCoeffReturnType minCoeff() const
+    { return MinCoeffReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the largest coefficient
       * of each column (or row) of the referenced expression.
-      * 
+      *
       * \warning the result is undefined if \c *this contains NaN.
       *
       * Example: \include PartialRedux_maxCoeff.cpp
       * Output: \verbinclude PartialRedux_maxCoeff.out
       *
       * \sa DenseBase::maxCoeff() */
-    const typename ReturnType<internal::member_maxCoeff>::Type maxCoeff() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const MaxCoeffReturnType maxCoeff() const
+    { return MaxCoeffReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the squared norm
       * of each column (or row) of the referenced expression.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * Example: \include PartialRedux_squaredNorm.cpp
       * Output: \verbinclude PartialRedux_squaredNorm.out
       *
       * \sa DenseBase::squaredNorm() */
-    const typename ReturnType<internal::member_squaredNorm,RealScalar>::Type squaredNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const SquaredNormReturnType squaredNorm() const
+    { return SquaredNormReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * Example: \include PartialRedux_norm.cpp
       * Output: \verbinclude PartialRedux_norm.out
       *
       * \sa DenseBase::norm() */
-    const typename ReturnType<internal::member_norm,RealScalar>::Type norm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const NormReturnType norm() const
+    { return NormReturnType(_expression()); }
+
+    /** \returns a row (or column) vector expression of the norm
+      * of each column (or row) of the referenced expression.
+      * This is a vector with real entries, even if the original matrix has complex entries.
+      *
+      * Example: \include PartialRedux_norm.cpp
+      * Output: \verbinclude PartialRedux_norm.out
+      *
+      * \sa DenseBase::norm() */
+    template<int p>
+    EIGEN_DEVICE_FUNC
+    const typename LpNormReturnType<p>::Type lpNorm() const
+    { return typename LpNormReturnType<p>::Type(_expression()); }
 
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression, using
-      * blue's algorithm.
+      * Blue's algorithm.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * \sa DenseBase::blueNorm() */
-    const typename ReturnType<internal::member_blueNorm,RealScalar>::Type blueNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const BlueNormReturnType blueNorm() const
+    { return BlueNormReturnType(_expression()); }
 
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression, avoiding
       * underflow and overflow.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * \sa DenseBase::stableNorm() */
-    const typename ReturnType<internal::member_stableNorm,RealScalar>::Type stableNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const StableNormReturnType stableNorm() const
+    { return StableNormReturnType(_expression()); }
 
 
     /** \returns a row (or column) vector expression of the norm
       * of each column (or row) of the referenced expression, avoiding
       * underflow and overflow using a concatenation of hypot() calls.
+      * This is a vector with real entries, even if the original matrix has complex entries.
       *
       * \sa DenseBase::hypotNorm() */
-    const typename ReturnType<internal::member_hypotNorm,RealScalar>::Type hypotNorm() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const HypotNormReturnType hypotNorm() const
+    { return HypotNormReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the sum
       * of each column (or row) of the referenced expression.
@@ -354,39 +394,48 @@
       * Output: \verbinclude PartialRedux_sum.out
       *
       * \sa DenseBase::sum() */
-    const typename ReturnType<internal::member_sum>::Type sum() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const SumReturnType sum() const
+    { return SumReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the mean
     * of each column (or row) of the referenced expression.
     *
     * \sa DenseBase::mean() */
-    const typename ReturnType<internal::member_mean>::Type mean() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const MeanReturnType mean() const
+    { return MeanReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression representing
       * whether \b all coefficients of each respective column (or row) are \c true.
+      * This expression can be assigned to a vector with entries of type \c bool.
       *
       * \sa DenseBase::all() */
-    const typename ReturnType<internal::member_all>::Type all() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const AllReturnType all() const
+    { return AllReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression representing
       * whether \b at \b least one coefficient of each respective column (or row) is \c true.
+      * This expression can be assigned to a vector with entries of type \c bool.
       *
       * \sa DenseBase::any() */
-    const typename ReturnType<internal::member_any>::Type any() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const AnyReturnType any() const
+    { return AnyReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression representing
       * the number of \c true coefficients of each respective column (or row).
+      * This expression can be assigned to a vector whose entries have the same type as is used to
+      * index entries of the original matrix; for dense matrices, this is \c std::ptrdiff_t .
       *
       * Example: \include PartialRedux_count.cpp
       * Output: \verbinclude PartialRedux_count.out
       *
       * \sa DenseBase::count() */
-    const PartialReduxExpr<ExpressionType, internal::member_count<Index>, Direction> count() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const CountReturnType count() const
+    { return CountReturnType(_expression()); }
 
     /** \returns a row (or column) vector expression of the product
       * of each column (or row) of the referenced expression.
@@ -395,8 +444,9 @@
       * Output: \verbinclude PartialRedux_prod.out
       *
       * \sa DenseBase::prod() */
-    const typename ReturnType<internal::member_prod>::Type prod() const
-    { return _expression(); }
+    EIGEN_DEVICE_FUNC
+    const ProdReturnType prod() const
+    { return ProdReturnType(_expression()); }
 
 
     /** \returns a matrix expression
@@ -406,10 +456,20 @@
       * Output: \verbinclude Vectorwise_reverse.out
       *
       * \sa DenseBase::reverse() */
-    const Reverse<ExpressionType, Direction> reverse() const
-    { return Reverse<ExpressionType, Direction>( _expression() ); }
+    EIGEN_DEVICE_FUNC
+    const ConstReverseReturnType reverse() const
+    { return ConstReverseReturnType( _expression() ); }
 
-    typedef Replicate<ExpressionType,Direction==Vertical?Dynamic:1,Direction==Horizontal?Dynamic:1> ReplicateReturnType;
+    /** \returns a writable matrix expression
+      * where each column (or row) are reversed.
+      *
+      * \sa reverse() const */
+    EIGEN_DEVICE_FUNC
+    ReverseReturnType reverse()
+    { return ReverseReturnType( _expression() ); }
+
+    typedef Replicate<ExpressionType,(isVertical?Dynamic:1),(isHorizontal?Dynamic:1)> ReplicateReturnType;
+    EIGEN_DEVICE_FUNC
     const ReplicateReturnType replicate(Index factor) const;
 
     /**
@@ -421,17 +481,20 @@
       * \sa VectorwiseOp::replicate(Index), DenseBase::replicate(), class Replicate
       */
     // NOTE implemented here because of sunstudio's compilation errors
-    template<int Factor> const Replicate<ExpressionType,(IsVertical?Factor:1),(IsHorizontal?Factor:1)>
+    // isVertical*Factor+isHorizontal instead of (isVertical?Factor:1) to handle CUDA bug with ternary operator
+    template<int Factor> const Replicate<ExpressionType,isVertical*Factor+isHorizontal,isHorizontal*Factor+isVertical>
+    EIGEN_DEVICE_FUNC
     replicate(Index factor = Factor) const
     {
-      return Replicate<ExpressionType,Direction==Vertical?Factor:1,Direction==Horizontal?Factor:1>
-          (_expression(),Direction==Vertical?factor:1,Direction==Horizontal?factor:1);
+      return Replicate<ExpressionType,(isVertical?Factor:1),(isHorizontal?Factor:1)>
+          (_expression(),isVertical?factor:1,isHorizontal?factor:1);
     }
 
 /////////// Artithmetic operators ///////////
 
     /** Copies the vector \a other to each subvector of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -442,6 +505,7 @@
 
     /** Adds the vector \a other to each subvector of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator+=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -451,6 +515,7 @@
 
     /** Substracts the vector \a other to each subvector of \c *this */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator-=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -460,6 +525,7 @@
 
     /** Multiples each subvector of \c *this by the vector \a other */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator*=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -471,6 +537,7 @@
 
     /** Divides each subvector of \c *this by the vector \a other */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     ExpressionType& operator/=(const DenseBase<OtherDerived>& other)
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -481,8 +548,8 @@
     }
 
     /** Returns the expression of the sum of the vector \a other to each subvector of \c *this */
-    template<typename OtherDerived> EIGEN_STRONG_INLINE
-    CwiseBinaryOp<internal::scalar_sum_op<Scalar>,
+    template<typename OtherDerived> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
+    CwiseBinaryOp<internal::scalar_sum_op<Scalar,typename OtherDerived::Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
     operator+(const DenseBase<OtherDerived>& other) const
@@ -494,7 +561,8 @@
 
     /** Returns the expression of the difference between each subvector of \c *this and the vector \a other */
     template<typename OtherDerived>
-    CwiseBinaryOp<internal::scalar_difference_op<Scalar>,
+    EIGEN_DEVICE_FUNC
+    CwiseBinaryOp<internal::scalar_difference_op<Scalar,typename OtherDerived::Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
     operator-(const DenseBase<OtherDerived>& other) const
@@ -506,10 +574,11 @@
 
     /** Returns the expression where each subvector is the product of the vector \a other
       * by the corresponding subvector of \c *this */
-    template<typename OtherDerived> EIGEN_STRONG_INLINE
+    template<typename OtherDerived> EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC
     CwiseBinaryOp<internal::scalar_product_op<Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
+    EIGEN_DEVICE_FUNC
     operator*(const DenseBase<OtherDerived>& other) const
     {
       EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived)
@@ -521,6 +590,7 @@
     /** Returns the expression where each subvector is the quotient of the corresponding
       * subvector of \c *this by the vector \a other */
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename ExtendedType<OtherDerived>::Type>
@@ -531,32 +601,36 @@
       EIGEN_STATIC_ASSERT_SAME_XPR_KIND(ExpressionType, OtherDerived)
       return m_matrix / extendedTo(other.derived());
     }
-    
-    /** \returns an expression where each column of row of the referenced matrix are normalized.
+
+    /** \returns an expression where each column (or row) of the referenced matrix are normalized.
       * The referenced matrix is \b not modified.
       * \sa MatrixBase::normalized(), normalize()
       */
+    EIGEN_DEVICE_FUNC
     CwiseBinaryOp<internal::scalar_quotient_op<Scalar>,
                   const ExpressionTypeNestedCleaned,
                   const typename OppositeExtendedType<typename ReturnType<internal::member_norm,RealScalar>::Type>::Type>
     normalized() const { return m_matrix.cwiseQuotient(extendedToOpposite(this->norm())); }
-    
-    
+
+
     /** Normalize in-place each row or columns of the referenced matrix.
       * \sa MatrixBase::normalize(), normalized()
       */
-    void normalize() {
+    EIGEN_DEVICE_FUNC void normalize() {
       m_matrix = this->normalized();
     }
 
+    EIGEN_DEVICE_FUNC inline void reverseInPlace();
+
 /////////// Geometry module ///////////
 
-    #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
-    Homogeneous<ExpressionType,Direction> homogeneous() const;
-    #endif
+    typedef Homogeneous<ExpressionType,Direction> HomogeneousReturnType;
+    EIGEN_DEVICE_FUNC
+    HomogeneousReturnType homogeneous() const;
 
     typedef typename ExpressionType::PlainObject CrossReturnType;
     template<typename OtherDerived>
+    EIGEN_DEVICE_FUNC
     const CrossReturnType cross(const MatrixBase<OtherDerived>& other) const;
 
     enum {
@@ -581,25 +655,15 @@
                   Direction==Horizontal ? HNormalized_SizeMinusOne : 1> >
             HNormalizedReturnType;
 
+    EIGEN_DEVICE_FUNC
     const HNormalizedReturnType hnormalized() const;
 
   protected:
     ExpressionTypeNested m_matrix;
 };
 
-/** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
-  *
-  * Example: \include MatrixBase_colwise.cpp
-  * Output: \verbinclude MatrixBase_colwise.out
-  *
-  * \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
-  */
-template<typename Derived>
-inline const typename DenseBase<Derived>::ConstColwiseReturnType
-DenseBase<Derived>::colwise() const
-{
-  return derived();
-}
+//const colwise moved to DenseBase.h due to CUDA compiler bug
+
 
 /** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
   *
@@ -609,22 +673,11 @@
 inline typename DenseBase<Derived>::ColwiseReturnType
 DenseBase<Derived>::colwise()
 {
-  return derived();
+  return ColwiseReturnType(derived());
 }
 
-/** \returns a VectorwiseOp wrapper of *this providing additional partial reduction operations
-  *
-  * Example: \include MatrixBase_rowwise.cpp
-  * Output: \verbinclude MatrixBase_rowwise.out
-  *
-  * \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting
-  */
-template<typename Derived>
-inline const typename DenseBase<Derived>::ConstRowwiseReturnType
-DenseBase<Derived>::rowwise() const
-{
-  return derived();
-}
+//const rowwise moved to DenseBase.h due to CUDA compiler bug
+
 
 /** \returns a writable VectorwiseOp wrapper of *this providing additional partial reduction operations
   *
@@ -634,7 +687,7 @@
 inline typename DenseBase<Derived>::RowwiseReturnType
 DenseBase<Derived>::rowwise()
 {
-  return derived();
+  return RowwiseReturnType(derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/Visitor.h b/Eigen/src/Core/Visitor.h
index 64867b7..54c1883 100644
--- a/Eigen/src/Core/Visitor.h
+++ b/Eigen/src/Core/Visitor.h
@@ -22,6 +22,7 @@
     row = (UnrollCount-1) % Derived::RowsAtCompileTime
   };
 
+  EIGEN_DEVICE_FUNC
   static inline void run(const Derived &mat, Visitor& visitor)
   {
     visitor_impl<Visitor, Derived, UnrollCount-1>::run(mat, visitor);
@@ -32,6 +33,7 @@
 template<typename Visitor, typename Derived>
 struct visitor_impl<Visitor, Derived, 1>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(const Derived &mat, Visitor& visitor)
   {
     return visitor.init(mat.coeff(0, 0), 0, 0);
@@ -41,7 +43,7 @@
 template<typename Visitor, typename Derived>
 struct visitor_impl<Visitor, Derived, Dynamic>
 {
-  typedef typename Derived::Index Index;
+  EIGEN_DEVICE_FUNC
   static inline void run(const Derived& mat, Visitor& visitor)
   {
     visitor.init(mat.coeff(0,0), 0, 0);
@@ -53,6 +55,33 @@
   }
 };
 
+// evaluator adaptor
+template<typename XprType>
+class visitor_evaluator
+{
+public:
+  EIGEN_DEVICE_FUNC
+  explicit visitor_evaluator(const XprType &xpr) : m_evaluator(xpr), m_xpr(xpr) {}
+  
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::CoeffReturnType CoeffReturnType;
+  
+  enum {
+    RowsAtCompileTime = XprType::RowsAtCompileTime,
+    CoeffReadCost = internal::evaluator<XprType>::CoeffReadCost
+  };
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_xpr.rows(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_xpr.cols(); }
+  EIGEN_DEVICE_FUNC Index size() const { return m_xpr.size(); }
+
+  EIGEN_DEVICE_FUNC CoeffReturnType coeff(Index row, Index col) const
+  { return m_evaluator.coeff(row, col); }
+  
+protected:
+  internal::evaluator<XprType> m_evaluator;
+  const XprType &m_xpr;
+};
 } // end namespace internal
 
 /** Applies the visitor \a visitor to the whole coefficients of the matrix or vector.
@@ -74,16 +103,17 @@
   */
 template<typename Derived>
 template<typename Visitor>
+EIGEN_DEVICE_FUNC
 void DenseBase<Derived>::visit(Visitor& visitor) const
 {
-  enum { unroll = SizeAtCompileTime != Dynamic
-                   && CoeffReadCost != Dynamic
-                   && (SizeAtCompileTime == 1 || internal::functor_traits<Visitor>::Cost != Dynamic)
-                   && SizeAtCompileTime * CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost
-                      <= EIGEN_UNROLLING_LIMIT };
-  return internal::visitor_impl<Visitor, Derived,
-      unroll ? int(SizeAtCompileTime) : Dynamic
-    >::run(derived(), visitor);
+  typedef typename internal::visitor_evaluator<Derived> ThisEvaluator;
+  ThisEvaluator thisEval(derived());
+  
+  enum {
+    unroll =  SizeAtCompileTime != Dynamic
+           && SizeAtCompileTime * ThisEvaluator::CoeffReadCost + (SizeAtCompileTime-1) * internal::functor_traits<Visitor>::Cost <= EIGEN_UNROLLING_LIMIT
+  };
+  return internal::visitor_impl<Visitor, ThisEvaluator, unroll ? int(SizeAtCompileTime) : Dynamic>::run(thisEval, visitor);
 }
 
 namespace internal {
@@ -94,10 +124,10 @@
 template <typename Derived>
 struct coeff_visitor
 {
-  typedef typename Derived::Index Index;
   typedef typename Derived::Scalar Scalar;
   Index row, col;
   Scalar res;
+  EIGEN_DEVICE_FUNC
   inline void init(const Scalar& value, Index i, Index j)
   {
     res = value;
@@ -114,8 +144,8 @@
 template <typename Derived>
 struct min_coeff_visitor : coeff_visitor<Derived>
 {
-  typedef typename Derived::Index Index;
   typedef typename Derived::Scalar Scalar;
+  EIGEN_DEVICE_FUNC
   void operator() (const Scalar& value, Index i, Index j)
   {
     if(value < this->res)
@@ -142,8 +172,8 @@
 template <typename Derived>
 struct max_coeff_visitor : coeff_visitor<Derived>
 {
-  typedef typename Derived::Index Index;
-  typedef typename Derived::Scalar Scalar;
+  typedef typename Derived::Scalar Scalar; 
+  EIGEN_DEVICE_FUNC
   void operator() (const Scalar& value, Index i, Index j)
   {
     if(value > this->res)
@@ -164,13 +194,15 @@
 
 } // end namespace internal
 
-/** \returns the minimum of all coefficients of *this and puts in *row and *col its location.
+/** \fn DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
+  * \returns the minimum of all coefficients of *this and puts in *row and *col its location.
   * \warning the result is undefined if \c *this contains NaN.
   *
-  * \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visitor(), DenseBase::minCoeff()
+  * \sa DenseBase::minCoeff(Index*), DenseBase::maxCoeff(Index*,Index*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* rowId, IndexType* colId) const
 {
@@ -184,27 +216,30 @@
 /** \returns the minimum of all coefficients of *this and puts in *index its location.
   * \warning the result is undefined if \c *this contains NaN. 
   *
-  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::visitor(), DenseBase::minCoeff()
+  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::maxCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::minCoeff()
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::minCoeff(IndexType* index) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   internal::min_coeff_visitor<Derived> minVisitor;
   this->visit(minVisitor);
-  *index = (RowsAtCompileTime==1) ? minVisitor.col : minVisitor.row;
+  *index = IndexType((RowsAtCompileTime==1) ? minVisitor.col : minVisitor.row);
   return minVisitor.res;
 }
 
-/** \returns the maximum of all coefficients of *this and puts in *row and *col its location.
+/** \fn DenseBase<Derived>::maxCoeff(IndexType* rowId, IndexType* colId) const
+  * \returns the maximum of all coefficients of *this and puts in *row and *col its location.
   * \warning the result is undefined if \c *this contains NaN. 
   *
-  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visitor(), DenseBase::maxCoeff()
+  * \sa DenseBase::minCoeff(IndexType*,IndexType*), DenseBase::visit(), DenseBase::maxCoeff()
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* rowPtr, IndexType* colPtr) const
 {
@@ -222,6 +257,7 @@
   */
 template<typename Derived>
 template<typename IndexType>
+EIGEN_DEVICE_FUNC
 typename internal::traits<Derived>::Scalar
 DenseBase<Derived>::maxCoeff(IndexType* index) const
 {
diff --git a/Eigen/src/Core/arch/AVX/Complex.h b/Eigen/src/Core/arch/AVX/Complex.h
new file mode 100644
index 0000000..7fa6196
--- /dev/null
+++ b/Eigen/src/Core/arch/AVX/Complex.h
@@ -0,0 +1,451 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner (benoit.steiner.goog@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLEX_AVX_H
+#define EIGEN_COMPLEX_AVX_H
+
+namespace Eigen {
+
+namespace internal {
+
+//---------- float ----------
+struct Packet4cf
+{
+  EIGEN_STRONG_INLINE Packet4cf() {}
+  EIGEN_STRONG_INLINE explicit Packet4cf(const __m256& a) : v(a) {}
+  __m256  v;
+};
+
+template<> struct packet_traits<std::complex<float> >  : default_packet_traits
+{
+  typedef Packet4cf type;
+  typedef Packet2cf half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 4,
+    HasHalfPacket = 1,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet4cf> { typedef std::complex<float> type; enum {size=4, alignment=Aligned32}; typedef Packet2cf half; };
+
+template<> EIGEN_STRONG_INLINE Packet4cf padd<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_add_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf psub<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_sub_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pnegate(const Packet4cf& a)
+{
+  return Packet4cf(pnegate(a.v));
+}
+template<> EIGEN_STRONG_INLINE Packet4cf pconj(const Packet4cf& a)
+{
+  const __m256 mask = _mm256_castsi256_ps(_mm256_setr_epi32(0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000,0x00000000,0x80000000));
+  return Packet4cf(_mm256_xor_ps(a.v,mask));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pmul<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
+{
+  __m256 tmp1 = _mm256_mul_ps(_mm256_moveldup_ps(a.v), b.v);
+  __m256 tmp2 = _mm256_mul_ps(_mm256_movehdup_ps(a.v), _mm256_permute_ps(b.v, _MM_SHUFFLE(2,3,0,1)));
+  __m256 result = _mm256_addsub_ps(tmp1, tmp2);
+  return Packet4cf(result);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pand   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_and_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf por    <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_or_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pxor   <Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_xor_ps(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet4cf pandnot<Packet4cf>(const Packet4cf& a, const Packet4cf& b) { return Packet4cf(_mm256_andnot_ps(a.v,b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet4cf pload <Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet4cf(pload<Packet8f>(&numext::real_ref(*from))); }
+template<> EIGEN_STRONG_INLINE Packet4cf ploadu<Packet4cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet4cf(ploadu<Packet8f>(&numext::real_ref(*from))); }
+
+
+template<> EIGEN_STRONG_INLINE Packet4cf pset1<Packet4cf>(const std::complex<float>& from)
+{
+  return Packet4cf(_mm256_castpd_ps(_mm256_broadcast_sd((const double*)(const void*)&from)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf ploaddup<Packet4cf>(const std::complex<float>* from)
+{
+  // FIXME The following might be optimized using _mm256_movedup_pd
+  Packet2cf a = ploaddup<Packet2cf>(from);
+  Packet2cf b = ploaddup<Packet2cf>(from+1);
+  return  Packet4cf(_mm256_insertf128_ps(_mm256_castps128_ps256(a.v), b.v, 1));
+}
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float>* to, const Packet4cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet4cf pgather<std::complex<float>, Packet4cf>(const std::complex<float>* from, Index stride)
+{
+  return Packet4cf(_mm256_set_ps(std::imag(from[3*stride]), std::real(from[3*stride]),
+                                 std::imag(from[2*stride]), std::real(from[2*stride]),
+                                 std::imag(from[1*stride]), std::real(from[1*stride]),
+                                 std::imag(from[0*stride]), std::real(from[0*stride])));
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet4cf>(std::complex<float>* to, const Packet4cf& from, Index stride)
+{
+  __m128 low = _mm256_extractf128_ps(from.v, 0);
+  to[stride*0] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 0)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(low, low, 1)));
+  to[stride*1] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(low, low, 2)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(low, low, 3)));
+
+  __m128 high = _mm256_extractf128_ps(from.v, 1);
+  to[stride*2] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 0)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(high, high, 1)));
+  to[stride*3] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(high, high, 2)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(high, high, 3)));
+
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet4cf>(const Packet4cf& a)
+{
+  return pfirst(Packet2cf(_mm256_castps256_ps128(a.v)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf preverse(const Packet4cf& a) {
+  __m128 low  = _mm256_extractf128_ps(a.v, 0);
+  __m128 high = _mm256_extractf128_ps(a.v, 1);
+  __m128d lowd  = _mm_castps_pd(low);
+  __m128d highd = _mm_castps_pd(high);
+  low  = _mm_castpd_ps(_mm_shuffle_pd(lowd,lowd,0x1));
+  high = _mm_castpd_ps(_mm_shuffle_pd(highd,highd,0x1));
+  __m256 result = _mm256_setzero_ps();
+  result = _mm256_insertf128_ps(result, low, 1);
+  result = _mm256_insertf128_ps(result, high, 0);
+  return Packet4cf(result);
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet4cf>(const Packet4cf& a)
+{
+  return predux(padd(Packet2cf(_mm256_extractf128_ps(a.v,0)),
+                     Packet2cf(_mm256_extractf128_ps(a.v,1))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf preduxp<Packet4cf>(const Packet4cf* vecs)
+{
+  Packet8f t0 = _mm256_shuffle_ps(vecs[0].v, vecs[0].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  Packet8f t1 = _mm256_shuffle_ps(vecs[1].v, vecs[1].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  t0 = _mm256_hadd_ps(t0,t1);
+  Packet8f t2 = _mm256_shuffle_ps(vecs[2].v, vecs[2].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  Packet8f t3 = _mm256_shuffle_ps(vecs[3].v, vecs[3].v, _MM_SHUFFLE(3, 1, 2 ,0));
+  t2 = _mm256_hadd_ps(t2,t3);
+  
+  t1 = _mm256_permute2f128_ps(t0,t2, 0 + (2<<4));
+  t3 = _mm256_permute2f128_ps(t0,t2, 1 + (3<<4));
+
+  return Packet4cf(_mm256_add_ps(t1,t3));
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet4cf>(const Packet4cf& a)
+{
+  return predux_mul(pmul(Packet2cf(_mm256_extractf128_ps(a.v, 0)),
+                         Packet2cf(_mm256_extractf128_ps(a.v, 1))));
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet4cf>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4cf& first, const Packet4cf& second)
+  {
+    if (Offset==0) return;
+    palign_impl<Offset*2,Packet8f>::run(first.v, second.v);
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, false,true>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, true,false>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet4cf, Packet4cf, true,true>
+{
+  EIGEN_STRONG_INLINE Packet4cf pmadd(const Packet4cf& x, const Packet4cf& y, const Packet4cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet4cf pmul(const Packet4cf& a, const Packet4cf& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet4cf,Packet8f)
+
+template<> EIGEN_STRONG_INLINE Packet4cf pdiv<Packet4cf>(const Packet4cf& a, const Packet4cf& b)
+{
+  Packet4cf num = pmul(a, pconj(b));
+  __m256 tmp = _mm256_mul_ps(b.v, b.v);
+  __m256 tmp2    = _mm256_shuffle_ps(tmp,tmp,0xB1);
+  __m256 denom = _mm256_add_ps(tmp, tmp2);
+  return Packet4cf(_mm256_div_ps(num.v, denom));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pcplxflip<Packet4cf>(const Packet4cf& x)
+{
+  return Packet4cf(_mm256_shuffle_ps(x.v, x.v, _MM_SHUFFLE(2, 3, 0 ,1)));
+}
+
+//---------- double ----------
+struct Packet2cd
+{
+  EIGEN_STRONG_INLINE Packet2cd() {}
+  EIGEN_STRONG_INLINE explicit Packet2cd(const __m256d& a) : v(a) {}
+  __m256d  v;
+};
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet2cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 0,
+    size = 2,
+    HasHalfPacket = 1,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet2cd> { typedef std::complex<double> type; enum {size=2, alignment=Aligned32}; typedef Packet1cd half; };
+
+template<> EIGEN_STRONG_INLINE Packet2cd padd<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_add_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd psub<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_sub_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pnegate(const Packet2cd& a) { return Packet2cd(pnegate(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pconj(const Packet2cd& a)
+{
+  const __m256d mask = _mm256_castsi256_pd(_mm256_set_epi32(0x80000000,0x0,0x0,0x0,0x80000000,0x0,0x0,0x0));
+  return Packet2cd(_mm256_xor_pd(a.v,mask));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pmul<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
+{
+  __m256d tmp1 = _mm256_shuffle_pd(a.v,a.v,0x0);
+  __m256d even = _mm256_mul_pd(tmp1, b.v);
+  __m256d tmp2 = _mm256_shuffle_pd(a.v,a.v,0xF);
+  __m256d tmp3 = _mm256_shuffle_pd(b.v,b.v,0x5);
+  __m256d odd  = _mm256_mul_pd(tmp2, tmp3);
+  return Packet2cd(_mm256_addsub_pd(even, odd));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pand   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_and_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd por    <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_or_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pxor   <Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_xor_pd(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cd pandnot<Packet2cd>(const Packet2cd& a, const Packet2cd& b) { return Packet2cd(_mm256_andnot_pd(a.v,b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet2cd pload <Packet2cd>(const std::complex<double>* from)
+{ EIGEN_DEBUG_ALIGNED_LOAD return Packet2cd(pload<Packet4d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cd ploadu<Packet2cd>(const std::complex<double>* from)
+{ EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cd(ploadu<Packet4d>((const double*)from)); }
+
+template<> EIGEN_STRONG_INLINE Packet2cd pset1<Packet2cd>(const std::complex<double>& from)
+{
+  // in case casting to a __m128d* is really not safe, then we can still fallback to this version: (much slower though)
+//   return Packet2cd(_mm256_loadu2_m128d((const double*)&from,(const double*)&from));
+    return Packet2cd(_mm256_broadcast_pd((const __m128d*)(const void*)&from));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd ploaddup<Packet2cd>(const std::complex<double>* from) { return pset1<Packet2cd>(*from); }
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet2cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet2cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2cd pgather<std::complex<double>, Packet2cd>(const std::complex<double>* from, Index stride)
+{
+  return Packet2cd(_mm256_set_pd(std::imag(from[1*stride]), std::real(from[1*stride]),
+				 std::imag(from[0*stride]), std::real(from[0*stride])));
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet2cd>(std::complex<double>* to, const Packet2cd& from, Index stride)
+{
+  __m128d low = _mm256_extractf128_pd(from.v, 0);
+  to[stride*0] = std::complex<double>(_mm_cvtsd_f64(low), _mm_cvtsd_f64(_mm_shuffle_pd(low, low, 1)));
+  __m128d high = _mm256_extractf128_pd(from.v, 1);
+  to[stride*1] = std::complex<double>(_mm_cvtsd_f64(high), _mm_cvtsd_f64(_mm_shuffle_pd(high, high, 1)));
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> pfirst<Packet2cd>(const Packet2cd& a)
+{
+  __m128d low = _mm256_extractf128_pd(a.v, 0);
+  EIGEN_ALIGN16 double res[2];
+  _mm_store_pd(res, low);
+  return std::complex<double>(res[0],res[1]);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd preverse(const Packet2cd& a) {
+  __m256d result = _mm256_permute2f128_pd(a.v, a.v, 1);
+  return Packet2cd(result);
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet2cd>(const Packet2cd& a)
+{
+  return predux(padd(Packet1cd(_mm256_extractf128_pd(a.v,0)),
+                     Packet1cd(_mm256_extractf128_pd(a.v,1))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd preduxp<Packet2cd>(const Packet2cd* vecs)
+{
+  Packet4d t0 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 0 + (2<<4));
+  Packet4d t1 = _mm256_permute2f128_pd(vecs[0].v,vecs[1].v, 1 + (3<<4));
+
+  return Packet2cd(_mm256_add_pd(t0,t1));
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet2cd>(const Packet2cd& a)
+{
+  return predux(pmul(Packet1cd(_mm256_extractf128_pd(a.v,0)),
+                     Packet1cd(_mm256_extractf128_pd(a.v,1))));
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet2cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2cd& first, const Packet2cd& second)
+  {
+    if (Offset==0) return;
+    palign_impl<Offset*2,Packet4d>::run(first.v, second.v);
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet2cd, Packet2cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet2cd pmadd(const Packet2cd& x, const Packet2cd& y, const Packet2cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cd pmul(const Packet2cd& a, const Packet2cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cd,Packet4d)
+
+template<> EIGEN_STRONG_INLINE Packet2cd pdiv<Packet2cd>(const Packet2cd& a, const Packet2cd& b)
+{
+  Packet2cd num = pmul(a, pconj(b));
+  __m256d tmp = _mm256_mul_pd(b.v, b.v);
+  __m256d denom = _mm256_hadd_pd(tmp, tmp);
+  return Packet2cd(_mm256_div_pd(num.v, denom));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pcplxflip<Packet2cd>(const Packet2cd& x)
+{
+  return Packet2cd(_mm256_shuffle_pd(x.v, x.v, 0x5));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4cf,4>& kernel) {
+  __m256d P0 = _mm256_castps_pd(kernel.packet[0].v);
+  __m256d P1 = _mm256_castps_pd(kernel.packet[1].v);
+  __m256d P2 = _mm256_castps_pd(kernel.packet[2].v);
+  __m256d P3 = _mm256_castps_pd(kernel.packet[3].v);
+
+  __m256d T0 = _mm256_shuffle_pd(P0, P1, 15);
+  __m256d T1 = _mm256_shuffle_pd(P0, P1, 0);
+  __m256d T2 = _mm256_shuffle_pd(P2, P3, 15);
+  __m256d T3 = _mm256_shuffle_pd(P2, P3, 0);
+
+  kernel.packet[1].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 32));
+  kernel.packet[3].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T0, T2, 49));
+  kernel.packet[0].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 32));
+  kernel.packet[2].v = _mm256_castpd_ps(_mm256_permute2f128_pd(T1, T3, 49));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2cd,2>& kernel) {
+  __m256d tmp = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 0+(2<<4));
+  kernel.packet[1].v = _mm256_permute2f128_pd(kernel.packet[0].v, kernel.packet[1].v, 1+(3<<4));
+ kernel.packet[0].v = tmp;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertfirst(const Packet4cf& a, std::complex<float> b)
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,1|2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertfirst(const Packet2cd& a, std::complex<double> b)
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,1|2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4cf pinsertlast(const Packet4cf& a, std::complex<float> b)
+{
+  return Packet4cf(_mm256_blend_ps(a.v,pset1<Packet4cf>(b).v,(1<<7)|(1<<6)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cd pinsertlast(const Packet2cd& a, std::complex<double> b)
+{
+  return Packet2cd(_mm256_blend_pd(a.v,pset1<Packet2cd>(b).v,(1<<3)|(1<<2)));
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COMPLEX_AVX_H
diff --git a/Eigen/src/Core/arch/AVX/MathFunctions.h b/Eigen/src/Core/arch/AVX/MathFunctions.h
new file mode 100644
index 0000000..6af67ce
--- /dev/null
+++ b/Eigen/src/Core/arch/AVX/MathFunctions.h
@@ -0,0 +1,439 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_MATH_FUNCTIONS_AVX_H
+#define EIGEN_MATH_FUNCTIONS_AVX_H
+
+/* The sin, cos, exp, and log functions of this file are loosely derived from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+namespace Eigen {
+
+namespace internal {
+
+inline Packet8i pshiftleft(Packet8i v, int n)
+{
+#ifdef EIGEN_VECTORIZE_AVX2
+  return _mm256_slli_epi32(v, n);
+#else
+  __m128i lo = _mm_slli_epi32(_mm256_extractf128_si256(v, 0), n);
+  __m128i hi = _mm_slli_epi32(_mm256_extractf128_si256(v, 1), n);
+  return _mm256_insertf128_si256(_mm256_castsi128_si256(lo), (hi), 1);
+#endif
+}
+
+inline Packet8f pshiftright(Packet8f v, int n)
+{
+#ifdef EIGEN_VECTORIZE_AVX2
+  return _mm256_cvtepi32_ps(_mm256_srli_epi32(_mm256_castps_si256(v), n));
+#else
+  __m128i lo = _mm_srli_epi32(_mm256_extractf128_si256(_mm256_castps_si256(v), 0), n);
+  __m128i hi = _mm_srli_epi32(_mm256_extractf128_si256(_mm256_castps_si256(v), 1), n);
+  return _mm256_cvtepi32_ps(_mm256_insertf128_si256(_mm256_castsi128_si256(lo), (hi), 1));
+#endif
+}
+
+// Sine function
+// Computes sin(x) by wrapping x to the interval [-Pi/4,3*Pi/4] and
+// evaluating interpolants in [-Pi/4,Pi/4] or [Pi/4,3*Pi/4]. The interpolants
+// are (anti-)symmetric and thus have only odd/even coefficients
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+psin<Packet8f>(const Packet8f& _x) {
+  Packet8f x = _x;
+
+  // Some useful values.
+  _EIGEN_DECLARE_CONST_Packet8i(one, 1);
+  _EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(two, 2.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(one_over_four, 0.25f);
+  _EIGEN_DECLARE_CONST_Packet8f(one_over_pi, 3.183098861837907e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_first, -3.140625000000000e+00f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_second, -9.670257568359375e-04f);
+  _EIGEN_DECLARE_CONST_Packet8f(neg_pi_third, -6.278329571784980e-07f);
+  _EIGEN_DECLARE_CONST_Packet8f(four_over_pi, 1.273239544735163e+00f);
+
+  // Map x from [-Pi/4,3*Pi/4] to z in [-1,3] and subtract the shifted period.
+  Packet8f z = pmul(x, p8f_one_over_pi);
+  Packet8f shift = _mm256_floor_ps(padd(z, p8f_one_over_four));
+  x = pmadd(shift, p8f_neg_pi_first, x);
+  x = pmadd(shift, p8f_neg_pi_second, x);
+  x = pmadd(shift, p8f_neg_pi_third, x);
+  z = pmul(x, p8f_four_over_pi);
+
+  // Make a mask for the entries that need flipping, i.e. wherever the shift
+  // is odd.
+  Packet8i shift_ints = _mm256_cvtps_epi32(shift);
+  Packet8i shift_isodd = _mm256_castps_si256(_mm256_and_ps(_mm256_castsi256_ps(shift_ints), _mm256_castsi256_ps(p8i_one)));
+  Packet8i sign_flip_mask = pshiftleft(shift_isodd, 31);
+
+  // Create a mask for which interpolant to use, i.e. if z > 1, then the mask
+  // is set to ones for that entry.
+  Packet8f ival_mask = _mm256_cmp_ps(z, p8f_one, _CMP_GT_OQ);
+
+  // Evaluate the polynomial for the interval [1,3] in z.
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_0, 9.999999724233232e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_2, -3.084242535619928e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_4, 1.584991525700324e-02f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_right_6, -3.188805084631342e-04f);
+  Packet8f z_minus_two = psub(z, p8f_two);
+  Packet8f z_minus_two2 = pmul(z_minus_two, z_minus_two);
+  Packet8f right = pmadd(p8f_coeff_right_6, z_minus_two2, p8f_coeff_right_4);
+  right = pmadd(right, z_minus_two2, p8f_coeff_right_2);
+  right = pmadd(right, z_minus_two2, p8f_coeff_right_0);
+
+  // Evaluate the polynomial for the interval [-1,1] in z.
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_1, 7.853981525427295e-01f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_3, -8.074536727092352e-02f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_5, 2.489871967827018e-03f);
+  _EIGEN_DECLARE_CONST_Packet8f(coeff_left_7, -3.587725841214251e-05f);
+  Packet8f z2 = pmul(z, z);
+  Packet8f left = pmadd(p8f_coeff_left_7, z2, p8f_coeff_left_5);
+  left = pmadd(left, z2, p8f_coeff_left_3);
+  left = pmadd(left, z2, p8f_coeff_left_1);
+  left = pmul(left, z);
+
+  // Assemble the results, i.e. select the left and right polynomials.
+  left = _mm256_andnot_ps(ival_mask, left);
+  right = _mm256_and_ps(ival_mask, right);
+  Packet8f res = _mm256_or_ps(left, right);
+
+  // Flip the sign on the odd intervals and return the result.
+  res = _mm256_xor_ps(res, _mm256_castsi256_ps(sign_flip_mask));
+  return res;
+}
+
+// Natural logarithm
+// Computes log(x) as log(2^e * m) = C*e + log(m), where the constant C =log(2)
+// and m is in the range [sqrt(1/2),sqrt(2)). In this range, the logarithm can
+// be easily approximated by a polynomial centered on m=1 for stability.
+// TODO(gonnet): Further reduce the interval allowing for lower-degree
+//               polynomial interpolants -> ... -> profit!
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+plog<Packet8f>(const Packet8f& _x) {
+  Packet8f x = _x;
+  _EIGEN_DECLARE_CONST_Packet8f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(126f, 126.0f);
+
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inv_mant_mask, ~0x7f800000);
+
+  // The smallest non denormalized float number.
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(min_norm_pos, 0x00800000);
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(minus_inf, 0xff800000);
+
+  // Polynomial coefficients.
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_SQRTHF, 0.707106781186547524f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p0, 7.0376836292E-2f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p1, -1.1514610310E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p2, 1.1676998740E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p3, -1.2420140846E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p4, +1.4249322787E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p5, -1.6668057665E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p6, +2.0000714765E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p7, -2.4999993993E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_p8, +3.3333331174E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_q1, -2.12194440e-4f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_log_q2, 0.693359375f);
+
+  Packet8f invalid_mask = _mm256_cmp_ps(x, _mm256_setzero_ps(), _CMP_NGE_UQ); // not greater equal is true if x is NaN
+  Packet8f iszero_mask = _mm256_cmp_ps(x, _mm256_setzero_ps(), _CMP_EQ_OQ);
+
+  // Truncate input values to the minimum positive normal.
+  x = pmax(x, p8f_min_norm_pos);
+
+  Packet8f emm0 = pshiftright(x,23);
+  Packet8f e = _mm256_sub_ps(emm0, p8f_126f);
+
+  // Set the exponents to -1, i.e. x are in the range [0.5,1).
+  x = _mm256_and_ps(x, p8f_inv_mant_mask);
+  x = _mm256_or_ps(x, p8f_half);
+
+  // part2: Shift the inputs from the range [0.5,1) to [sqrt(1/2),sqrt(2))
+  // and shift by -1. The values are then centered around 0, which improves
+  // the stability of the polynomial evaluation.
+  //   if( x < SQRTHF ) {
+  //     e -= 1;
+  //     x = x + x - 1.0;
+  //   } else { x = x - 1.0; }
+  Packet8f mask = _mm256_cmp_ps(x, p8f_cephes_SQRTHF, _CMP_LT_OQ);
+  Packet8f tmp = _mm256_and_ps(x, mask);
+  x = psub(x, p8f_1);
+  e = psub(e, _mm256_and_ps(p8f_1, mask));
+  x = padd(x, tmp);
+
+  Packet8f x2 = pmul(x, x);
+  Packet8f x3 = pmul(x2, x);
+
+  // Evaluate the polynomial approximant of degree 8 in three parts, probably
+  // to improve instruction-level parallelism.
+  Packet8f y, y1, y2;
+  y = pmadd(p8f_cephes_log_p0, x, p8f_cephes_log_p1);
+  y1 = pmadd(p8f_cephes_log_p3, x, p8f_cephes_log_p4);
+  y2 = pmadd(p8f_cephes_log_p6, x, p8f_cephes_log_p7);
+  y = pmadd(y, x, p8f_cephes_log_p2);
+  y1 = pmadd(y1, x, p8f_cephes_log_p5);
+  y2 = pmadd(y2, x, p8f_cephes_log_p8);
+  y = pmadd(y, x3, y1);
+  y = pmadd(y, x3, y2);
+  y = pmul(y, x3);
+
+  // Add the logarithm of the exponent back to the result of the interpolation.
+  y1 = pmul(e, p8f_cephes_log_q1);
+  tmp = pmul(x2, p8f_half);
+  y = padd(y, y1);
+  x = psub(x, tmp);
+  y2 = pmul(e, p8f_cephes_log_q2);
+  x = padd(x, y);
+  x = padd(x, y2);
+
+  // Filter out invalid inputs, i.e. negative arg will be NAN, 0 will be -INF.
+  return _mm256_or_ps(
+      _mm256_andnot_ps(iszero_mask, _mm256_or_ps(x, invalid_mask)),
+      _mm256_and_ps(iszero_mask, p8f_minus_inf));
+}
+
+// Exponential function. Works by writing "x = m*log(2) + r" where
+// "m = floor(x/log(2)+1/2)" and "r" is the remainder. The result is then
+// "exp(x) = 2^m*exp(r)" where exp(r) is in the range [-1,1).
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+pexp<Packet8f>(const Packet8f& _x) {
+  _EIGEN_DECLARE_CONST_Packet8f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet8f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(127, 127.0f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(exp_hi, 88.3762626647950f);
+  _EIGEN_DECLARE_CONST_Packet8f(exp_lo, -88.3762626647949f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_LOG2EF, 1.44269504088896341f);
+
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p0, 1.9875691500E-4f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p1, 1.3981999507E-3f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p2, 8.3334519073E-3f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p3, 4.1665795894E-2f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p4, 1.6666665459E-1f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_p5, 5.0000001201E-1f);
+
+  // Clamp x.
+  Packet8f x = pmax(pmin(_x, p8f_exp_hi), p8f_exp_lo);
+
+  // Express exp(x) as exp(m*ln(2) + r), start by extracting
+  // m = floor(x/ln(2) + 0.5).
+  Packet8f m = _mm256_floor_ps(pmadd(x, p8f_cephes_LOG2EF, p8f_half));
+
+// Get r = x - m*ln(2). If no FMA instructions are available, m*ln(2) is
+// subtracted out in two parts, m*C1+m*C2 = m*ln(2), to avoid accumulating
+// truncation errors. Note that we don't use the "pmadd" function here to
+// ensure that a precision-preserving FMA instruction is used.
+#ifdef EIGEN_VECTORIZE_FMA
+  _EIGEN_DECLARE_CONST_Packet8f(nln2, -0.6931471805599453f);
+  Packet8f r = _mm256_fmadd_ps(m, p8f_nln2, x);
+#else
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_C1, 0.693359375f);
+  _EIGEN_DECLARE_CONST_Packet8f(cephes_exp_C2, -2.12194440e-4f);
+  Packet8f r = psub(x, pmul(m, p8f_cephes_exp_C1));
+  r = psub(r, pmul(m, p8f_cephes_exp_C2));
+#endif
+
+  Packet8f r2 = pmul(r, r);
+
+  // TODO(gonnet): Split into odd/even polynomials and try to exploit
+  //               instruction-level parallelism.
+  Packet8f y = p8f_cephes_exp_p0;
+  y = pmadd(y, r, p8f_cephes_exp_p1);
+  y = pmadd(y, r, p8f_cephes_exp_p2);
+  y = pmadd(y, r, p8f_cephes_exp_p3);
+  y = pmadd(y, r, p8f_cephes_exp_p4);
+  y = pmadd(y, r, p8f_cephes_exp_p5);
+  y = pmadd(y, r2, r);
+  y = padd(y, p8f_1);
+
+  // Build emm0 = 2^m.
+  Packet8i emm0 = _mm256_cvttps_epi32(padd(m, p8f_127));
+  emm0 = pshiftleft(emm0, 23);
+
+  // Return 2^m * exp(r).
+  return pmax(pmul(y, _mm256_castsi256_ps(emm0)), _x);
+}
+
+// Hyperbolic Tangent function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+ptanh<Packet8f>(const Packet8f& x) {
+  return internal::generic_fast_tanh_float(x);
+}
+
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4d
+pexp<Packet4d>(const Packet4d& _x) {
+  Packet4d x = _x;
+
+  _EIGEN_DECLARE_CONST_Packet4d(1, 1.0);
+  _EIGEN_DECLARE_CONST_Packet4d(2, 2.0);
+  _EIGEN_DECLARE_CONST_Packet4d(half, 0.5);
+
+  _EIGEN_DECLARE_CONST_Packet4d(exp_hi, 709.437);
+  _EIGEN_DECLARE_CONST_Packet4d(exp_lo, -709.436139303);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_LOG2EF, 1.4426950408889634073599);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p0, 1.26177193074810590878e-4);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p1, 3.02994407707441961300e-2);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q0, 3.00198505138664455042e-6);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q1, 2.52448340349684104192e-3);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q2, 2.27265548208155028766e-1);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_q3, 2.00000000000000000009e0);
+
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_C1, 0.693145751953125);
+  _EIGEN_DECLARE_CONST_Packet4d(cephes_exp_C2, 1.42860682030941723212e-6);
+  _EIGEN_DECLARE_CONST_Packet4i(1023, 1023);
+
+  Packet4d tmp, fx;
+
+  // clamp x
+  x = pmax(pmin(x, p4d_exp_hi), p4d_exp_lo);
+  // Express exp(x) as exp(g + n*log(2)).
+  fx = pmadd(p4d_cephes_LOG2EF, x, p4d_half);
+
+  // Get the integer modulus of log(2), i.e. the "n" described above.
+  fx = _mm256_floor_pd(fx);
+
+  // Get the remainder modulo log(2), i.e. the "g" described above. Subtract
+  // n*log(2) out in two steps, i.e. n*C1 + n*C2, C1+C2=log2 to get the last
+  // digits right.
+  tmp = pmul(fx, p4d_cephes_exp_C1);
+  Packet4d z = pmul(fx, p4d_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  Packet4d x2 = pmul(x, x);
+
+  // Evaluate the numerator polynomial of the rational interpolant.
+  Packet4d px = p4d_cephes_exp_p0;
+  px = pmadd(px, x2, p4d_cephes_exp_p1);
+  px = pmadd(px, x2, p4d_cephes_exp_p2);
+  px = pmul(px, x);
+
+  // Evaluate the denominator polynomial of the rational interpolant.
+  Packet4d qx = p4d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p4d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p4d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p4d_cephes_exp_q3);
+
+  // I don't really get this bit, copied from the SSE2 routines, so...
+  // TODO(gonnet): Figure out what is going on here, perhaps find a better
+  // rational interpolant?
+  x = _mm256_div_pd(px, psub(qx, px));
+  x = pmadd(p4d_2, x, p4d_1);
+
+  // Build e=2^n by constructing the exponents in a 128-bit vector and
+  // shifting them to where they belong in double-precision values.
+  __m128i emm0 = _mm256_cvtpd_epi32(fx);
+  emm0 = _mm_add_epi32(emm0, p4i_1023);
+  emm0 = _mm_shuffle_epi32(emm0, _MM_SHUFFLE(3, 1, 2, 0));
+  __m128i lo = _mm_slli_epi64(emm0, 52);
+  __m128i hi = _mm_slli_epi64(_mm_srli_epi64(emm0, 32), 52);
+  __m256i e = _mm256_insertf128_si256(_mm256_setzero_si256(), lo, 0);
+  e = _mm256_insertf128_si256(e, hi, 1);
+
+  // Construct the result 2^n * exp(g) = e * x. The max is used to catch
+  // non-finite values in the input.
+  return pmax(pmul(x, _mm256_castsi256_pd(e)), _x);
+}
+
+// Functions for sqrt.
+// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
+// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
+// exact solution. It does not handle +inf, or denormalized numbers correctly.
+// The main advantage of this approach is not just speed, but also the fact that
+// it can be inlined and pipelined with other computations, further reducing its
+// effective latency. This is similar to Quake3's fast inverse square root.
+// For detail see here: http://www.beyond3d.com/content/articles/8/
+#if EIGEN_FAST_MATH
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8f
+psqrt<Packet8f>(const Packet8f& _x) {
+  Packet8f half = pmul(_x, pset1<Packet8f>(.5f));
+  Packet8f denormal_mask = _mm256_and_ps(
+      _mm256_cmp_ps(_x, pset1<Packet8f>((std::numeric_limits<float>::min)()),
+                    _CMP_LT_OQ),
+      _mm256_cmp_ps(_x, _mm256_setzero_ps(), _CMP_GE_OQ));
+
+  // Compute approximate reciprocal sqrt.
+  Packet8f x = _mm256_rsqrt_ps(_x);
+  // Do a single step of Newton's iteration.
+  x = pmul(x, psub(pset1<Packet8f>(1.5f), pmul(half, pmul(x,x))));
+  // Flush results for denormals to zero.
+  return _mm256_andnot_ps(denormal_mask, pmul(_x,x));
+}
+#else
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f psqrt<Packet8f>(const Packet8f& x) {
+  return _mm256_sqrt_ps(x);
+}
+#endif
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4d psqrt<Packet4d>(const Packet4d& x) {
+  return _mm256_sqrt_pd(x);
+}
+#if EIGEN_FAST_MATH
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f prsqrt<Packet8f>(const Packet8f& _x) {
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(inf, 0x7f800000);
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(nan, 0x7fc00000);
+  _EIGEN_DECLARE_CONST_Packet8f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet8f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(flt_min, 0x00800000);
+
+  Packet8f neg_half = pmul(_x, p8f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  Packet8f le_zero_mask = _mm256_cmp_ps(_x, p8f_flt_min, _CMP_LT_OQ);
+  Packet8f x = _mm256_andnot_ps(le_zero_mask, _mm256_rsqrt_ps(_x));
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  Packet8f neg_mask = _mm256_cmp_ps(_x, _mm256_setzero_ps(), _CMP_LT_OQ);
+  Packet8f zero_mask = _mm256_andnot_ps(neg_mask, le_zero_mask);
+  Packet8f infs_and_nans = _mm256_or_ps(_mm256_and_ps(neg_mask, p8f_nan),
+                                        _mm256_and_ps(zero_mask, p8f_inf));
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8f_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm256_or_ps(x, infs_and_nans);
+}
+
+#else
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet8f prsqrt<Packet8f>(const Packet8f& x) {
+  _EIGEN_DECLARE_CONST_Packet8f(one, 1.0f);
+  return _mm256_div_ps(p8f_one, _mm256_sqrt_ps(x));
+}
+#endif
+
+template <> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4d prsqrt<Packet4d>(const Packet4d& x) {
+  _EIGEN_DECLARE_CONST_Packet4d(one, 1.0);
+  return _mm256_div_pd(p4d_one, _mm256_sqrt_pd(x));
+}
+
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_MATH_FUNCTIONS_AVX_H
diff --git a/Eigen/src/Core/arch/AVX/PacketMath.h b/Eigen/src/Core/arch/AVX/PacketMath.h
new file mode 100644
index 0000000..923a124
--- /dev/null
+++ b/Eigen/src/Core/arch/AVX/PacketMath.h
@@ -0,0 +1,637 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner (benoit.steiner.goog@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_AVX_H
+#define EIGEN_PACKET_MATH_AVX_H
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8
+#endif
+
+#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS (2*sizeof(void*))
+#endif
+
+#ifdef __FMA__
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+#endif
+
+typedef __m256  Packet8f;
+typedef __m256i Packet8i;
+typedef __m256d Packet4d;
+
+template<> struct is_arithmetic<__m256>  { enum { value = true }; };
+template<> struct is_arithmetic<__m256i> { enum { value = true }; };
+template<> struct is_arithmetic<__m256d> { enum { value = true }; };
+
+#define _EIGEN_DECLARE_CONST_Packet8f(NAME,X) \
+  const Packet8f p8f_##NAME = pset1<Packet8f>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet4d(NAME,X) \
+  const Packet4d p4d_##NAME = pset1<Packet4d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet8f_FROM_INT(NAME,X) \
+  const Packet8f p8f_##NAME = _mm256_castsi256_ps(pset1<Packet8i>(X))
+
+#define _EIGEN_DECLARE_CONST_Packet8i(NAME,X) \
+  const Packet8i p8i_##NAME = pset1<Packet8i>(X)
+
+// Use the packet_traits defined in AVX512/PacketMath.h instead if we're going
+// to leverage AVX512 instructions.
+#ifndef EIGEN_VECTORIZE_AVX512
+template<> struct packet_traits<float>  : default_packet_traits
+{
+  typedef Packet8f type;
+  typedef Packet4f half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=8,
+    HasHalfPacket = 1,
+
+    HasDiv  = 1,
+    HasSin  = EIGEN_FAST_MATH,
+    HasCos  = 0,
+    HasLog  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasTanh  = EIGEN_FAST_MATH,
+    HasBlend = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+  };
+};
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet4d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=4,
+    HasHalfPacket = 1,
+
+    HasDiv  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasBlend = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+  };
+};
+#endif
+
+template<> struct scalar_div_cost<float,true> { enum { value = 14 }; };
+template<> struct scalar_div_cost<double,true> { enum { value = 16 }; };
+
+/* Proper support for integers is only provided by AVX2. In the meantime, we'll
+   use SSE instructions and packets to deal with integers.
+template<> struct packet_traits<int>    : default_packet_traits
+{
+  typedef Packet8i type;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=8
+  };
+};
+*/
+
+template<> struct unpacket_traits<Packet8f> { typedef float  type; typedef Packet4f half; enum {size=8, alignment=Aligned32}; };
+template<> struct unpacket_traits<Packet4d> { typedef double type; typedef Packet2d half; enum {size=4, alignment=Aligned32}; };
+template<> struct unpacket_traits<Packet8i> { typedef int    type; typedef Packet4i half; enum {size=8, alignment=Aligned32}; };
+
+template<> EIGEN_STRONG_INLINE Packet8f pset1<Packet8f>(const float&  from) { return _mm256_set1_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4d pset1<Packet4d>(const double& from) { return _mm256_set1_pd(from); }
+template<> EIGEN_STRONG_INLINE Packet8i pset1<Packet8i>(const int&    from) { return _mm256_set1_epi32(from); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pload1<Packet8f>(const float*  from) { return _mm256_broadcast_ss(from); }
+template<> EIGEN_STRONG_INLINE Packet4d pload1<Packet4d>(const double* from) { return _mm256_broadcast_sd(from); }
+
+template<> EIGEN_STRONG_INLINE Packet8f plset<Packet8f>(const float& a) { return _mm256_add_ps(_mm256_set1_ps(a), _mm256_set_ps(7.0,6.0,5.0,4.0,3.0,2.0,1.0,0.0)); }
+template<> EIGEN_STRONG_INLINE Packet4d plset<Packet4d>(const double& a) { return _mm256_add_pd(_mm256_set1_pd(a), _mm256_set_pd(3.0,2.0,1.0,0.0)); }
+
+template<> EIGEN_STRONG_INLINE Packet8f padd<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_add_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d padd<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_add_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f psub<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_sub_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d psub<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_sub_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pnegate(const Packet8f& a)
+{
+  return _mm256_sub_ps(_mm256_set1_ps(0.0),a);
+}
+template<> EIGEN_STRONG_INLINE Packet4d pnegate(const Packet4d& a)
+{
+  return _mm256_sub_pd(_mm256_set1_pd(0.0),a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pconj(const Packet8f& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet4d pconj(const Packet4d& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet8i pconj(const Packet8i& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet8f pmul<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_mul_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pmul<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_mul_pd(a,b); }
+
+
+template<> EIGEN_STRONG_INLINE Packet8f pdiv<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_div_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pdiv<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_div_pd(a,b); }
+template<> EIGEN_STRONG_INLINE Packet8i pdiv<Packet8i>(const Packet8i& /*a*/, const Packet8i& /*b*/)
+{ eigen_assert(false && "packet integer division are not supported by AVX");
+  return pset1<Packet8i>(0);
+}
+
+#ifdef __FMA__
+template<> EIGEN_STRONG_INLINE Packet8f pmadd(const Packet8f& a, const Packet8f& b, const Packet8f& c) {
+#if ( (EIGEN_COMP_GNUC_STRICT && EIGEN_COMP_GNUC<80) || (EIGEN_COMP_CLANG) )
+  // Clang stupidly generates a vfmadd213ps instruction plus some vmovaps on registers,
+  //  and even register spilling with clang>=6.0 (bug 1637).
+  // Gcc stupidly generates a vfmadd132ps instruction.
+  // So let's enforce it to generate a vfmadd231ps instruction since the most common use
+  //  case is to accumulate the result of the product.
+  Packet8f res = c;
+  __asm__("vfmadd231ps %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
+  return res;
+#else
+  return _mm256_fmadd_ps(a,b,c);
+#endif
+}
+template<> EIGEN_STRONG_INLINE Packet4d pmadd(const Packet4d& a, const Packet4d& b, const Packet4d& c) {
+#if ( (EIGEN_COMP_GNUC_STRICT && EIGEN_COMP_GNUC<80) || (EIGEN_COMP_CLANG) )
+  // see above
+  Packet4d res = c;
+  __asm__("vfmadd231pd %[a], %[b], %[c]" : [c] "+x" (res) : [a] "x" (a), [b] "x" (b));
+  return res;
+#else
+  return _mm256_fmadd_pd(a,b,c);
+#endif
+}
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet8f pmin<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_min_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pmin<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_min_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pmax<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_max_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pmax<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_max_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pround<Packet8f>(const Packet8f& a) { return _mm256_round_ps(a, _MM_FROUND_CUR_DIRECTION); }
+template<> EIGEN_STRONG_INLINE Packet4d pround<Packet4d>(const Packet4d& a) { return _mm256_round_pd(a, _MM_FROUND_CUR_DIRECTION); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pceil<Packet8f>(const Packet8f& a) { return _mm256_ceil_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet4d pceil<Packet4d>(const Packet4d& a) { return _mm256_ceil_pd(a); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pfloor<Packet8f>(const Packet8f& a) { return _mm256_floor_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet4d pfloor<Packet4d>(const Packet4d& a) { return _mm256_floor_pd(a); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pand<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_and_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pand<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_and_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f por<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_or_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d por<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_or_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pxor<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_xor_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pxor<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_xor_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pandnot<Packet8f>(const Packet8f& a, const Packet8f& b) { return _mm256_andnot_ps(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4d pandnot<Packet4d>(const Packet4d& a, const Packet4d& b) { return _mm256_andnot_pd(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet8f pload<Packet8f>(const float*   from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm256_load_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4d pload<Packet4d>(const double*  from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm256_load_pd(from); }
+template<> EIGEN_STRONG_INLINE Packet8i pload<Packet8i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm256_load_si256(reinterpret_cast<const __m256i*>(from)); }
+
+template<> EIGEN_STRONG_INLINE Packet8f ploadu<Packet8f>(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm256_loadu_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4d ploadu<Packet4d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm256_loadu_pd(from); }
+template<> EIGEN_STRONG_INLINE Packet8i ploadu<Packet8i>(const int* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm256_loadu_si256(reinterpret_cast<const __m256i*>(from)); }
+
+// Loads 4 floats from memory a returns the packet {a0, a0  a1, a1, a2, a2, a3, a3}
+template<> EIGEN_STRONG_INLINE Packet8f ploaddup<Packet8f>(const float* from)
+{
+  // TODO try to find a way to avoid the need of a temporary register
+//   Packet8f tmp  = _mm256_castps128_ps256(_mm_loadu_ps(from));
+//   tmp = _mm256_insertf128_ps(tmp, _mm_movehl_ps(_mm256_castps256_ps128(tmp),_mm256_castps256_ps128(tmp)), 1);
+//   return _mm256_unpacklo_ps(tmp,tmp);
+  
+  // _mm256_insertf128_ps is very slow on Haswell, thus:
+  Packet8f tmp = _mm256_broadcast_ps((const __m128*)(const void*)from);
+  // mimic an "inplace" permutation of the lower 128bits using a blend
+  tmp = _mm256_blend_ps(tmp,_mm256_castps128_ps256(_mm_permute_ps( _mm256_castps256_ps128(tmp), _MM_SHUFFLE(1,0,1,0))), 15);
+  // then we can perform a consistent permutation on the global register to get everything in shape:
+  return  _mm256_permute_ps(tmp, _MM_SHUFFLE(3,3,2,2));
+}
+// Loads 2 doubles from memory a returns the packet {a0, a0  a1, a1}
+template<> EIGEN_STRONG_INLINE Packet4d ploaddup<Packet4d>(const double* from)
+{
+  Packet4d tmp = _mm256_broadcast_pd((const __m128d*)(const void*)from);
+  return  _mm256_permute_pd(tmp, 3<<2);
+}
+
+// Loads 2 floats from memory a returns the packet {a0, a0  a0, a0, a1, a1, a1, a1}
+template<> EIGEN_STRONG_INLINE Packet8f ploadquad<Packet8f>(const float* from)
+{
+  Packet8f tmp = _mm256_castps128_ps256(_mm_broadcast_ss(from));
+  return _mm256_insertf128_ps(tmp, _mm_broadcast_ss(from+1), 1);
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet8f& from) { EIGEN_DEBUG_ALIGNED_STORE _mm256_store_ps(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet4d& from) { EIGEN_DEBUG_ALIGNED_STORE _mm256_store_pd(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet8i& from) { EIGEN_DEBUG_ALIGNED_STORE _mm256_storeu_si256(reinterpret_cast<__m256i*>(to), from); }
+
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*   to, const Packet8f& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm256_storeu_ps(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet4d& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm256_storeu_pd(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*       to, const Packet8i& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm256_storeu_si256(reinterpret_cast<__m256i*>(to), from); }
+
+// NOTE: leverage _mm256_i32gather_ps and _mm256_i32gather_pd if AVX2 instructions are available
+// NOTE: for the record the following seems to be slower: return _mm256_i32gather_ps(from, _mm256_set1_epi32(stride), 4);
+template<> EIGEN_DEVICE_FUNC inline Packet8f pgather<float, Packet8f>(const float* from, Index stride)
+{
+  return _mm256_set_ps(from[7*stride], from[6*stride], from[5*stride], from[4*stride],
+                       from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4d pgather<double, Packet4d>(const double* from, Index stride)
+{
+  return _mm256_set_pd(from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet8f>(float* to, const Packet8f& from, Index stride)
+{
+  __m128 low = _mm256_extractf128_ps(from, 0);
+  to[stride*0] = _mm_cvtss_f32(low);
+  to[stride*1] = _mm_cvtss_f32(_mm_shuffle_ps(low, low, 1));
+  to[stride*2] = _mm_cvtss_f32(_mm_shuffle_ps(low, low, 2));
+  to[stride*3] = _mm_cvtss_f32(_mm_shuffle_ps(low, low, 3));
+
+  __m128 high = _mm256_extractf128_ps(from, 1);
+  to[stride*4] = _mm_cvtss_f32(high);
+  to[stride*5] = _mm_cvtss_f32(_mm_shuffle_ps(high, high, 1));
+  to[stride*6] = _mm_cvtss_f32(_mm_shuffle_ps(high, high, 2));
+  to[stride*7] = _mm_cvtss_f32(_mm_shuffle_ps(high, high, 3));
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet4d>(double* to, const Packet4d& from, Index stride)
+{
+  __m128d low = _mm256_extractf128_pd(from, 0);
+  to[stride*0] = _mm_cvtsd_f64(low);
+  to[stride*1] = _mm_cvtsd_f64(_mm_shuffle_pd(low, low, 1));
+  __m128d high = _mm256_extractf128_pd(from, 1);
+  to[stride*2] = _mm_cvtsd_f64(high);
+  to[stride*3] = _mm_cvtsd_f64(_mm_shuffle_pd(high, high, 1));
+}
+
+template<> EIGEN_STRONG_INLINE void pstore1<Packet8f>(float* to, const float& a)
+{
+  Packet8f pa = pset1<Packet8f>(a);
+  pstore(to, pa);
+}
+template<> EIGEN_STRONG_INLINE void pstore1<Packet4d>(double* to, const double& a)
+{
+  Packet4d pa = pset1<Packet4d>(a);
+  pstore(to, pa);
+}
+template<> EIGEN_STRONG_INLINE void pstore1<Packet8i>(int* to, const int& a)
+{
+  Packet8i pa = pset1<Packet8i>(a);
+  pstore(to, pa);
+}
+
+#ifndef EIGEN_VECTORIZE_AVX512
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+#endif
+
+template<> EIGEN_STRONG_INLINE float  pfirst<Packet8f>(const Packet8f& a) {
+  return _mm_cvtss_f32(_mm256_castps256_ps128(a));
+}
+template<> EIGEN_STRONG_INLINE double pfirst<Packet4d>(const Packet4d& a) {
+  return _mm_cvtsd_f64(_mm256_castpd256_pd128(a));
+}
+template<> EIGEN_STRONG_INLINE int    pfirst<Packet8i>(const Packet8i& a) {
+  return _mm_cvtsi128_si32(_mm256_castsi256_si128(a));
+}
+
+
+template<> EIGEN_STRONG_INLINE Packet8f preverse(const Packet8f& a)
+{
+  __m256 tmp = _mm256_shuffle_ps(a,a,0x1b);
+  return _mm256_permute2f128_ps(tmp, tmp, 1);
+}
+template<> EIGEN_STRONG_INLINE Packet4d preverse(const Packet4d& a)
+{
+   __m256d tmp = _mm256_shuffle_pd(a,a,5);
+  return _mm256_permute2f128_pd(tmp, tmp, 1);
+  #if 0
+  // This version is unlikely to be faster as _mm256_shuffle_ps and _mm256_permute_pd
+  // exhibit the same latency/throughput, but it is here for future reference/benchmarking...
+  __m256d swap_halves = _mm256_permute2f128_pd(a,a,1);
+    return _mm256_permute_pd(swap_halves,5);
+  #endif
+}
+
+// pabs should be ok
+template<> EIGEN_STRONG_INLINE Packet8f pabs(const Packet8f& a)
+{
+  const Packet8f mask = _mm256_castsi256_ps(_mm256_setr_epi32(0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF));
+  return _mm256_and_ps(a,mask);
+}
+template<> EIGEN_STRONG_INLINE Packet4d pabs(const Packet4d& a)
+{
+  const Packet4d mask = _mm256_castsi256_pd(_mm256_setr_epi32(0xFFFFFFFF,0x7FFFFFFF,0xFFFFFFFF,0x7FFFFFFF,0xFFFFFFFF,0x7FFFFFFF,0xFFFFFFFF,0x7FFFFFFF));
+  return _mm256_and_pd(a,mask);
+}
+
+// preduxp should be ok
+// FIXME: why is this ok? why isn't the simply implementation working as expected?
+template<> EIGEN_STRONG_INLINE Packet8f preduxp<Packet8f>(const Packet8f* vecs)
+{
+    __m256 hsum1 = _mm256_hadd_ps(vecs[0], vecs[1]);
+    __m256 hsum2 = _mm256_hadd_ps(vecs[2], vecs[3]);
+    __m256 hsum3 = _mm256_hadd_ps(vecs[4], vecs[5]);
+    __m256 hsum4 = _mm256_hadd_ps(vecs[6], vecs[7]);
+
+    __m256 hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+    __m256 hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+    __m256 hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+    __m256 hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+    __m256 perm1 =  _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+    __m256 perm2 =  _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+    __m256 perm3 =  _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+    __m256 perm4 =  _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+    __m256 sum1 = _mm256_add_ps(perm1, hsum5);
+    __m256 sum2 = _mm256_add_ps(perm2, hsum6);
+    __m256 sum3 = _mm256_add_ps(perm3, hsum7);
+    __m256 sum4 = _mm256_add_ps(perm4, hsum8);
+
+    __m256 blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+    __m256 blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+    __m256 final = _mm256_blend_ps(blend1, blend2, 0xf0);
+    return final;
+}
+template<> EIGEN_STRONG_INLINE Packet4d preduxp<Packet4d>(const Packet4d* vecs)
+{
+ Packet4d tmp0, tmp1;
+
+  tmp0 = _mm256_hadd_pd(vecs[0], vecs[1]);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs[2], vecs[3]);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  return _mm256_blend_pd(tmp0, tmp1, 0xC);
+}
+
+template<> EIGEN_STRONG_INLINE float predux<Packet8f>(const Packet8f& a)
+{
+  return predux(Packet4f(_mm_add_ps(_mm256_castps256_ps128(a),_mm256_extractf128_ps(a,1))));
+}
+template<> EIGEN_STRONG_INLINE double predux<Packet4d>(const Packet4d& a)
+{
+  return predux(Packet2d(_mm_add_pd(_mm256_castpd256_pd128(a),_mm256_extractf128_pd(a,1))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f predux_downto4<Packet8f>(const Packet8f& a)
+{
+  return _mm_add_ps(_mm256_castps256_ps128(a),_mm256_extractf128_ps(a,1));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_mul<Packet8f>(const Packet8f& a)
+{
+  Packet8f tmp;
+  tmp = _mm256_mul_ps(a, _mm256_permute2f128_ps(a,a,1));
+  tmp = _mm256_mul_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
+  return pfirst(_mm256_mul_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
+}
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet4d>(const Packet4d& a)
+{
+  Packet4d tmp;
+  tmp = _mm256_mul_pd(a, _mm256_permute2f128_pd(a,a,1));
+  return pfirst(_mm256_mul_pd(tmp, _mm256_shuffle_pd(tmp,tmp,1)));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_min<Packet8f>(const Packet8f& a)
+{
+  Packet8f tmp = _mm256_min_ps(a, _mm256_permute2f128_ps(a,a,1));
+  tmp = _mm256_min_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
+  return pfirst(_mm256_min_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
+}
+template<> EIGEN_STRONG_INLINE double predux_min<Packet4d>(const Packet4d& a)
+{
+  Packet4d tmp = _mm256_min_pd(a, _mm256_permute2f128_pd(a,a,1));
+  return pfirst(_mm256_min_pd(tmp, _mm256_shuffle_pd(tmp, tmp, 1)));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_max<Packet8f>(const Packet8f& a)
+{
+  Packet8f tmp = _mm256_max_ps(a, _mm256_permute2f128_ps(a,a,1));
+  tmp = _mm256_max_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
+  return pfirst(_mm256_max_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
+}
+
+template<> EIGEN_STRONG_INLINE double predux_max<Packet4d>(const Packet4d& a)
+{
+  Packet4d tmp = _mm256_max_pd(a, _mm256_permute2f128_pd(a,a,1));
+  return pfirst(_mm256_max_pd(tmp, _mm256_shuffle_pd(tmp, tmp, 1)));
+}
+
+
+template<int Offset>
+struct palign_impl<Offset,Packet8f>
+{
+  static EIGEN_STRONG_INLINE void run(Packet8f& first, const Packet8f& second)
+  {
+    if (Offset==1)
+    {
+      first = _mm256_blend_ps(first, second, 1);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(0,3,2,1));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_blend_ps(tmp1, tmp2, 0x88);
+    }
+    else if (Offset==2)
+    {
+      first = _mm256_blend_ps(first, second, 3);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(1,0,3,2));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_blend_ps(tmp1, tmp2, 0xcc);
+    }
+    else if (Offset==3)
+    {
+      first = _mm256_blend_ps(first, second, 7);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(2,1,0,3));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_blend_ps(tmp1, tmp2, 0xee);
+    }
+    else if (Offset==4)
+    {
+      first = _mm256_blend_ps(first, second, 15);
+      Packet8f tmp1 = _mm256_permute_ps (first, _MM_SHUFFLE(3,2,1,0));
+      Packet8f tmp2 = _mm256_permute2f128_ps (tmp1, tmp1, 1);
+      first = _mm256_permute_ps(tmp2, _MM_SHUFFLE(3,2,1,0));
+    }
+    else if (Offset==5)
+    {
+      first = _mm256_blend_ps(first, second, 31);
+      first = _mm256_permute2f128_ps(first, first, 1);
+      Packet8f tmp = _mm256_permute_ps (first, _MM_SHUFFLE(0,3,2,1));
+      first = _mm256_permute2f128_ps(tmp, tmp, 1);
+      first = _mm256_blend_ps(tmp, first, 0x88);
+    }
+    else if (Offset==6)
+    {
+      first = _mm256_blend_ps(first, second, 63);
+      first = _mm256_permute2f128_ps(first, first, 1);
+      Packet8f tmp = _mm256_permute_ps (first, _MM_SHUFFLE(1,0,3,2));
+      first = _mm256_permute2f128_ps(tmp, tmp, 1);
+      first = _mm256_blend_ps(tmp, first, 0xcc);
+    }
+    else if (Offset==7)
+    {
+      first = _mm256_blend_ps(first, second, 127);
+      first = _mm256_permute2f128_ps(first, first, 1);
+      Packet8f tmp = _mm256_permute_ps (first, _MM_SHUFFLE(2,1,0,3));
+      first = _mm256_permute2f128_ps(tmp, tmp, 1);
+      first = _mm256_blend_ps(tmp, first, 0xee);
+    }
+  }
+};
+
+template<int Offset>
+struct palign_impl<Offset,Packet4d>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4d& first, const Packet4d& second)
+  {
+    if (Offset==1)
+    {
+      first = _mm256_blend_pd(first, second, 1);
+      __m256d tmp = _mm256_permute_pd(first, 5);
+      first = _mm256_permute2f128_pd(tmp, tmp, 1);
+      first = _mm256_blend_pd(tmp, first, 0xA);
+    }
+    else if (Offset==2)
+    {
+      first = _mm256_blend_pd(first, second, 3);
+      first = _mm256_permute2f128_pd(first, first, 1);
+    }
+    else if (Offset==3)
+    {
+      first = _mm256_blend_pd(first, second, 7);
+      __m256d tmp = _mm256_permute_pd(first, 5);
+      first = _mm256_permute2f128_pd(tmp, tmp, 1);
+      first = _mm256_blend_pd(tmp, first, 5);
+    }
+  }
+};
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet8f,8>& kernel) {
+  __m256 T0 = _mm256_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T1 = _mm256_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T2 = _mm256_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m256 T3 = _mm256_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+  __m256 T4 = _mm256_unpacklo_ps(kernel.packet[4], kernel.packet[5]);
+  __m256 T5 = _mm256_unpackhi_ps(kernel.packet[4], kernel.packet[5]);
+  __m256 T6 = _mm256_unpacklo_ps(kernel.packet[6], kernel.packet[7]);
+  __m256 T7 = _mm256_unpackhi_ps(kernel.packet[6], kernel.packet[7]);
+  __m256 S0 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(1,0,1,0));
+  __m256 S1 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(3,2,3,2));
+  __m256 S2 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(1,0,1,0));
+  __m256 S3 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(3,2,3,2));
+  __m256 S4 = _mm256_shuffle_ps(T4,T6,_MM_SHUFFLE(1,0,1,0));
+  __m256 S5 = _mm256_shuffle_ps(T4,T6,_MM_SHUFFLE(3,2,3,2));
+  __m256 S6 = _mm256_shuffle_ps(T5,T7,_MM_SHUFFLE(1,0,1,0));
+  __m256 S7 = _mm256_shuffle_ps(T5,T7,_MM_SHUFFLE(3,2,3,2));
+  kernel.packet[0] = _mm256_permute2f128_ps(S0, S4, 0x20);
+  kernel.packet[1] = _mm256_permute2f128_ps(S1, S5, 0x20);
+  kernel.packet[2] = _mm256_permute2f128_ps(S2, S6, 0x20);
+  kernel.packet[3] = _mm256_permute2f128_ps(S3, S7, 0x20);
+  kernel.packet[4] = _mm256_permute2f128_ps(S0, S4, 0x31);
+  kernel.packet[5] = _mm256_permute2f128_ps(S1, S5, 0x31);
+  kernel.packet[6] = _mm256_permute2f128_ps(S2, S6, 0x31);
+  kernel.packet[7] = _mm256_permute2f128_ps(S3, S7, 0x31);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet8f,4>& kernel) {
+  __m256 T0 = _mm256_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T1 = _mm256_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m256 T2 = _mm256_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m256 T3 = _mm256_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+
+  __m256 S0 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(1,0,1,0));
+  __m256 S1 = _mm256_shuffle_ps(T0,T2,_MM_SHUFFLE(3,2,3,2));
+  __m256 S2 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(1,0,1,0));
+  __m256 S3 = _mm256_shuffle_ps(T1,T3,_MM_SHUFFLE(3,2,3,2));
+
+  kernel.packet[0] = _mm256_permute2f128_ps(S0, S1, 0x20);
+  kernel.packet[1] = _mm256_permute2f128_ps(S2, S3, 0x20);
+  kernel.packet[2] = _mm256_permute2f128_ps(S0, S1, 0x31);
+  kernel.packet[3] = _mm256_permute2f128_ps(S2, S3, 0x31);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4d,4>& kernel) {
+  __m256d T0 = _mm256_shuffle_pd(kernel.packet[0], kernel.packet[1], 15);
+  __m256d T1 = _mm256_shuffle_pd(kernel.packet[0], kernel.packet[1], 0);
+  __m256d T2 = _mm256_shuffle_pd(kernel.packet[2], kernel.packet[3], 15);
+  __m256d T3 = _mm256_shuffle_pd(kernel.packet[2], kernel.packet[3], 0);
+
+  kernel.packet[1] = _mm256_permute2f128_pd(T0, T2, 32);
+  kernel.packet[3] = _mm256_permute2f128_pd(T0, T2, 49);
+  kernel.packet[0] = _mm256_permute2f128_pd(T1, T3, 32);
+  kernel.packet[2] = _mm256_permute2f128_pd(T1, T3, 49);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pblend(const Selector<8>& ifPacket, const Packet8f& thenPacket, const Packet8f& elsePacket) {
+  const __m256 zero = _mm256_setzero_ps();
+  const __m256 select = _mm256_set_ps(ifPacket.select[7], ifPacket.select[6], ifPacket.select[5], ifPacket.select[4], ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m256 false_mask = _mm256_cmp_ps(select, zero, _CMP_EQ_UQ);
+  return _mm256_blendv_ps(thenPacket, elsePacket, false_mask);
+}
+template<> EIGEN_STRONG_INLINE Packet4d pblend(const Selector<4>& ifPacket, const Packet4d& thenPacket, const Packet4d& elsePacket) {
+  const __m256d zero = _mm256_setzero_pd();
+  const __m256d select = _mm256_set_pd(ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m256d false_mask = _mm256_cmp_pd(select, zero, _CMP_EQ_UQ);
+  return _mm256_blendv_pd(thenPacket, elsePacket, false_mask);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pinsertfirst(const Packet8f& a, float b)
+{
+  return _mm256_blend_ps(a,pset1<Packet8f>(b),1);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4d pinsertfirst(const Packet4d& a, double b)
+{
+  return _mm256_blend_pd(a,pset1<Packet4d>(b),1);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pinsertlast(const Packet8f& a, float b)
+{
+  return _mm256_blend_ps(a,pset1<Packet8f>(b),(1<<7));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4d pinsertlast(const Packet4d& a, double b)
+{
+  return _mm256_blend_pd(a,pset1<Packet4d>(b),(1<<3));
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PACKET_MATH_AVX_H
diff --git a/Eigen/src/Core/arch/AVX/TypeCasting.h b/Eigen/src/Core/arch/AVX/TypeCasting.h
new file mode 100644
index 0000000..83bfdc6
--- /dev/null
+++ b/Eigen/src/Core/arch/AVX/TypeCasting.h
@@ -0,0 +1,51 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TYPE_CASTING_AVX_H
+#define EIGEN_TYPE_CASTING_AVX_H
+
+namespace Eigen {
+
+namespace internal {
+
+// For now we use SSE to handle integers, so we can't use AVX instructions to cast
+// from int to float
+template <>
+struct type_casting_traits<float, int> {
+  enum {
+    VectorizedCast = 0,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<int, float> {
+  enum {
+    VectorizedCast = 0,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+
+
+template<> EIGEN_STRONG_INLINE Packet8i pcast<Packet8f, Packet8i>(const Packet8f& a) {
+  return _mm256_cvtps_epi32(a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8i, Packet8f>(const Packet8i& a) {
+  return _mm256_cvtepi32_ps(a);
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TYPE_CASTING_AVX_H
diff --git a/Eigen/src/Core/arch/AVX512/MathFunctions.h b/Eigen/src/Core/arch/AVX512/MathFunctions.h
new file mode 100644
index 0000000..9c1717f
--- /dev/null
+++ b/Eigen/src/Core/arch/AVX512/MathFunctions.h
@@ -0,0 +1,391 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Pedro Gonnet (pedro.gonnet@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
+#define THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
+
+namespace Eigen {
+
+namespace internal {
+
+// Disable the code for older versions of gcc that don't support many of the required avx512 instrinsics.
+#if EIGEN_GNUC_AT_LEAST(5, 3)
+
+#define _EIGEN_DECLARE_CONST_Packet16f(NAME, X) \
+  const Packet16f p16f_##NAME = pset1<Packet16f>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(NAME, X) \
+  const Packet16f p16f_##NAME = (__m512)pset1<Packet16i>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet8d(NAME, X) \
+  const Packet8d p8d_##NAME = pset1<Packet8d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(NAME, X) \
+  const Packet8d p8d_##NAME = _mm512_castsi512_pd(_mm512_set1_epi64(X))
+
+// Natural logarithm
+// Computes log(x) as log(2^e * m) = C*e + log(m), where the constant C =log(2)
+// and m is in the range [sqrt(1/2),sqrt(2)). In this range, the logarithm can
+// be easily approximated by a polynomial centered on m=1 for stability.
+#if defined(EIGEN_VECTORIZE_AVX512DQ)
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+plog<Packet16f>(const Packet16f& _x) {
+  Packet16f x = _x;
+  _EIGEN_DECLARE_CONST_Packet16f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet16f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(126f, 126.0f);
+
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(inv_mant_mask, ~0x7f800000);
+
+  // The smallest non denormalized float number.
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(min_norm_pos, 0x00800000);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(minus_inf, 0xff800000);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(nan, 0x7fc00000);
+
+  // Polynomial coefficients.
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_SQRTHF, 0.707106781186547524f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p0, 7.0376836292E-2f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p1, -1.1514610310E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p2, 1.1676998740E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p3, -1.2420140846E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p4, +1.4249322787E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p5, -1.6668057665E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p6, +2.0000714765E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p7, -2.4999993993E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_p8, +3.3333331174E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_q1, -2.12194440e-4f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_log_q2, 0.693359375f);
+
+  // invalid_mask is set to true when x is NaN
+  __mmask16 invalid_mask =
+      _mm512_cmp_ps_mask(x, _mm512_setzero_ps(), _CMP_NGE_UQ);
+  __mmask16 iszero_mask =
+      _mm512_cmp_ps_mask(x, _mm512_setzero_ps(), _CMP_EQ_UQ);
+
+  // Truncate input values to the minimum positive normal.
+  x = pmax(x, p16f_min_norm_pos);
+
+  // Extract the shifted exponents.
+  Packet16f emm0 = _mm512_cvtepi32_ps(_mm512_srli_epi32((__m512i)x, 23));
+  Packet16f e = _mm512_sub_ps(emm0, p16f_126f);
+
+  // Set the exponents to -1, i.e. x are in the range [0.5,1).
+  x = _mm512_and_ps(x, p16f_inv_mant_mask);
+  x = _mm512_or_ps(x, p16f_half);
+
+  // part2: Shift the inputs from the range [0.5,1) to [sqrt(1/2),sqrt(2))
+  // and shift by -1. The values are then centered around 0, which improves
+  // the stability of the polynomial evaluation.
+  //   if( x < SQRTHF ) {
+  //     e -= 1;
+  //     x = x + x - 1.0;
+  //   } else { x = x - 1.0; }
+  __mmask16 mask = _mm512_cmp_ps_mask(x, p16f_cephes_SQRTHF, _CMP_LT_OQ);
+  Packet16f tmp = _mm512_mask_blend_ps(mask, _mm512_setzero_ps(), x);
+  x = psub(x, p16f_1);
+  e = psub(e, _mm512_mask_blend_ps(mask, _mm512_setzero_ps(), p16f_1));
+  x = padd(x, tmp);
+
+  Packet16f x2 = pmul(x, x);
+  Packet16f x3 = pmul(x2, x);
+
+  // Evaluate the polynomial approximant of degree 8 in three parts, probably
+  // to improve instruction-level parallelism.
+  Packet16f y, y1, y2;
+  y = pmadd(p16f_cephes_log_p0, x, p16f_cephes_log_p1);
+  y1 = pmadd(p16f_cephes_log_p3, x, p16f_cephes_log_p4);
+  y2 = pmadd(p16f_cephes_log_p6, x, p16f_cephes_log_p7);
+  y = pmadd(y, x, p16f_cephes_log_p2);
+  y1 = pmadd(y1, x, p16f_cephes_log_p5);
+  y2 = pmadd(y2, x, p16f_cephes_log_p8);
+  y = pmadd(y, x3, y1);
+  y = pmadd(y, x3, y2);
+  y = pmul(y, x3);
+
+  // Add the logarithm of the exponent back to the result of the interpolation.
+  y1 = pmul(e, p16f_cephes_log_q1);
+  tmp = pmul(x2, p16f_half);
+  y = padd(y, y1);
+  x = psub(x, tmp);
+  y2 = pmul(e, p16f_cephes_log_q2);
+  x = padd(x, y);
+  x = padd(x, y2);
+
+  // Filter out invalid inputs, i.e. negative arg will be NAN, 0 will be -INF.
+  return _mm512_mask_blend_ps(iszero_mask,
+                              _mm512_mask_blend_ps(invalid_mask, x, p16f_nan),
+                              p16f_minus_inf);
+}
+#endif
+
+// Exponential function. Works by writing "x = m*log(2) + r" where
+// "m = floor(x/log(2)+1/2)" and "r" is the remainder. The result is then
+// "exp(x) = 2^m*exp(r)" where exp(r) is in the range [-1,1).
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+pexp<Packet16f>(const Packet16f& _x) {
+  _EIGEN_DECLARE_CONST_Packet16f(1, 1.0f);
+  _EIGEN_DECLARE_CONST_Packet16f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(127, 127.0f);
+
+  _EIGEN_DECLARE_CONST_Packet16f(exp_hi, 88.3762626647950f);
+  _EIGEN_DECLARE_CONST_Packet16f(exp_lo, -88.3762626647949f);
+
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_LOG2EF, 1.44269504088896341f);
+
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p0, 1.9875691500E-4f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p1, 1.3981999507E-3f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p2, 8.3334519073E-3f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p3, 4.1665795894E-2f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p4, 1.6666665459E-1f);
+  _EIGEN_DECLARE_CONST_Packet16f(cephes_exp_p5, 5.0000001201E-1f);
+
+  // Clamp x.
+  Packet16f x = pmax(pmin(_x, p16f_exp_hi), p16f_exp_lo);
+
+  // Express exp(x) as exp(m*ln(2) + r), start by extracting
+  // m = floor(x/ln(2) + 0.5).
+  Packet16f m = _mm512_floor_ps(pmadd(x, p16f_cephes_LOG2EF, p16f_half));
+
+  // Get r = x - m*ln(2). Note that we can do this without losing more than one
+  // ulp precision due to the FMA instruction.
+  _EIGEN_DECLARE_CONST_Packet16f(nln2, -0.6931471805599453f);
+  Packet16f r = _mm512_fmadd_ps(m, p16f_nln2, x);
+  Packet16f r2 = pmul(r, r);
+
+  // TODO(gonnet): Split into odd/even polynomials and try to exploit
+  //               instruction-level parallelism.
+  Packet16f y = p16f_cephes_exp_p0;
+  y = pmadd(y, r, p16f_cephes_exp_p1);
+  y = pmadd(y, r, p16f_cephes_exp_p2);
+  y = pmadd(y, r, p16f_cephes_exp_p3);
+  y = pmadd(y, r, p16f_cephes_exp_p4);
+  y = pmadd(y, r, p16f_cephes_exp_p5);
+  y = pmadd(y, r2, r);
+  y = padd(y, p16f_1);
+
+  // Build emm0 = 2^m.
+  Packet16i emm0 = _mm512_cvttps_epi32(padd(m, p16f_127));
+  emm0 = _mm512_slli_epi32(emm0, 23);
+
+  // Return 2^m * exp(r).
+  return pmax(pmul(y, _mm512_castsi512_ps(emm0)), _x);
+}
+
+/*template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
+pexp<Packet8d>(const Packet8d& _x) {
+  Packet8d x = _x;
+
+  _EIGEN_DECLARE_CONST_Packet8d(1, 1.0);
+  _EIGEN_DECLARE_CONST_Packet8d(2, 2.0);
+
+  _EIGEN_DECLARE_CONST_Packet8d(exp_hi, 709.437);
+  _EIGEN_DECLARE_CONST_Packet8d(exp_lo, -709.436139303);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_LOG2EF, 1.4426950408889634073599);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_p0, 1.26177193074810590878e-4);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_p1, 3.02994407707441961300e-2);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q0, 3.00198505138664455042e-6);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q1, 2.52448340349684104192e-3);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q2, 2.27265548208155028766e-1);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_q3, 2.00000000000000000009e0);
+
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_C1, 0.693145751953125);
+  _EIGEN_DECLARE_CONST_Packet8d(cephes_exp_C2, 1.42860682030941723212e-6);
+
+  // clamp x
+  x = pmax(pmin(x, p8d_exp_hi), p8d_exp_lo);
+
+  // Express exp(x) as exp(g + n*log(2)).
+  const Packet8d n =
+      _mm512_mul_round_pd(p8d_cephes_LOG2EF, x, _MM_FROUND_TO_NEAREST_INT);
+
+  // Get the remainder modulo log(2), i.e. the "g" described above. Subtract
+  // n*log(2) out in two steps, i.e. n*C1 + n*C2, C1+C2=log2 to get the last
+  // digits right.
+  const Packet8d nC1 = pmul(n, p8d_cephes_exp_C1);
+  const Packet8d nC2 = pmul(n, p8d_cephes_exp_C2);
+  x = psub(x, nC1);
+  x = psub(x, nC2);
+
+  const Packet8d x2 = pmul(x, x);
+
+  // Evaluate the numerator polynomial of the rational interpolant.
+  Packet8d px = p8d_cephes_exp_p0;
+  px = pmadd(px, x2, p8d_cephes_exp_p1);
+  px = pmadd(px, x2, p8d_cephes_exp_p2);
+  px = pmul(px, x);
+
+  // Evaluate the denominator polynomial of the rational interpolant.
+  Packet8d qx = p8d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p8d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p8d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p8d_cephes_exp_q3);
+
+  // I don't really get this bit, copied from the SSE2 routines, so...
+  // TODO(gonnet): Figure out what is going on here, perhaps find a better
+  // rational interpolant?
+  x = _mm512_div_pd(px, psub(qx, px));
+  x = pmadd(p8d_2, x, p8d_1);
+
+  // Build e=2^n.
+  const Packet8d e = _mm512_castsi512_pd(_mm512_slli_epi64(
+      _mm512_add_epi64(_mm512_cvtpd_epi64(n), _mm512_set1_epi64(1023)), 52));
+
+  // Construct the result 2^n * exp(g) = e * x. The max is used to catch
+  // non-finite values in the input.
+  return pmax(pmul(x, e), _x);
+  }*/
+
+// Functions for sqrt.
+// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
+// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
+// exact solution. The main advantage of this approach is not just speed, but
+// also the fact that it can be inlined and pipelined with other computations,
+// further reducing its effective latency.
+#if EIGEN_FAST_MATH
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+psqrt<Packet16f>(const Packet16f& _x) {
+  _EIGEN_DECLARE_CONST_Packet16f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(flt_min, 0x00800000);
+
+  Packet16f neg_half = pmul(_x, p16f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask16 non_zero_mask = _mm512_cmp_ps_mask(_x, p16f_flt_min, _CMP_GE_OQ);
+  Packet16f x = _mm512_mask_blend_ps(non_zero_mask, _mm512_setzero_ps(), _mm512_rsqrt14_ps(_x));
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p16f_one_point_five));
+
+  // Multiply the original _x by it's reciprocal square root to extract the
+  // square root.
+  return pmul(_x, x);
+}
+
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
+psqrt<Packet8d>(const Packet8d& _x) {
+  _EIGEN_DECLARE_CONST_Packet8d(one_point_five, 1.5);
+  _EIGEN_DECLARE_CONST_Packet8d(minus_half, -0.5);
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(dbl_min, 0x0010000000000000LL);
+
+  Packet8d neg_half = pmul(_x, p8d_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask8 non_zero_mask = _mm512_cmp_pd_mask(_x, p8d_dbl_min, _CMP_GE_OQ);
+  Packet8d x = _mm512_mask_blend_pd(non_zero_mask, _mm512_setzero_pd(), _mm512_rsqrt14_pd(_x));
+
+  // Do a first step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Do a second step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Multiply the original _x by it's reciprocal square root to extract the
+  // square root.
+  return pmul(_x, x);
+}
+#else
+template <>
+EIGEN_STRONG_INLINE Packet16f psqrt<Packet16f>(const Packet16f& x) {
+  return _mm512_sqrt_ps(x);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d psqrt<Packet8d>(const Packet8d& x) {
+  return _mm512_sqrt_pd(x);
+}
+#endif
+
+// Functions for rsqrt.
+// Almost identical to the sqrt routine, just leave out the last multiplication
+// and fill in NaN/Inf where needed. Note that this function only exists as an
+// iterative version for doubles since there is no instruction for diretly
+// computing the reciprocal square root in AVX-512.
+#ifdef EIGEN_FAST_MATH
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet16f
+prsqrt<Packet16f>(const Packet16f& _x) {
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(inf, 0x7f800000);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(nan, 0x7fc00000);
+  _EIGEN_DECLARE_CONST_Packet16f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet16f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet16f_FROM_INT(flt_min, 0x00800000);
+
+  Packet16f neg_half = pmul(_x, p16f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask16 le_zero_mask = _mm512_cmp_ps_mask(_x, p16f_flt_min, _CMP_LT_OQ);
+  Packet16f x = _mm512_mask_blend_ps(le_zero_mask, _mm512_rsqrt14_ps(_x), _mm512_setzero_ps());
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  __mmask16 neg_mask = _mm512_cmp_ps_mask(_x, _mm512_setzero_ps(), _CMP_LT_OQ);
+  Packet16f infs_and_nans = _mm512_mask_blend_ps(
+      neg_mask, _mm512_mask_blend_ps(le_zero_mask, _mm512_setzero_ps(), p16f_inf), p16f_nan);
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p16f_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm512_mask_blend_ps(le_zero_mask, x, infs_and_nans);
+}
+
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet8d
+prsqrt<Packet8d>(const Packet8d& _x) {
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(inf, 0x7ff0000000000000LL);
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(nan, 0x7ff1000000000000LL);
+  _EIGEN_DECLARE_CONST_Packet8d(one_point_five, 1.5);
+  _EIGEN_DECLARE_CONST_Packet8d(minus_half, -0.5);
+  _EIGEN_DECLARE_CONST_Packet8d_FROM_INT64(dbl_min, 0x0010000000000000LL);
+
+  Packet8d neg_half = pmul(_x, p8d_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  __mmask8 le_zero_mask = _mm512_cmp_pd_mask(_x, p8d_dbl_min, _CMP_LT_OQ);
+  Packet8d x = _mm512_mask_blend_pd(le_zero_mask, _mm512_rsqrt14_pd(_x), _mm512_setzero_pd());
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  __mmask8 neg_mask = _mm512_cmp_pd_mask(_x, _mm512_setzero_pd(), _CMP_LT_OQ);
+  Packet8d infs_and_nans = _mm512_mask_blend_pd(
+      neg_mask, _mm512_mask_blend_pd(le_zero_mask, _mm512_setzero_pd(), p8d_inf), p8d_nan);
+
+  // Do a first step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Do a second step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p8d_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm512_mask_blend_pd(le_zero_mask, x, infs_and_nans);
+}
+#elif defined(EIGEN_VECTORIZE_AVX512ER)
+template <>
+EIGEN_STRONG_INLINE Packet16f prsqrt<Packet16f>(const Packet16f& x) {
+  return _mm512_rsqrt28_ps(x);
+}
+#endif
+#endif
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // THIRD_PARTY_EIGEN3_EIGEN_SRC_CORE_ARCH_AVX512_MATHFUNCTIONS_H_
diff --git a/Eigen/src/Core/arch/AVX512/PacketMath.h b/Eigen/src/Core/arch/AVX512/PacketMath.h
new file mode 100644
index 0000000..5adddc7
--- /dev/null
+++ b/Eigen/src/Core/arch/AVX512/PacketMath.h
@@ -0,0 +1,1316 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Benoit Steiner (benoit.steiner.goog@gmail.com)
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_AVX512_H
+#define EIGEN_PACKET_MATH_AVX512_H
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8
+#endif
+
+#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS (2*sizeof(void*))
+#endif
+
+#ifdef __FMA__
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+#endif
+
+typedef __m512 Packet16f;
+typedef __m512i Packet16i;
+typedef __m512d Packet8d;
+
+template <>
+struct is_arithmetic<__m512> {
+  enum { value = true };
+};
+template <>
+struct is_arithmetic<__m512i> {
+  enum { value = true };
+};
+template <>
+struct is_arithmetic<__m512d> {
+  enum { value = true };
+};
+
+template<> struct packet_traits<float>  : default_packet_traits
+{
+  typedef Packet16f type;
+  typedef Packet8f half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 16,
+    HasHalfPacket = 1,
+#if EIGEN_GNUC_AT_LEAST(5, 3)
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+    HasLog = 1,
+#endif
+    HasExp = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+#endif
+    HasDiv = 1
+  };
+ };
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet8d type;
+  typedef Packet4d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 8,
+    HasHalfPacket = 1,
+#if EIGEN_GNUC_AT_LEAST(5, 3)
+    HasSqrt = 1,
+    HasRsqrt = EIGEN_FAST_MATH,
+#endif
+    HasDiv = 1
+  };
+};
+
+/* TODO Implement AVX512 for integers
+template<> struct packet_traits<int>    : default_packet_traits
+{
+  typedef Packet16i type;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=8
+  };
+};
+*/
+
+template <>
+struct unpacket_traits<Packet16f> {
+  typedef float type;
+  typedef Packet8f half;
+  enum { size = 16, alignment=Aligned64 };
+};
+template <>
+struct unpacket_traits<Packet8d> {
+  typedef double type;
+  typedef Packet4d half;
+  enum { size = 8, alignment=Aligned64 };
+};
+template <>
+struct unpacket_traits<Packet16i> {
+  typedef int type;
+  typedef Packet8i half;
+  enum { size = 16, alignment=Aligned64 };
+};
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pset1<Packet16f>(const float& from) {
+  return _mm512_set1_ps(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pset1<Packet8d>(const double& from) {
+  return _mm512_set1_pd(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i pset1<Packet16i>(const int& from) {
+  return _mm512_set1_epi32(from);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pload1<Packet16f>(const float* from) {
+  return _mm512_broadcastss_ps(_mm_load_ps1(from));
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pload1<Packet8d>(const double* from) {
+  return _mm512_broadcastsd_pd(_mm_load_pd1(from));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f plset<Packet16f>(const float& a) {
+  return _mm512_add_ps(
+      _mm512_set1_ps(a),
+      _mm512_set_ps(15.0f, 14.0f, 13.0f, 12.0f, 11.0f, 10.0f, 9.0f, 8.0f, 7.0f, 6.0f, 5.0f,
+                    4.0f, 3.0f, 2.0f, 1.0f, 0.0f));
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d plset<Packet8d>(const double& a) {
+  return _mm512_add_pd(_mm512_set1_pd(a),
+                       _mm512_set_pd(7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0, 0.0));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f padd<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_add_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d padd<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_add_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f psub<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_sub_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d psub<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_sub_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pnegate(const Packet16f& a) {
+  return _mm512_sub_ps(_mm512_set1_ps(0.0), a);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pnegate(const Packet8d& a) {
+  return _mm512_sub_pd(_mm512_set1_pd(0.0), a);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pconj(const Packet16f& a) {
+  return a;
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pconj(const Packet8d& a) {
+  return a;
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i pconj(const Packet16i& a) {
+  return a;
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pmul<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_mul_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmul<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_mul_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pdiv<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_div_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pdiv<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_div_pd(a, b);
+}
+
+#ifdef __FMA__
+template <>
+EIGEN_STRONG_INLINE Packet16f pmadd(const Packet16f& a, const Packet16f& b,
+                                    const Packet16f& c) {
+  return _mm512_fmadd_ps(a, b, c);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmadd(const Packet8d& a, const Packet8d& b,
+                                   const Packet8d& c) {
+  return _mm512_fmadd_pd(a, b, c);
+}
+#endif
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pmin<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_min_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmin<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_min_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pmax<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+  return _mm512_max_ps(a, b);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pmax<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+  return _mm512_max_pd(a, b);
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pand<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_and_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_and_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pand<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_and_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_and_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_and_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet16f por<Packet16f>(const Packet16f& a,
+                                             const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_or_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_or_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet8d por<Packet8d>(const Packet8d& a,
+                                           const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_or_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_or_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_or_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pxor<Packet16f>(const Packet16f& a,
+                                              const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_xor_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_xor_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pxor<Packet8d>(const Packet8d& a,
+                                            const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_xor_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_xor_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_xor_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pandnot<Packet16f>(const Packet16f& a,
+                                                 const Packet16f& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_andnot_ps(a, b);
+#else
+  Packet16f res = _mm512_undefined_ps();
+  Packet4f lane0_a = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane0_b = _mm512_extractf32x4_ps(b, 0);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane0_a, lane0_b), 0);
+
+  Packet4f lane1_a = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane1_b = _mm512_extractf32x4_ps(b, 1);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane1_a, lane1_b), 1);
+
+  Packet4f lane2_a = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane2_b = _mm512_extractf32x4_ps(b, 2);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane2_a, lane2_b), 2);
+
+  Packet4f lane3_a = _mm512_extractf32x4_ps(a, 3);
+  Packet4f lane3_b = _mm512_extractf32x4_ps(b, 3);
+  res = _mm512_insertf32x4(res, _mm_andnot_ps(lane3_a, lane3_b), 3);
+
+  return res;
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pandnot<Packet8d>(const Packet8d& a,
+                                               const Packet8d& b) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  return _mm512_andnot_pd(a, b);
+#else
+  Packet8d res = _mm512_undefined_pd();
+  Packet4d lane0_a = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane0_b = _mm512_extractf64x4_pd(b, 0);
+  res = _mm512_insertf64x4(res, _mm256_andnot_pd(lane0_a, lane0_b), 0);
+
+  Packet4d lane1_a = _mm512_extractf64x4_pd(a, 1);
+  Packet4d lane1_b = _mm512_extractf64x4_pd(b, 1);
+  res = _mm512_insertf64x4(res, _mm256_andnot_pd(lane1_a, lane1_b), 1);
+
+  return res;
+#endif
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f pload<Packet16f>(const float* from) {
+  EIGEN_DEBUG_ALIGNED_LOAD return _mm512_load_ps(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pload<Packet8d>(const double* from) {
+  EIGEN_DEBUG_ALIGNED_LOAD return _mm512_load_pd(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i pload<Packet16i>(const int* from) {
+  EIGEN_DEBUG_ALIGNED_LOAD return _mm512_load_si512(
+      reinterpret_cast<const __m512i*>(from));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet16f ploadu<Packet16f>(const float* from) {
+  EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_ps(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d ploadu<Packet8d>(const double* from) {
+  EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_pd(from);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16i ploadu<Packet16i>(const int* from) {
+  EIGEN_DEBUG_UNALIGNED_LOAD return _mm512_loadu_si512(
+      reinterpret_cast<const __m512i*>(from));
+}
+
+// Loads 8 floats from memory a returns the packet
+// {a0, a0  a1, a1, a2, a2, a3, a3, a4, a4, a5, a5, a6, a6, a7, a7}
+template <>
+EIGEN_STRONG_INLINE Packet16f ploaddup<Packet16f>(const float* from) {
+  Packet8f lane0 = _mm256_broadcast_ps((const __m128*)(const void*)from);
+  // mimic an "inplace" permutation of the lower 128bits using a blend
+  lane0 = _mm256_blend_ps(
+      lane0, _mm256_castps128_ps256(_mm_permute_ps(
+                 _mm256_castps256_ps128(lane0), _MM_SHUFFLE(1, 0, 1, 0))),
+      15);
+  // then we can perform a consistent permutation on the global register to get
+  // everything in shape:
+  lane0 = _mm256_permute_ps(lane0, _MM_SHUFFLE(3, 3, 2, 2));
+
+  Packet8f lane1 = _mm256_broadcast_ps((const __m128*)(const void*)(from + 4));
+  // mimic an "inplace" permutation of the lower 128bits using a blend
+  lane1 = _mm256_blend_ps(
+      lane1, _mm256_castps128_ps256(_mm_permute_ps(
+                 _mm256_castps256_ps128(lane1), _MM_SHUFFLE(1, 0, 1, 0))),
+      15);
+  // then we can perform a consistent permutation on the global register to get
+  // everything in shape:
+  lane1 = _mm256_permute_ps(lane1, _MM_SHUFFLE(3, 3, 2, 2));
+
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  Packet16f res = _mm512_undefined_ps();
+  return _mm512_insertf32x8(res, lane0, 0);
+  return _mm512_insertf32x8(res, lane1, 1);
+  return res;
+#else
+  Packet16f res = _mm512_undefined_ps();
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane0, 0), 0);
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane0, 1), 1);
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane1, 0), 2);
+  res = _mm512_insertf32x4(res, _mm256_extractf128_ps(lane1, 1), 3);
+  return res;
+#endif
+}
+// Loads 4 doubles from memory a returns the packet {a0, a0  a1, a1, a2, a2, a3,
+// a3}
+template <>
+EIGEN_STRONG_INLINE Packet8d ploaddup<Packet8d>(const double* from) {
+  Packet4d lane0 = _mm256_broadcast_pd((const __m128d*)(const void*)from);
+  lane0 = _mm256_permute_pd(lane0, 3 << 2);
+
+  Packet4d lane1 = _mm256_broadcast_pd((const __m128d*)(const void*)(from + 2));
+  lane1 = _mm256_permute_pd(lane1, 3 << 2);
+
+  Packet8d res = _mm512_undefined_pd();
+  res = _mm512_insertf64x4(res, lane0, 0);
+  return _mm512_insertf64x4(res, lane1, 1);
+}
+
+// Loads 4 floats from memory a returns the packet
+// {a0, a0  a0, a0, a1, a1, a1, a1, a2, a2, a2, a2, a3, a3, a3, a3}
+template <>
+EIGEN_STRONG_INLINE Packet16f ploadquad<Packet16f>(const float* from) {
+  Packet16f tmp = _mm512_undefined_ps();
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from), 0);
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 1), 1);
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 2), 2);
+  tmp = _mm512_insertf32x4(tmp, _mm_load_ps1(from + 3), 3);
+  return tmp;
+}
+// Loads 2 doubles from memory a returns the packet
+// {a0, a0  a0, a0, a1, a1, a1, a1}
+template <>
+EIGEN_STRONG_INLINE Packet8d ploadquad<Packet8d>(const double* from) {
+  Packet8d tmp = _mm512_undefined_pd();
+  Packet2d tmp0 = _mm_load_pd1(from);
+  Packet2d tmp1 = _mm_load_pd1(from + 1);
+  Packet4d lane0 = _mm256_broadcastsd_pd(tmp0);
+  Packet4d lane1 = _mm256_broadcastsd_pd(tmp1);
+  tmp = _mm512_insertf64x4(tmp, lane0, 0);
+  return _mm512_insertf64x4(tmp, lane1, 1);
+}
+
+template <>
+EIGEN_STRONG_INLINE void pstore<float>(float* to, const Packet16f& from) {
+  EIGEN_DEBUG_ALIGNED_STORE _mm512_store_ps(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet8d& from) {
+  EIGEN_DEBUG_ALIGNED_STORE _mm512_store_pd(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore<int>(int* to, const Packet16i& from) {
+  EIGEN_DEBUG_ALIGNED_STORE _mm512_storeu_si512(reinterpret_cast<__m512i*>(to),
+                                                from);
+}
+
+template <>
+EIGEN_STRONG_INLINE void pstoreu<float>(float* to, const Packet16f& from) {
+  EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_ps(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet8d& from) {
+  EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_pd(to, from);
+}
+template <>
+EIGEN_STRONG_INLINE void pstoreu<int>(int* to, const Packet16i& from) {
+  EIGEN_DEBUG_UNALIGNED_STORE _mm512_storeu_si512(
+      reinterpret_cast<__m512i*>(to), from);
+}
+
+template <>
+EIGEN_DEVICE_FUNC inline Packet16f pgather<float, Packet16f>(const float* from,
+                                                             Index stride) {
+  Packet16i stride_vector = _mm512_set1_epi32(stride);
+  Packet16i stride_multiplier =
+      _mm512_set_epi32(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+  Packet16i indices = _mm512_mullo_epi32(stride_vector, stride_multiplier);
+
+  return _mm512_i32gather_ps(indices, from, 4);
+}
+template <>
+EIGEN_DEVICE_FUNC inline Packet8d pgather<double, Packet8d>(const double* from,
+                                                            Index stride) {
+  Packet8i stride_vector = _mm256_set1_epi32(stride);
+  Packet8i stride_multiplier = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0);
+  Packet8i indices = _mm256_mullo_epi32(stride_vector, stride_multiplier);
+
+  return _mm512_i32gather_pd(indices, from, 8);
+}
+
+template <>
+EIGEN_DEVICE_FUNC inline void pscatter<float, Packet16f>(float* to,
+                                                         const Packet16f& from,
+                                                         Index stride) {
+  Packet16i stride_vector = _mm512_set1_epi32(stride);
+  Packet16i stride_multiplier =
+      _mm512_set_epi32(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0);
+  Packet16i indices = _mm512_mullo_epi32(stride_vector, stride_multiplier);
+  _mm512_i32scatter_ps(to, indices, from, 4);
+}
+template <>
+EIGEN_DEVICE_FUNC inline void pscatter<double, Packet8d>(double* to,
+                                                         const Packet8d& from,
+                                                         Index stride) {
+  Packet8i stride_vector = _mm256_set1_epi32(stride);
+  Packet8i stride_multiplier = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0);
+  Packet8i indices = _mm256_mullo_epi32(stride_vector, stride_multiplier);
+  _mm512_i32scatter_pd(to, indices, from, 8);
+}
+
+template <>
+EIGEN_STRONG_INLINE void pstore1<Packet16f>(float* to, const float& a) {
+  Packet16f pa = pset1<Packet16f>(a);
+  pstore(to, pa);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore1<Packet8d>(double* to, const double& a) {
+  Packet8d pa = pset1<Packet8d>(a);
+  pstore(to, pa);
+}
+template <>
+EIGEN_STRONG_INLINE void pstore1<Packet16i>(int* to, const int& a) {
+  Packet16i pa = pset1<Packet16i>(a);
+  pstore(to, pa);
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+
+template <>
+EIGEN_STRONG_INLINE float pfirst<Packet16f>(const Packet16f& a) {
+  return _mm_cvtss_f32(_mm512_extractf32x4_ps(a, 0));
+}
+template <>
+EIGEN_STRONG_INLINE double pfirst<Packet8d>(const Packet8d& a) {
+  return _mm_cvtsd_f64(_mm256_extractf128_pd(_mm512_extractf64x4_pd(a, 0), 0));
+}
+template <>
+EIGEN_STRONG_INLINE int pfirst<Packet16i>(const Packet16i& a) {
+  return _mm_extract_epi32(_mm512_extracti32x4_epi32(a, 0), 0);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16f preverse(const Packet16f& a)
+{
+  return _mm512_permutexvar_ps(_mm512_set_epi32(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15), a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8d preverse(const Packet8d& a)
+{
+  return _mm512_permutexvar_pd(_mm512_set_epi32(0, 0, 0, 1, 0, 2, 0, 3, 0, 4, 0, 5, 0, 6, 0, 7), a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16f pabs(const Packet16f& a)
+{
+  // _mm512_abs_ps intrinsic not found, so hack around it
+  return _mm512_castsi512_ps(_mm512_and_si512(_mm512_castps_si512(a), _mm512_set1_epi32(0x7fffffff)));
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pabs(const Packet8d& a) {
+  // _mm512_abs_ps intrinsic not found, so hack around it
+  return _mm512_castsi512_pd(_mm512_and_si512(_mm512_castpd_si512(a),
+                                   _mm512_set1_epi64(0x7fffffffffffffff)));
+}
+
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+// AVX512F does not define _mm512_extractf32x8_ps to extract _m256 from _m512
+#define EIGEN_EXTRACT_8f_FROM_16f(INPUT, OUTPUT)                           \
+  __m256 OUTPUT##_0 = _mm512_extractf32x8_ps(INPUT, 0) __m256 OUTPUT##_1 = \
+      _mm512_extractf32x8_ps(INPUT, 1)
+#else
+#define EIGEN_EXTRACT_8f_FROM_16f(INPUT, OUTPUT)                \
+  __m256 OUTPUT##_0 = _mm256_insertf128_ps(                     \
+      _mm256_castps128_ps256(_mm512_extractf32x4_ps(INPUT, 0)), \
+      _mm512_extractf32x4_ps(INPUT, 1), 1);                     \
+  __m256 OUTPUT##_1 = _mm256_insertf128_ps(                     \
+      _mm256_castps128_ps256(_mm512_extractf32x4_ps(INPUT, 2)), \
+      _mm512_extractf32x4_ps(INPUT, 3), 1);
+#endif
+
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+#define EIGEN_INSERT_8f_INTO_16f(OUTPUT, INPUTA, INPUTB) \
+  OUTPUT = _mm512_insertf32x8(OUTPUT, INPUTA, 0);        \
+  OUTPUT = _mm512_insertf32x8(OUTPUT, INPUTB, 1);
+#else
+#define EIGEN_INSERT_8f_INTO_16f(OUTPUT, INPUTA, INPUTB)                    \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTA, 0), 0); \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTA, 1), 1); \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTB, 0), 2); \
+  OUTPUT = _mm512_insertf32x4(OUTPUT, _mm256_extractf128_ps(INPUTB, 1), 3);
+#endif
+template<> EIGEN_STRONG_INLINE Packet16f preduxp<Packet16f>(const Packet16f*
+vecs)
+{
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[0], vecs0);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[1], vecs1);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[2], vecs2);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[3], vecs3);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[4], vecs4);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[5], vecs5);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[6], vecs6);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[7], vecs7);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[8], vecs8);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[9], vecs9);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[10], vecs10);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[11], vecs11);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[12], vecs12);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[13], vecs13);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[14], vecs14);
+  EIGEN_EXTRACT_8f_FROM_16f(vecs[15], vecs15);
+
+  __m256 hsum1 = _mm256_hadd_ps(vecs0_0, vecs1_0);
+  __m256 hsum2 = _mm256_hadd_ps(vecs2_0, vecs3_0);
+  __m256 hsum3 = _mm256_hadd_ps(vecs4_0, vecs5_0);
+  __m256 hsum4 = _mm256_hadd_ps(vecs6_0, vecs7_0);
+
+  __m256 hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  __m256 hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  __m256 hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  __m256 hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  __m256 perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  __m256 perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  __m256 perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  __m256 perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  __m256 sum1 = _mm256_add_ps(perm1, hsum5);
+  __m256 sum2 = _mm256_add_ps(perm2, hsum6);
+  __m256 sum3 = _mm256_add_ps(perm3, hsum7);
+  __m256 sum4 = _mm256_add_ps(perm4, hsum8);
+
+  __m256 blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  __m256 blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  __m256 final = _mm256_blend_ps(blend1, blend2, 0xf0);
+
+  hsum1 = _mm256_hadd_ps(vecs0_1, vecs1_1);
+  hsum2 = _mm256_hadd_ps(vecs2_1, vecs3_1);
+  hsum3 = _mm256_hadd_ps(vecs4_1, vecs5_1);
+  hsum4 = _mm256_hadd_ps(vecs6_1, vecs7_1);
+
+  hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  sum1 = _mm256_add_ps(perm1, hsum5);
+  sum2 = _mm256_add_ps(perm2, hsum6);
+  sum3 = _mm256_add_ps(perm3, hsum7);
+  sum4 = _mm256_add_ps(perm4, hsum8);
+
+  blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  final = padd(final, _mm256_blend_ps(blend1, blend2, 0xf0));
+
+  hsum1 = _mm256_hadd_ps(vecs8_0, vecs9_0);
+  hsum2 = _mm256_hadd_ps(vecs10_0, vecs11_0);
+  hsum3 = _mm256_hadd_ps(vecs12_0, vecs13_0);
+  hsum4 = _mm256_hadd_ps(vecs14_0, vecs15_0);
+
+  hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  sum1 = _mm256_add_ps(perm1, hsum5);
+  sum2 = _mm256_add_ps(perm2, hsum6);
+  sum3 = _mm256_add_ps(perm3, hsum7);
+  sum4 = _mm256_add_ps(perm4, hsum8);
+
+  blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  __m256 final_1 = _mm256_blend_ps(blend1, blend2, 0xf0);
+
+  hsum1 = _mm256_hadd_ps(vecs8_1, vecs9_1);
+  hsum2 = _mm256_hadd_ps(vecs10_1, vecs11_1);
+  hsum3 = _mm256_hadd_ps(vecs12_1, vecs13_1);
+  hsum4 = _mm256_hadd_ps(vecs14_1, vecs15_1);
+
+  hsum5 = _mm256_hadd_ps(hsum1, hsum1);
+  hsum6 = _mm256_hadd_ps(hsum2, hsum2);
+  hsum7 = _mm256_hadd_ps(hsum3, hsum3);
+  hsum8 = _mm256_hadd_ps(hsum4, hsum4);
+
+  perm1 = _mm256_permute2f128_ps(hsum5, hsum5, 0x23);
+  perm2 = _mm256_permute2f128_ps(hsum6, hsum6, 0x23);
+  perm3 = _mm256_permute2f128_ps(hsum7, hsum7, 0x23);
+  perm4 = _mm256_permute2f128_ps(hsum8, hsum8, 0x23);
+
+  sum1 = _mm256_add_ps(perm1, hsum5);
+  sum2 = _mm256_add_ps(perm2, hsum6);
+  sum3 = _mm256_add_ps(perm3, hsum7);
+  sum4 = _mm256_add_ps(perm4, hsum8);
+
+  blend1 = _mm256_blend_ps(sum1, sum2, 0xcc);
+  blend2 = _mm256_blend_ps(sum3, sum4, 0xcc);
+
+  final_1 = padd(final_1, _mm256_blend_ps(blend1, blend2, 0xf0));
+
+  __m512 final_output;
+
+  EIGEN_INSERT_8f_INTO_16f(final_output, final, final_1);
+  return final_output;
+}
+
+template<> EIGEN_STRONG_INLINE Packet8d preduxp<Packet8d>(const Packet8d* vecs)
+{
+  Packet4d vecs0_0 = _mm512_extractf64x4_pd(vecs[0], 0);
+  Packet4d vecs0_1 = _mm512_extractf64x4_pd(vecs[0], 1);
+
+  Packet4d vecs1_0 = _mm512_extractf64x4_pd(vecs[1], 0);
+  Packet4d vecs1_1 = _mm512_extractf64x4_pd(vecs[1], 1);
+
+  Packet4d vecs2_0 = _mm512_extractf64x4_pd(vecs[2], 0);
+  Packet4d vecs2_1 = _mm512_extractf64x4_pd(vecs[2], 1);
+
+  Packet4d vecs3_0 = _mm512_extractf64x4_pd(vecs[3], 0);
+  Packet4d vecs3_1 = _mm512_extractf64x4_pd(vecs[3], 1);
+
+  Packet4d vecs4_0 = _mm512_extractf64x4_pd(vecs[4], 0);
+  Packet4d vecs4_1 = _mm512_extractf64x4_pd(vecs[4], 1);
+
+  Packet4d vecs5_0 = _mm512_extractf64x4_pd(vecs[5], 0);
+  Packet4d vecs5_1 = _mm512_extractf64x4_pd(vecs[5], 1);
+
+  Packet4d vecs6_0 = _mm512_extractf64x4_pd(vecs[6], 0);
+  Packet4d vecs6_1 = _mm512_extractf64x4_pd(vecs[6], 1);
+
+  Packet4d vecs7_0 = _mm512_extractf64x4_pd(vecs[7], 0);
+  Packet4d vecs7_1 = _mm512_extractf64x4_pd(vecs[7], 1);
+
+  Packet4d tmp0, tmp1;
+
+  tmp0 = _mm256_hadd_pd(vecs0_0, vecs1_0);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs2_0, vecs3_0);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  __m256d final_0 = _mm256_blend_pd(tmp0, tmp1, 0xC);
+
+  tmp0 = _mm256_hadd_pd(vecs0_1, vecs1_1);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs2_1, vecs3_1);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  final_0 = padd(final_0, _mm256_blend_pd(tmp0, tmp1, 0xC));
+
+  tmp0 = _mm256_hadd_pd(vecs4_0, vecs5_0);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs6_0, vecs7_0);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  __m256d final_1 = _mm256_blend_pd(tmp0, tmp1, 0xC);
+
+  tmp0 = _mm256_hadd_pd(vecs4_1, vecs5_1);
+  tmp0 = _mm256_add_pd(tmp0, _mm256_permute2f128_pd(tmp0, tmp0, 1));
+
+  tmp1 = _mm256_hadd_pd(vecs6_1, vecs7_1);
+  tmp1 = _mm256_add_pd(tmp1, _mm256_permute2f128_pd(tmp1, tmp1, 1));
+
+  final_1 = padd(final_1, _mm256_blend_pd(tmp0, tmp1, 0xC));
+
+  __m512d final_output = _mm512_insertf64x4(final_output, final_0, 0);
+
+  return _mm512_insertf64x4(final_output, final_1, 1);
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux<Packet16f>(const Packet16f& a) {
+  //#ifdef EIGEN_VECTORIZE_AVX512DQ
+#if 0
+  Packet8f lane0 = _mm512_extractf32x8_ps(a, 0);
+  Packet8f lane1 = _mm512_extractf32x8_ps(a, 1);
+  Packet8f sum = padd(lane0, lane1);
+  Packet8f tmp0 = _mm256_hadd_ps(sum, _mm256_permute2f128_ps(a, a, 1));
+  tmp0 = _mm256_hadd_ps(tmp0, tmp0);
+  return pfirst(_mm256_hadd_ps(tmp0, tmp0));
+#else
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f sum = padd(padd(lane0, lane1), padd(lane2, lane3));
+  sum = _mm_hadd_ps(sum, sum);
+  sum = _mm_hadd_ps(sum, _mm_permute_ps(sum, 1));
+  return pfirst(sum);
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE double predux<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d sum = padd(lane0, lane1);
+  Packet4d tmp0 = _mm256_hadd_pd(sum, _mm256_permute2f128_pd(sum, sum, 1));
+  return pfirst(_mm256_hadd_pd(tmp0, tmp0));
+}
+
+template <>
+EIGEN_STRONG_INLINE Packet8f predux_downto4<Packet16f>(const Packet16f& a) {
+#ifdef EIGEN_VECTORIZE_AVX512DQ
+  Packet8f lane0 = _mm512_extractf32x8_ps(a, 0);
+  Packet8f lane1 = _mm512_extractf32x8_ps(a, 1);
+  return padd(lane0, lane1);
+#else
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f sum0 = padd(lane0, lane2);
+  Packet4f sum1 = padd(lane1, lane3);
+  return _mm256_insertf128_ps(_mm256_castps128_ps256(sum0), sum1, 1);
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE Packet4d predux_downto4<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = padd(lane0, lane1);
+  return res;
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux_mul<Packet16f>(const Packet16f& a) {
+//#ifdef EIGEN_VECTORIZE_AVX512DQ
+#if 0
+  Packet8f lane0 = _mm512_extractf32x8_ps(a, 0);
+  Packet8f lane1 = _mm512_extractf32x8_ps(a, 1);
+  Packet8f res = pmul(lane0, lane1);
+  res = pmul(res, _mm256_permute2f128_ps(res, res, 1));
+  res = pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+#else
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f res = pmul(pmul(lane0, lane1), pmul(lane2, lane3));
+  res = pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(pmul(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+#endif
+}
+template <>
+EIGEN_STRONG_INLINE double predux_mul<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = pmul(lane0, lane1);
+  res = pmul(res, _mm256_permute2f128_pd(res, res, 1));
+  return pfirst(pmul(res, _mm256_shuffle_pd(res, res, 1)));
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux_min<Packet16f>(const Packet16f& a) {
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f res = _mm_min_ps(_mm_min_ps(lane0, lane1), _mm_min_ps(lane2, lane3));
+  res = _mm_min_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(_mm_min_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+}
+template <>
+EIGEN_STRONG_INLINE double predux_min<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = _mm256_min_pd(lane0, lane1);
+  res = _mm256_min_pd(res, _mm256_permute2f128_pd(res, res, 1));
+  return pfirst(_mm256_min_pd(res, _mm256_shuffle_pd(res, res, 1)));
+}
+
+template <>
+EIGEN_STRONG_INLINE float predux_max<Packet16f>(const Packet16f& a) {
+  Packet4f lane0 = _mm512_extractf32x4_ps(a, 0);
+  Packet4f lane1 = _mm512_extractf32x4_ps(a, 1);
+  Packet4f lane2 = _mm512_extractf32x4_ps(a, 2);
+  Packet4f lane3 = _mm512_extractf32x4_ps(a, 3);
+  Packet4f res = _mm_max_ps(_mm_max_ps(lane0, lane1), _mm_max_ps(lane2, lane3));
+  res = _mm_max_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 3, 2)));
+  return pfirst(_mm_max_ps(res, _mm_permute_ps(res, _MM_SHUFFLE(0, 0, 0, 1))));
+}
+template <>
+EIGEN_STRONG_INLINE double predux_max<Packet8d>(const Packet8d& a) {
+  Packet4d lane0 = _mm512_extractf64x4_pd(a, 0);
+  Packet4d lane1 = _mm512_extractf64x4_pd(a, 1);
+  Packet4d res = _mm256_max_pd(lane0, lane1);
+  res = _mm256_max_pd(res, _mm256_permute2f128_pd(res, res, 1));
+  return pfirst(_mm256_max_pd(res, _mm256_shuffle_pd(res, res, 1)));
+}
+
+template <int Offset>
+struct palign_impl<Offset, Packet16f> {
+  static EIGEN_STRONG_INLINE void run(Packet16f& first,
+                                      const Packet16f& second) {
+    if (Offset != 0) {
+      __m512i first_idx = _mm512_set_epi32(
+          Offset + 15, Offset + 14, Offset + 13, Offset + 12, Offset + 11,
+          Offset + 10, Offset + 9, Offset + 8, Offset + 7, Offset + 6,
+          Offset + 5, Offset + 4, Offset + 3, Offset + 2, Offset + 1, Offset);
+
+      __m512i second_idx =
+          _mm512_set_epi32(Offset - 1, Offset - 2, Offset - 3, Offset - 4,
+                           Offset - 5, Offset - 6, Offset - 7, Offset - 8,
+                           Offset - 9, Offset - 10, Offset - 11, Offset - 12,
+                           Offset - 13, Offset - 14, Offset - 15, Offset - 16);
+
+      unsigned short mask = 0xFFFF;
+      mask <<= (16 - Offset);
+
+      first = _mm512_permutexvar_ps(first_idx, first);
+      Packet16f tmp = _mm512_permutexvar_ps(second_idx, second);
+      first = _mm512_mask_blend_ps(mask, first, tmp);
+    }
+  }
+};
+template <int Offset>
+struct palign_impl<Offset, Packet8d> {
+  static EIGEN_STRONG_INLINE void run(Packet8d& first, const Packet8d& second) {
+    if (Offset != 0) {
+      __m512i first_idx = _mm512_set_epi32(
+          0, Offset + 7, 0, Offset + 6, 0, Offset + 5, 0, Offset + 4, 0,
+          Offset + 3, 0, Offset + 2, 0, Offset + 1, 0, Offset);
+
+      __m512i second_idx = _mm512_set_epi32(
+          0, Offset - 1, 0, Offset - 2, 0, Offset - 3, 0, Offset - 4, 0,
+          Offset - 5, 0, Offset - 6, 0, Offset - 7, 0, Offset - 8);
+
+      unsigned char mask = 0xFF;
+      mask <<= (8 - Offset);
+
+      first = _mm512_permutexvar_pd(first_idx, first);
+      Packet8d tmp = _mm512_permutexvar_pd(second_idx, second);
+      first = _mm512_mask_blend_pd(mask, first, tmp);
+    }
+  }
+};
+
+
+#define PACK_OUTPUT(OUTPUT, INPUT, INDEX, STRIDE) \
+  EIGEN_INSERT_8f_INTO_16f(OUTPUT[INDEX], INPUT[INDEX], INPUT[INDEX + STRIDE]);
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet16f, 16>& kernel) {
+  __m512 T0 = _mm512_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T1 = _mm512_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T2 = _mm512_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m512 T3 = _mm512_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+  __m512 T4 = _mm512_unpacklo_ps(kernel.packet[4], kernel.packet[5]);
+  __m512 T5 = _mm512_unpackhi_ps(kernel.packet[4], kernel.packet[5]);
+  __m512 T6 = _mm512_unpacklo_ps(kernel.packet[6], kernel.packet[7]);
+  __m512 T7 = _mm512_unpackhi_ps(kernel.packet[6], kernel.packet[7]);
+  __m512 T8 = _mm512_unpacklo_ps(kernel.packet[8], kernel.packet[9]);
+  __m512 T9 = _mm512_unpackhi_ps(kernel.packet[8], kernel.packet[9]);
+  __m512 T10 = _mm512_unpacklo_ps(kernel.packet[10], kernel.packet[11]);
+  __m512 T11 = _mm512_unpackhi_ps(kernel.packet[10], kernel.packet[11]);
+  __m512 T12 = _mm512_unpacklo_ps(kernel.packet[12], kernel.packet[13]);
+  __m512 T13 = _mm512_unpackhi_ps(kernel.packet[12], kernel.packet[13]);
+  __m512 T14 = _mm512_unpacklo_ps(kernel.packet[14], kernel.packet[15]);
+  __m512 T15 = _mm512_unpackhi_ps(kernel.packet[14], kernel.packet[15]);
+  __m512 S0 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S1 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S2 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S3 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S4 = _mm512_shuffle_ps(T4, T6, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S5 = _mm512_shuffle_ps(T4, T6, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S6 = _mm512_shuffle_ps(T5, T7, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S7 = _mm512_shuffle_ps(T5, T7, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S8 = _mm512_shuffle_ps(T8, T10, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S9 = _mm512_shuffle_ps(T8, T10, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S10 = _mm512_shuffle_ps(T9, T11, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S11 = _mm512_shuffle_ps(T9, T11, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S12 = _mm512_shuffle_ps(T12, T14, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S13 = _mm512_shuffle_ps(T12, T14, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S14 = _mm512_shuffle_ps(T13, T15, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S15 = _mm512_shuffle_ps(T13, T15, _MM_SHUFFLE(3, 2, 3, 2));
+
+  EIGEN_EXTRACT_8f_FROM_16f(S0, S0);
+  EIGEN_EXTRACT_8f_FROM_16f(S1, S1);
+  EIGEN_EXTRACT_8f_FROM_16f(S2, S2);
+  EIGEN_EXTRACT_8f_FROM_16f(S3, S3);
+  EIGEN_EXTRACT_8f_FROM_16f(S4, S4);
+  EIGEN_EXTRACT_8f_FROM_16f(S5, S5);
+  EIGEN_EXTRACT_8f_FROM_16f(S6, S6);
+  EIGEN_EXTRACT_8f_FROM_16f(S7, S7);
+  EIGEN_EXTRACT_8f_FROM_16f(S8, S8);
+  EIGEN_EXTRACT_8f_FROM_16f(S9, S9);
+  EIGEN_EXTRACT_8f_FROM_16f(S10, S10);
+  EIGEN_EXTRACT_8f_FROM_16f(S11, S11);
+  EIGEN_EXTRACT_8f_FROM_16f(S12, S12);
+  EIGEN_EXTRACT_8f_FROM_16f(S13, S13);
+  EIGEN_EXTRACT_8f_FROM_16f(S14, S14);
+  EIGEN_EXTRACT_8f_FROM_16f(S15, S15);
+
+  PacketBlock<Packet8f, 32> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_ps(S0_0, S4_0, 0x20);
+  tmp.packet[1] = _mm256_permute2f128_ps(S1_0, S5_0, 0x20);
+  tmp.packet[2] = _mm256_permute2f128_ps(S2_0, S6_0, 0x20);
+  tmp.packet[3] = _mm256_permute2f128_ps(S3_0, S7_0, 0x20);
+  tmp.packet[4] = _mm256_permute2f128_ps(S0_0, S4_0, 0x31);
+  tmp.packet[5] = _mm256_permute2f128_ps(S1_0, S5_0, 0x31);
+  tmp.packet[6] = _mm256_permute2f128_ps(S2_0, S6_0, 0x31);
+  tmp.packet[7] = _mm256_permute2f128_ps(S3_0, S7_0, 0x31);
+
+  tmp.packet[8] = _mm256_permute2f128_ps(S0_1, S4_1, 0x20);
+  tmp.packet[9] = _mm256_permute2f128_ps(S1_1, S5_1, 0x20);
+  tmp.packet[10] = _mm256_permute2f128_ps(S2_1, S6_1, 0x20);
+  tmp.packet[11] = _mm256_permute2f128_ps(S3_1, S7_1, 0x20);
+  tmp.packet[12] = _mm256_permute2f128_ps(S0_1, S4_1, 0x31);
+  tmp.packet[13] = _mm256_permute2f128_ps(S1_1, S5_1, 0x31);
+  tmp.packet[14] = _mm256_permute2f128_ps(S2_1, S6_1, 0x31);
+  tmp.packet[15] = _mm256_permute2f128_ps(S3_1, S7_1, 0x31);
+
+  // Second set of _m256 outputs
+  tmp.packet[16] = _mm256_permute2f128_ps(S8_0, S12_0, 0x20);
+  tmp.packet[17] = _mm256_permute2f128_ps(S9_0, S13_0, 0x20);
+  tmp.packet[18] = _mm256_permute2f128_ps(S10_0, S14_0, 0x20);
+  tmp.packet[19] = _mm256_permute2f128_ps(S11_0, S15_0, 0x20);
+  tmp.packet[20] = _mm256_permute2f128_ps(S8_0, S12_0, 0x31);
+  tmp.packet[21] = _mm256_permute2f128_ps(S9_0, S13_0, 0x31);
+  tmp.packet[22] = _mm256_permute2f128_ps(S10_0, S14_0, 0x31);
+  tmp.packet[23] = _mm256_permute2f128_ps(S11_0, S15_0, 0x31);
+
+  tmp.packet[24] = _mm256_permute2f128_ps(S8_1, S12_1, 0x20);
+  tmp.packet[25] = _mm256_permute2f128_ps(S9_1, S13_1, 0x20);
+  tmp.packet[26] = _mm256_permute2f128_ps(S10_1, S14_1, 0x20);
+  tmp.packet[27] = _mm256_permute2f128_ps(S11_1, S15_1, 0x20);
+  tmp.packet[28] = _mm256_permute2f128_ps(S8_1, S12_1, 0x31);
+  tmp.packet[29] = _mm256_permute2f128_ps(S9_1, S13_1, 0x31);
+  tmp.packet[30] = _mm256_permute2f128_ps(S10_1, S14_1, 0x31);
+  tmp.packet[31] = _mm256_permute2f128_ps(S11_1, S15_1, 0x31);
+
+  // Pack them into the output
+  PACK_OUTPUT(kernel.packet, tmp.packet, 0, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 1, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 2, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 3, 16);
+
+  PACK_OUTPUT(kernel.packet, tmp.packet, 4, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 5, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 6, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 7, 16);
+
+  PACK_OUTPUT(kernel.packet, tmp.packet, 8, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 9, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 10, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 11, 16);
+
+  PACK_OUTPUT(kernel.packet, tmp.packet, 12, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 13, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 14, 16);
+  PACK_OUTPUT(kernel.packet, tmp.packet, 15, 16);
+}
+#define PACK_OUTPUT_2(OUTPUT, INPUT, INDEX, STRIDE)         \
+  EIGEN_INSERT_8f_INTO_16f(OUTPUT[INDEX], INPUT[2 * INDEX], \
+                           INPUT[2 * INDEX + STRIDE]);
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet16f, 4>& kernel) {
+  __m512 T0 = _mm512_unpacklo_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T1 = _mm512_unpackhi_ps(kernel.packet[0], kernel.packet[1]);
+  __m512 T2 = _mm512_unpacklo_ps(kernel.packet[2], kernel.packet[3]);
+  __m512 T3 = _mm512_unpackhi_ps(kernel.packet[2], kernel.packet[3]);
+
+  __m512 S0 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S1 = _mm512_shuffle_ps(T0, T2, _MM_SHUFFLE(3, 2, 3, 2));
+  __m512 S2 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(1, 0, 1, 0));
+  __m512 S3 = _mm512_shuffle_ps(T1, T3, _MM_SHUFFLE(3, 2, 3, 2));
+
+  EIGEN_EXTRACT_8f_FROM_16f(S0, S0);
+  EIGEN_EXTRACT_8f_FROM_16f(S1, S1);
+  EIGEN_EXTRACT_8f_FROM_16f(S2, S2);
+  EIGEN_EXTRACT_8f_FROM_16f(S3, S3);
+
+  PacketBlock<Packet8f, 8> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_ps(S0_0, S1_0, 0x20);
+  tmp.packet[1] = _mm256_permute2f128_ps(S2_0, S3_0, 0x20);
+  tmp.packet[2] = _mm256_permute2f128_ps(S0_0, S1_0, 0x31);
+  tmp.packet[3] = _mm256_permute2f128_ps(S2_0, S3_0, 0x31);
+
+  tmp.packet[4] = _mm256_permute2f128_ps(S0_1, S1_1, 0x20);
+  tmp.packet[5] = _mm256_permute2f128_ps(S2_1, S3_1, 0x20);
+  tmp.packet[6] = _mm256_permute2f128_ps(S0_1, S1_1, 0x31);
+  tmp.packet[7] = _mm256_permute2f128_ps(S2_1, S3_1, 0x31);
+
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 0, 1);
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 1, 1);
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 2, 1);
+  PACK_OUTPUT_2(kernel.packet, tmp.packet, 3, 1);
+}
+
+#define PACK_OUTPUT_SQ_D(OUTPUT, INPUT, INDEX, STRIDE)                \
+  OUTPUT[INDEX] = _mm512_insertf64x4(OUTPUT[INDEX], INPUT[INDEX], 0); \
+  OUTPUT[INDEX] = _mm512_insertf64x4(OUTPUT[INDEX], INPUT[INDEX + STRIDE], 1);
+
+#define PACK_OUTPUT_D(OUTPUT, INPUT, INDEX, STRIDE)                         \
+  OUTPUT[INDEX] = _mm512_insertf64x4(OUTPUT[INDEX], INPUT[(2 * INDEX)], 0); \
+  OUTPUT[INDEX] =                                                           \
+      _mm512_insertf64x4(OUTPUT[INDEX], INPUT[(2 * INDEX) + STRIDE], 1);
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet8d, 4>& kernel) {
+  __m512d T0 = _mm512_shuffle_pd(kernel.packet[0], kernel.packet[1], 0);
+  __m512d T1 = _mm512_shuffle_pd(kernel.packet[0], kernel.packet[1], 0xff);
+  __m512d T2 = _mm512_shuffle_pd(kernel.packet[2], kernel.packet[3], 0);
+  __m512d T3 = _mm512_shuffle_pd(kernel.packet[2], kernel.packet[3], 0xff);
+
+  PacketBlock<Packet4d, 8> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x20);
+  tmp.packet[1] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x20);
+  tmp.packet[2] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x31);
+  tmp.packet[3] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x31);
+
+  tmp.packet[4] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x20);
+  tmp.packet[5] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x20);
+  tmp.packet[6] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x31);
+  tmp.packet[7] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x31);
+
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 0, 1);
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 1, 1);
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 2, 1);
+  PACK_OUTPUT_D(kernel.packet, tmp.packet, 3, 1);
+}
+
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet8d, 8>& kernel) {
+  __m512d T0 = _mm512_unpacklo_pd(kernel.packet[0], kernel.packet[1]);
+  __m512d T1 = _mm512_unpackhi_pd(kernel.packet[0], kernel.packet[1]);
+  __m512d T2 = _mm512_unpacklo_pd(kernel.packet[2], kernel.packet[3]);
+  __m512d T3 = _mm512_unpackhi_pd(kernel.packet[2], kernel.packet[3]);
+  __m512d T4 = _mm512_unpacklo_pd(kernel.packet[4], kernel.packet[5]);
+  __m512d T5 = _mm512_unpackhi_pd(kernel.packet[4], kernel.packet[5]);
+  __m512d T6 = _mm512_unpacklo_pd(kernel.packet[6], kernel.packet[7]);
+  __m512d T7 = _mm512_unpackhi_pd(kernel.packet[6], kernel.packet[7]);
+
+  PacketBlock<Packet4d, 16> tmp;
+
+  tmp.packet[0] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x20);
+  tmp.packet[1] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x20);
+  tmp.packet[2] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 0),
+                                         _mm512_extractf64x4_pd(T2, 0), 0x31);
+  tmp.packet[3] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 0),
+                                         _mm512_extractf64x4_pd(T3, 0), 0x31);
+
+  tmp.packet[4] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x20);
+  tmp.packet[5] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x20);
+  tmp.packet[6] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T0, 1),
+                                         _mm512_extractf64x4_pd(T2, 1), 0x31);
+  tmp.packet[7] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T1, 1),
+                                         _mm512_extractf64x4_pd(T3, 1), 0x31);
+
+  tmp.packet[8] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 0),
+                                         _mm512_extractf64x4_pd(T6, 0), 0x20);
+  tmp.packet[9] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 0),
+                                         _mm512_extractf64x4_pd(T7, 0), 0x20);
+  tmp.packet[10] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 0),
+                                          _mm512_extractf64x4_pd(T6, 0), 0x31);
+  tmp.packet[11] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 0),
+                                          _mm512_extractf64x4_pd(T7, 0), 0x31);
+
+  tmp.packet[12] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 1),
+                                          _mm512_extractf64x4_pd(T6, 1), 0x20);
+  tmp.packet[13] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 1),
+                                          _mm512_extractf64x4_pd(T7, 1), 0x20);
+  tmp.packet[14] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T4, 1),
+                                          _mm512_extractf64x4_pd(T6, 1), 0x31);
+  tmp.packet[15] = _mm256_permute2f128_pd(_mm512_extractf64x4_pd(T5, 1),
+                                          _mm512_extractf64x4_pd(T7, 1), 0x31);
+
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 0, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 1, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 2, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 3, 8);
+
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 4, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 5, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 6, 8);
+  PACK_OUTPUT_SQ_D(kernel.packet, tmp.packet, 7, 8);
+}
+template <>
+EIGEN_STRONG_INLINE Packet16f pblend(const Selector<16>& /*ifPacket*/,
+                                     const Packet16f& /*thenPacket*/,
+                                     const Packet16f& /*elsePacket*/) {
+  assert(false && "To be implemented");
+  return Packet16f();
+}
+template <>
+EIGEN_STRONG_INLINE Packet8d pblend(const Selector<8>& /*ifPacket*/,
+                                    const Packet8d& /*thenPacket*/,
+                                    const Packet8d& /*elsePacket*/) {
+  assert(false && "To be implemented");
+  return Packet8d();
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PACKET_MATH_AVX512_H
diff --git a/Eigen/src/Core/arch/AltiVec/CMakeLists.txt b/Eigen/src/Core/arch/AltiVec/CMakeLists.txt
deleted file mode 100644
index 9f8d2e9..0000000
--- a/Eigen/src/Core/arch/AltiVec/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Core_arch_AltiVec_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Core_arch_AltiVec_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core/arch/AltiVec COMPONENT Devel
-)
diff --git a/Eigen/src/Core/arch/AltiVec/Complex.h b/Eigen/src/Core/arch/AltiVec/Complex.h
index 68d9a2b..3e66573 100644
--- a/Eigen/src/Core/arch/AltiVec/Complex.h
+++ b/Eigen/src/Core/arch/AltiVec/Complex.h
@@ -2,30 +2,34 @@
 // for linear algebra.
 //
 // Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010-2016 Konstantinos Margaritis <markos@freevec.org>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-#ifndef EIGEN_COMPLEX_ALTIVEC_H
-#define EIGEN_COMPLEX_ALTIVEC_H
+#ifndef EIGEN_COMPLEX32_ALTIVEC_H
+#define EIGEN_COMPLEX32_ALTIVEC_H
 
 namespace Eigen {
 
 namespace internal {
 
-static Packet4ui  p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_ZERO_);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
-static Packet16uc p16uc_COMPLEX_RE   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
-static Packet16uc p16uc_COMPLEX_IM   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 1), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
-static Packet16uc p16uc_COMPLEX_REV  = vec_sld(p16uc_REVERSE, p16uc_REVERSE, 8);//{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
-static Packet16uc p16uc_COMPLEX_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8);//{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
-static Packet16uc p16uc_PSET_HI = (Packet16uc) vec_mergeh((Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 1));//{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
-static Packet16uc p16uc_PSET_LO = (Packet16uc) vec_mergeh((Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 2), (Packet4ui) vec_splat((Packet4ui)p16uc_FORWARD, 3));//{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 };
+static Packet4ui  p4ui_CONJ_XOR = vec_mergeh((Packet4ui)p4i_ZERO, (Packet4ui)p4f_MZERO);//{ 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+#ifdef __VSX__
+#if defined(_BIG_ENDIAN)
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+#else
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_MZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2d_MZERO, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+#endif
+#endif
 
 //---------- float ----------
 struct Packet2cf
 {
-  EIGEN_STRONG_INLINE Packet2cf() {}
+  EIGEN_STRONG_INLINE explicit Packet2cf() : v(p4f_ZERO) {}
   EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
   Packet4f  v;
 };
@@ -33,10 +37,12 @@
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet2cf type;
+  typedef Packet2cf half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 2,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -47,65 +53,78 @@
     HasAbs2   = 0,
     HasMin    = 0,
     HasMax    = 0,
+#ifdef __VSX__
+    HasBlend  = 1,
+#endif
     HasSetLinear = 0
   };
 };
 
-template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
 
 template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
 {
   Packet2cf res;
-  /* On AltiVec we cannot load 64-bit registers, so wa have to take care of alignment */
-  if((ptrdiff_t(&from) % 16) == 0)
+  if((std::ptrdiff_t(&from) % 16) == 0)
     res.v = pload<Packet4f>((const float *)&from);
   else
     res.v = ploadu<Packet4f>((const float *)&from);
-  res.v = vec_perm(res.v, res.v, p16uc_PSET_HI);
+  res.v = vec_perm(res.v, res.v, p16uc_PSET64_HI);
   return res;
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_add(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_sub(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pload<Packet2cf>(const std::complex<float>*        from) { return Packet2cf(pload<Packet4f>((const float *) from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>*       from) { return Packet2cf(ploadu<Packet4f>((const float*) from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>*     from) { return pset1<Packet2cf>(*from); }
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { pstore((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { pstoreu((float*)to, from.v); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  return pload<Packet2cf>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  pstore<std::complex<float> >((std::complex<float> *) af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(a.v + b.v); }
+template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(a.v - b.v); }
 template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(a.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf((Packet4f)vec_xor((Packet4ui)a.v, p4ui_CONJ_XOR)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a) { return Packet2cf(pxor<Packet4f>(a.v, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR))); }
 
 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
   Packet4f v1, v2;
 
   // Permute and multiply the real parts of a and b
-  v1 = vec_perm(a.v, a.v, p16uc_COMPLEX_RE);
+  v1 = vec_perm(a.v, a.v, p16uc_PSET32_WODD);
   // Get the imaginary parts of a
-  v2 = vec_perm(a.v, a.v, p16uc_COMPLEX_IM);
+  v2 = vec_perm(a.v, a.v, p16uc_PSET32_WEVEN);
   // multiply a_re * b 
   v1 = vec_madd(v1, b.v, p4f_ZERO);
   // multiply a_im * b and get the conjugate result
   v2 = vec_madd(v2, b.v, p4f_ZERO);
-  v2 = (Packet4f) vec_xor((Packet4ui)v2, p4ui_CONJ_XOR);
+  v2 = reinterpret_cast<Packet4f>(pxor(v2, reinterpret_cast<Packet4f>(p4ui_CONJ_XOR)));
   // permute back to a proper order
-  v2 = vec_perm(v2, v2, p16uc_COMPLEX_REV);
+  v2 = vec_perm(v2, v2, p16uc_COMPLEX32_REV);
   
-  return Packet2cf(vec_add(v1, v2));
+  return Packet2cf(padd<Packet4f>(v1, v2));
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_and(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_or(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_xor(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(vec_and(a.v, vec_nor(b.v,b.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v, b.v)); }
 
-template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
-template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
-
-template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>*     from)
-{
-  return pset1<Packet2cf>(*from);
-}
-
-template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
-template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
-
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { vec_dstt((float *)addr, DST_CTRL(2,2,32), DST_CHAN); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> * addr)    { EIGEN_PPC_PREFETCH(addr); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
 {
@@ -118,26 +137,30 @@
 template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
 {
   Packet4f rev_a;
-  rev_a = vec_perm(a.v, a.v, p16uc_COMPLEX_REV2);
+  rev_a = vec_perm(a.v, a.v, p16uc_COMPLEX32_REV2);
   return Packet2cf(rev_a);
 }
 
 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
 {
   Packet4f b;
-  b = (Packet4f) vec_sld(a.v, a.v, 8);
-  b = padd(a.v, b);
-  return pfirst(Packet2cf(b));
+  b = vec_sld(a.v, a.v, 8);
+  b = padd<Packet4f>(a.v, b);
+  return pfirst<Packet2cf>(Packet2cf(b));
 }
 
 template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
 {
   Packet4f b1, b2;
-  
-  b1 = (Packet4f) vec_sld(vecs[0].v, vecs[1].v, 8);
-  b2 = (Packet4f) vec_sld(vecs[1].v, vecs[0].v, 8);
-  b2 = (Packet4f) vec_sld(b2, b2, 8);
-  b2 = padd(b1, b2);
+#ifdef _BIG_ENDIAN  
+  b1 = vec_sld(vecs[0].v, vecs[1].v, 8);
+  b2 = vec_sld(vecs[1].v, vecs[0].v, 8);
+#else
+  b1 = vec_sld(vecs[1].v, vecs[0].v, 8);
+  b2 = vec_sld(vecs[0].v, vecs[1].v, 8);
+#endif
+  b2 = vec_sld(b2, b2, 8);
+  b2 = padd<Packet4f>(b1, b2);
 
   return Packet2cf(b2);
 }
@@ -146,10 +169,10 @@
 {
   Packet4f b;
   Packet2cf prod;
-  b = (Packet4f) vec_sld(a.v, a.v, 8);
-  prod = pmul(a, Packet2cf(b));
+  b = vec_sld(a.v, a.v, 8);
+  prod = pmul<Packet2cf>(a, Packet2cf(b));
 
-  return pfirst(prod);
+  return pfirst<Packet2cf>(prod);
 }
 
 template<int Offset>
@@ -159,7 +182,11 @@
   {
     if (Offset==1)
     {
+#ifdef _BIG_ENDIAN
       first.v = vec_sld(first.v, second.v, 8);
+#else
+      first.v = vec_sld(second.v, first.v, 8);
+#endif
     }
   }
 };
@@ -197,21 +224,207 @@
   }
 };
 
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
+
 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
   // TODO optimize it for AltiVec
-  Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a,b);
-  Packet4f s = vec_madd(b.v, b.v, p4f_ZERO);
-  return Packet2cf(pdiv(res.v, vec_add(s,vec_perm(s, s, p16uc_COMPLEX_REV))));
+  Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a, b);
+  Packet4f s = pmul<Packet4f>(b.v, b.v);
+  return Packet2cf(pdiv(res.v, padd<Packet4f>(s, vec_perm(s, s, p16uc_COMPLEX32_REV))));
 }
 
 template<> EIGEN_STRONG_INLINE Packet2cf pcplxflip<Packet2cf>(const Packet2cf& x)
 {
-  return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX_REV));
+  return Packet2cf(vec_perm(x.v, x.v, p16uc_COMPLEX32_REV));
 }
 
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
+{
+  Packet4f tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+  kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+  kernel.packet[0].v = tmp;
+}
+
+#ifdef __VSX__
+template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+  Packet2cf result;
+  result.v = reinterpret_cast<Packet4f>(pblend<Packet2d>(ifPacket, reinterpret_cast<Packet2d>(thenPacket.v), reinterpret_cast<Packet2d>(elsePacket.v)));
+  return result;
+}
+#endif
+
+//---------- double ----------
+#ifdef __VSX__
+struct Packet1cd
+{
+  EIGEN_STRONG_INLINE Packet1cd() {}
+  EIGEN_STRONG_INLINE explicit Packet1cd(const Packet2d& a) : v(a) {}
+  Packet2d v;
+};
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet1cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 0,
+    size = 1,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
+
+template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { return Packet1cd(pload<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { return Packet1cd(ploadu<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { pstoreu((double*)to, from.v); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
+{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride)
+{
+  std::complex<double> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  return pload<Packet1cd>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride)
+{
+  std::complex<double> EIGEN_ALIGN16 af[2];
+  pstore<std::complex<double> >(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
+template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
+template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd(pxor(a.v, reinterpret_cast<Packet2d>(p2ul_CONJ_XOR2))); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  Packet2d a_re, a_im, v1, v2;
+
+  // Permute and multiply the real parts of a and b
+  a_re = vec_perm(a.v, a.v, p16uc_PSET64_HI);
+  // Get the imaginary parts of a
+  a_im = vec_perm(a.v, a.v, p16uc_PSET64_LO);
+  // multiply a_re * b
+  v1 = vec_madd(a_re, b.v, p2d_ZERO);
+  // multiply a_im * b and get the conjugate result
+  v2 = vec_madd(a_im, b.v, p2d_ZERO);
+  v2 = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(v2), reinterpret_cast<Packet4ui>(v2), 8));
+  v2 = pxor(v2, reinterpret_cast<Packet2d>(p2ul_CONJ_XOR1));
+
+  return Packet1cd(padd<Packet2d>(v1, v2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pand(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(por(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pxor(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(pandnot(a.v, b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>*     from)  { return pset1<Packet1cd>(*from); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> * addr)    { EIGEN_PPC_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
+{
+  std::complex<double> EIGEN_ALIGN16 res[2];
+  pstore<std::complex<double> >(res, a);
+
+  return res[0];
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs)        { return vecs[0]; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+
+template<int Offset>
+struct palign_impl<Offset,Packet1cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/)
+  {
+    // FIXME is it sure we never have to align a Packet1cd?
+    // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary...
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
+
+template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  // TODO optimize it for AltiVec
+  Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
+  Packet2d s = pmul<Packet2d>(b.v, b.v);
+  return Packet1cd(pdiv(res.v, padd<Packet2d>(s, vec_perm(s, s, p16uc_REVERSE64))));
+}
+
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+{
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
+{
+  Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+  kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+  kernel.packet[0].v = tmp;
+}
+#endif // __VSX__
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_COMPLEX_ALTIVEC_H
+#endif // EIGEN_COMPLEX32_ALTIVEC_H
diff --git a/Eigen/src/Core/arch/AltiVec/MathFunctions.h b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
new file mode 100644
index 0000000..c5e4bed
--- /dev/null
+++ b/Eigen/src/Core/arch/AltiVec/MathFunctions.h
@@ -0,0 +1,322 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2007 Julien Pommier
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+/* The sin, cos, exp, and log functions of this file come from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+#ifndef EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+#define EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+
+namespace Eigen {
+
+namespace internal {
+
+static _EIGEN_DECLARE_CONST_Packet4f(1 , 1.0f);
+static _EIGEN_DECLARE_CONST_Packet4f(half, 0.5f);
+static _EIGEN_DECLARE_CONST_Packet4i(0x7f, 0x7f);
+static _EIGEN_DECLARE_CONST_Packet4i(23, 23);
+
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(inv_mant_mask, ~0x7f800000);
+
+/* the smallest non denormalized float number */
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos,  0x00800000);
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_inf,     0xff800000); // -1.f/0.f
+static _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_nan,     0xffffffff);
+  
+/* natural logarithm computed for 4 simultaneous float
+  return NaN for x <= 0
+*/
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_SQRTHF, 0.707106781186547524f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p0, 7.0376836292E-2f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p1, - 1.1514610310E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p2, 1.1676998740E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p3, - 1.2420140846E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p4, + 1.4249322787E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p5, - 1.6668057665E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p6, + 2.0000714765E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p7, - 2.4999993993E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_p8, + 3.3333331174E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_q1, -2.12194440e-4f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_log_q2, 0.693359375f);
+
+static _EIGEN_DECLARE_CONST_Packet4f(exp_hi,  88.3762626647950f);
+static _EIGEN_DECLARE_CONST_Packet4f(exp_lo, -88.3762626647949f);
+
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_LOG2EF, 1.44269504088896341f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C1, 0.693359375f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C2, -2.12194440e-4f);
+
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p0, 1.9875691500E-4f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p1, 1.3981999507E-3f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p2, 8.3334519073E-3f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p3, 4.1665795894E-2f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p4, 1.6666665459E-1f);
+static _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p5, 5.0000001201E-1f);
+
+#ifdef __VSX__
+static _EIGEN_DECLARE_CONST_Packet2d(1 , 1.0);
+static _EIGEN_DECLARE_CONST_Packet2d(2 , 2.0);
+static _EIGEN_DECLARE_CONST_Packet2d(half, 0.5);
+
+static _EIGEN_DECLARE_CONST_Packet2d(exp_hi,  709.437);
+static _EIGEN_DECLARE_CONST_Packet2d(exp_lo, -709.436139303);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_LOG2EF, 1.4426950408889634073599);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p0, 1.26177193074810590878e-4);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p1, 3.02994407707441961300e-2);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q0, 3.00198505138664455042e-6);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q1, 2.52448340349684104192e-3);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q2, 2.27265548208155028766e-1);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q3, 2.00000000000000000009e0);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C1, 0.693145751953125);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
+
+#ifdef __POWER8_VECTOR__
+static Packet2l p2l_1023 = { 1023, 1023 };
+static Packet2ul p2ul_52 = { 52, 52 };
+#endif
+
+#endif
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f plog<Packet4f>(const Packet4f& _x)
+{
+  Packet4f x = _x;
+
+  Packet4i emm0;
+
+  /* isvalid_mask is 0 if x < 0 or x is NaN. */
+  Packet4ui isvalid_mask = reinterpret_cast<Packet4ui>(vec_cmpge(x, p4f_ZERO));
+  Packet4ui iszero_mask = reinterpret_cast<Packet4ui>(vec_cmpeq(x, p4f_ZERO));
+
+  x = pmax(x, p4f_min_norm_pos);  /* cut off denormalized stuff */
+  emm0 = vec_sr(reinterpret_cast<Packet4i>(x),
+                reinterpret_cast<Packet4ui>(p4i_23));
+
+  /* keep only the fractional part */
+  x = pand(x, p4f_inv_mant_mask);
+  x = por(x, p4f_half);
+
+  emm0 = psub(emm0, p4i_0x7f);
+  Packet4f e = padd(vec_ctf(emm0, 0), p4f_1);
+
+  /* part2:
+     if( x < SQRTHF ) {
+       e -= 1;
+       x = x + x - 1.0;
+     } else { x = x - 1.0; }
+  */
+  Packet4f mask = reinterpret_cast<Packet4f>(vec_cmplt(x, p4f_cephes_SQRTHF));
+  Packet4f tmp = pand(x, mask);
+  x = psub(x, p4f_1);
+  e = psub(e, pand(p4f_1, mask));
+  x = padd(x, tmp);
+
+  Packet4f x2 = pmul(x,x);
+  Packet4f x3 = pmul(x2,x);
+
+  Packet4f y, y1, y2;
+  y  = pmadd(p4f_cephes_log_p0, x, p4f_cephes_log_p1);
+  y1 = pmadd(p4f_cephes_log_p3, x, p4f_cephes_log_p4);
+  y2 = pmadd(p4f_cephes_log_p6, x, p4f_cephes_log_p7);
+  y  = pmadd(y , x, p4f_cephes_log_p2);
+  y1 = pmadd(y1, x, p4f_cephes_log_p5);
+  y2 = pmadd(y2, x, p4f_cephes_log_p8);
+  y = pmadd(y, x3, y1);
+  y = pmadd(y, x3, y2);
+  y = pmul(y, x3);
+
+  y1 = pmul(e, p4f_cephes_log_q1);
+  tmp = pmul(x2, p4f_half);
+  y = padd(y, y1);
+  x = psub(x, tmp);
+  y2 = pmul(e, p4f_cephes_log_q2);
+  x = padd(x, y);
+  x = padd(x, y2);
+  // negative arg will be NAN, 0 will be -INF
+  x = vec_sel(x, p4f_minus_inf, iszero_mask);
+  x = vec_sel(p4f_minus_nan, x, isvalid_mask);
+  return x;
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pexp<Packet4f>(const Packet4f& _x)
+{
+  Packet4f x = _x;
+
+  Packet4f tmp, fx;
+  Packet4i emm0;
+
+  // clamp x
+  x = pmax(pmin(x, p4f_exp_hi), p4f_exp_lo);
+
+  // express exp(x) as exp(g + n*log(2))
+  fx = pmadd(x, p4f_cephes_LOG2EF, p4f_half);
+
+  fx = pfloor(fx);
+
+  tmp = pmul(fx, p4f_cephes_exp_C1);
+  Packet4f z = pmul(fx, p4f_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  z = pmul(x,x);
+
+  Packet4f y = p4f_cephes_exp_p0;
+  y = pmadd(y, x, p4f_cephes_exp_p1);
+  y = pmadd(y, x, p4f_cephes_exp_p2);
+  y = pmadd(y, x, p4f_cephes_exp_p3);
+  y = pmadd(y, x, p4f_cephes_exp_p4);
+  y = pmadd(y, x, p4f_cephes_exp_p5);
+  y = pmadd(y, z, x);
+  y = padd(y, p4f_1);
+
+  // build 2^n
+  emm0 = vec_cts(fx, 0);
+  emm0 = vec_add(emm0, p4i_0x7f);
+  emm0 = vec_sl(emm0, reinterpret_cast<Packet4ui>(p4i_23));
+
+  // Altivec's max & min operators just drop silent NaNs. Check NaNs in 
+  // inputs and return them unmodified.
+  Packet4ui isnumber_mask = reinterpret_cast<Packet4ui>(vec_cmpeq(_x, _x));
+  return vec_sel(_x, pmax(pmul(y, reinterpret_cast<Packet4f>(emm0)), _x),
+                 isnumber_mask);
+}
+
+#ifndef EIGEN_COMP_CLANG
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& x)
+{
+  return  vec_rsqrt(x);
+}
+#endif
+
+#ifdef __VSX__
+#ifndef EIGEN_COMP_CLANG
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d prsqrt<Packet2d>(const Packet2d& x)
+{
+  return  vec_rsqrt(x);
+}
+#endif
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psqrt<Packet4f>(const Packet4f& x)
+{
+  return  vec_sqrt(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d psqrt<Packet2d>(const Packet2d& x)
+{
+  return  vec_sqrt(x);
+}
+
+// VSX support varies between different compilers and even different
+// versions of the same compiler.  For gcc version >= 4.9.3, we can use
+// vec_cts to efficiently convert Packet2d to Packet2l.  Otherwise, use
+// a slow version that works with older compilers. 
+// Update: apparently vec_cts/vec_ctf intrinsics for 64-bit doubles
+// are buggy, https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70963
+static inline Packet2l ConvertToPacket2l(const Packet2d& x) {
+#if EIGEN_GNUC_AT_LEAST(5, 4) || \
+    (EIGEN_GNUC_AT(6, 1) && __GNUC_PATCHLEVEL__ >= 1)
+  return vec_cts(x, 0);    // TODO: check clang version.
+#else
+  double tmp[2];
+  memcpy(tmp, &x, sizeof(tmp));
+  Packet2l l = { static_cast<long long>(tmp[0]),
+                 static_cast<long long>(tmp[1]) };
+  return l;
+#endif
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d pexp<Packet2d>(const Packet2d& _x)
+{
+  Packet2d x = _x;
+
+  Packet2d tmp, fx;
+  Packet2l emm0;
+
+  // clamp x
+  x = pmax(pmin(x, p2d_exp_hi), p2d_exp_lo);
+
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = pmadd(x, p2d_cephes_LOG2EF, p2d_half);
+
+  fx = pfloor(fx);
+
+  tmp = pmul(fx, p2d_cephes_exp_C1);
+  Packet2d z = pmul(fx, p2d_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  Packet2d x2 = pmul(x,x);
+
+  Packet2d px = p2d_cephes_exp_p0;
+  px = pmadd(px, x2, p2d_cephes_exp_p1);
+  px = pmadd(px, x2, p2d_cephes_exp_p2);
+  px = pmul (px, x);
+
+  Packet2d qx = p2d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p2d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q3);
+
+  x = pdiv(px,psub(qx,px));
+  x = pmadd(p2d_2,x,p2d_1);
+
+  // build 2^n
+  emm0 = ConvertToPacket2l(fx);
+
+#ifdef __POWER8_VECTOR__ 
+  emm0 = vec_add(emm0, p2l_1023);
+  emm0 = vec_sl(emm0, p2ul_52);
+#else
+  // Code is a bit complex for POWER7.  There is actually a
+  // vec_xxsldi intrinsic but it is not supported by some gcc versions.
+  // So we shift (52-32) bits and do a word swap with zeros.
+  _EIGEN_DECLARE_CONST_Packet4i(1023, 1023);
+  _EIGEN_DECLARE_CONST_Packet4i(20, 20);    // 52 - 32
+
+  Packet4i emm04i = reinterpret_cast<Packet4i>(emm0);
+  emm04i = vec_add(emm04i, p4i_1023);
+  emm04i = vec_sl(emm04i, reinterpret_cast<Packet4ui>(p4i_20));
+  static const Packet16uc perm = {
+    0x14, 0x15, 0x16, 0x17, 0x00, 0x01, 0x02, 0x03, 
+    0x1c, 0x1d, 0x1e, 0x1f, 0x08, 0x09, 0x0a, 0x0b };
+#ifdef  _BIG_ENDIAN
+  emm0 = reinterpret_cast<Packet2l>(vec_perm(p4i_ZERO, emm04i, perm));
+#else
+  emm0 = reinterpret_cast<Packet2l>(vec_perm(emm04i, p4i_ZERO, perm));
+#endif
+
+#endif
+
+  // Altivec's max & min operators just drop silent NaNs. Check NaNs in 
+  // inputs and return them unmodified.
+  Packet2ul isnumber_mask = reinterpret_cast<Packet2ul>(vec_cmpeq(_x, _x));
+  return vec_sel(_x, pmax(pmul(x, reinterpret_cast<Packet2d>(emm0)), _x),
+                 isnumber_mask);
+}
+#endif
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_MATH_FUNCTIONS_ALTIVEC_H
diff --git a/Eigen/src/Core/arch/AltiVec/PacketMath.h b/Eigen/src/Core/arch/AltiVec/PacketMath.h
old mode 100644
new mode 100755
index e408996..08a27d1
--- a/Eigen/src/Core/arch/AltiVec/PacketMath.h
+++ b/Eigen/src/Core/arch/AltiVec/PacketMath.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Konstantinos Margaritis <markos@codex.gr>
+// Copyright (C) 2008-2016 Konstantinos Margaritis <markos@freevec.org>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,13 +18,17 @@
 #define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 4
 #endif
 
-#ifndef EIGEN_HAS_FUSE_CJMADD
-#define EIGEN_HAS_FUSE_CJMADD 1
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
 #endif
 
 // NOTE Altivec has 32 registers, but Eigen only accepts a value of 8 or 16
 #ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
-#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 16
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS  32
 #endif
 
 typedef __vector float          Packet4f;
@@ -38,7 +42,7 @@
 // and it doesn't really work to declare them global, so we define macros instead
 
 #define _EIGEN_DECLARE_CONST_FAST_Packet4f(NAME,X) \
-  Packet4f p4f_##NAME = (Packet4f) vec_splat_s32(X)
+  Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(vec_splat_s32(X))
 
 #define _EIGEN_DECLARE_CONST_FAST_Packet4i(NAME,X) \
   Packet4i p4i_##NAME = vec_splat_s32(X)
@@ -46,60 +50,158 @@
 #define _EIGEN_DECLARE_CONST_Packet4f(NAME,X) \
   Packet4f p4f_##NAME = pset1<Packet4f>(X)
 
-#define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
-  Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int>(X))
-
 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
   Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
+#define _EIGEN_DECLARE_CONST_Packet2d(NAME,X) \
+  Packet2d p2d_##NAME = pset1<Packet2d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet2l(NAME,X) \
+  Packet2l p2l_##NAME = pset1<Packet2l>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
+  const Packet4f p4f_##NAME = reinterpret_cast<Packet4f>(pset1<Packet4i>(X))
+
 #define DST_CHAN 1
 #define DST_CTRL(size, count, stride) (((size) << 24) | ((count) << 16) | (stride))
 
-// Define global static constants:
-static Packet4f p4f_COUNTDOWN = { 3.0, 2.0, 1.0, 0.0 };
-static Packet4i p4i_COUNTDOWN = { 3, 2, 1, 0 };
-static Packet16uc p16uc_REVERSE = {12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3};
-static Packet16uc p16uc_FORWARD = vec_lvsl(0, (float*)0);
-static Packet16uc p16uc_DUPLICATE = {0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7};
 
-static _EIGEN_DECLARE_CONST_FAST_Packet4f(ZERO, 0);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE,1);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS16,-16);
-static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS1,-1);
-static Packet4f p4f_ONE = vec_ctf(p4i_ONE, 0);
-static Packet4f p4f_ZERO_ = (Packet4f) vec_sl((Packet4ui)p4i_MINUS1, (Packet4ui)p4i_MINUS1);
+// These constants are endian-agnostic
+static _EIGEN_DECLARE_CONST_FAST_Packet4f(ZERO, 0); //{ 0.0, 0.0, 0.0, 0.0}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE,1); //{ 1, 1, 1, 1}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS16,-16); //{ -16, -16, -16, -16}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(MINUS1,-1); //{ -1, -1, -1, -1}
+static Packet4f p4f_MZERO = (Packet4f) vec_sl((Packet4ui)p4i_MINUS1, (Packet4ui)p4i_MINUS1); //{ 0x80000000, 0x80000000, 0x80000000, 0x80000000}
+#ifndef __VSX__
+static Packet4f p4f_ONE = vec_ctf(p4i_ONE, 0); //{ 1.0, 1.0, 1.0, 1.0}
+#endif
+
+static Packet4f p4f_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 };
+static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 };
+
+static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 };
+static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
+
+// Mask alignment
+#ifdef __PPC64__
+#define _EIGEN_MASK_ALIGNMENT	0xfffffffffffffff0
+#else
+#define _EIGEN_MASK_ALIGNMENT	0xfffffff0
+#endif
+
+#define _EIGEN_ALIGNED_PTR(x)	((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
+
+// Handle endianness properly while loading constants
+// Define global static constants:
+#ifdef _BIG_ENDIAN
+static Packet16uc p16uc_FORWARD = vec_lvsl(0, (float*)0);
+#ifdef __VSX__
+static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+#endif
+static Packet16uc p16uc_PSET32_WODD   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
+static Packet16uc p16uc_PSET32_WEVEN  = vec_sld(p16uc_DUPLICATE32_HI, (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
+static Packet16uc p16uc_HALF64_0_16 = vec_sld((Packet16uc)p4i_ZERO, vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 3), 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
+#else
+static Packet16uc p16uc_FORWARD = p16uc_REVERSE32;
+static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+static Packet16uc p16uc_PSET32_WODD = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 1), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
+static Packet16uc p16uc_PSET32_WEVEN = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
+static Packet16uc p16uc_HALF64_0_16 = vec_sld(vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 0), (Packet16uc)p4i_ZERO, 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
+#endif // _BIG_ENDIAN
+
+static Packet16uc p16uc_PSET64_HI = (Packet16uc) vec_mergeh((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
+static Packet16uc p16uc_PSET64_LO = (Packet16uc) vec_mergel((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 };
+static Packet16uc p16uc_TRANSPOSE64_HI = p16uc_PSET64_HI + p16uc_HALF64_0_16;                                         //{ 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
+static Packet16uc p16uc_TRANSPOSE64_LO = p16uc_PSET64_LO + p16uc_HALF64_0_16;                                         //{ 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};
+
+static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32, 8);                                         //{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
+
+#ifdef _BIG_ENDIAN
+static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8);                                            //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+#else
+static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_PSET64_HI, p16uc_PSET64_LO, 8);                                            //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+#endif // _BIG_ENDIAN
+
+#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
+  #define EIGEN_PPC_PREFETCH(ADDR) __builtin_prefetch(ADDR);
+#else
+  #define EIGEN_PPC_PREFETCH(ADDR) asm( "   dcbt [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
+#endif
 
 template<> struct packet_traits<float>  : default_packet_traits
 {
   typedef Packet4f type;
+  typedef Packet4f half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size=4,
+    HasHalfPacket = 1,
 
-    // FIXME check the Has*
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
     HasSin  = 0,
     HasCos  = 0,
     HasLog  = 0,
-    HasExp  = 0,
-    HasSqrt = 0
+    HasExp  = 1,
+#ifdef __VSX__
+    HasSqrt = 1,
+#if !EIGEN_COMP_CLANG
+    HasRsqrt = 1,
+#else
+    HasRsqrt = 0,
+#endif
+#else
+    HasSqrt = 0,
+    HasRsqrt = 0,
+#endif
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
   };
 };
 template<> struct packet_traits<int>    : default_packet_traits
 {
   typedef Packet4i type;
+  typedef Packet4i half;
   enum {
-    // FIXME check the Has*
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4
+    size = 4,
+    HasHalfPacket = 0,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 0,
+    HasBlend = 1
   };
 };
 
-template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4}; };
-template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4}; };
-/*
+
+template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
+
+inline std::ostream & operator <<(std::ostream & s, const Packet16uc & v)
+{
+  union {
+    Packet16uc   v;
+    unsigned char n[16];
+  } vt;
+  vt.v = v;
+  for (int i=0; i< 16; i++)
+    s << (int)vt.n[i] << ", ";
+  return s;
+}
+
 inline std::ostream & operator <<(std::ostream & s, const Packet4f & v)
 {
   union {
@@ -133,89 +235,136 @@
   return s;
 }
 
-inline std::ostream & operator <<(std::ostream & s, const Packetbi & v)
+// Need to define them first or we get specialization after instantiation errors
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from)
 {
-  union {
-    Packet4bi v;
-    unsigned int n[4];
-  } vt;
-  vt.v = v;
-  s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
-  return s;
+  EIGEN_DEBUG_ALIGNED_LOAD
+#ifdef __VSX__
+  return vec_vsx_ld(0, from);
+#else
+  return vec_ld(0, from);
+#endif
 }
-*/
+
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from)
+{
+  EIGEN_DEBUG_ALIGNED_LOAD
+#ifdef __VSX__
+  return vec_vsx_ld(0, from);
+#else
+  return vec_ld(0, from);
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+#ifdef __VSX__
+  vec_vsx_st(from, 0, to);
+#else
+  vec_st(from, 0, to);
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+#ifdef __VSX__
+  vec_vsx_st(from, 0, to);
+#else
+  vec_st(from, 0, to);
+#endif
+}
+
 template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) {
-  // Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
-  float EIGEN_ALIGN16 af[4];
-  af[0] = from;
-  Packet4f vc = vec_ld(0, af);
-  vc = vec_splat(vc, 0);
-  return vc;
+  Packet4f v = {from, from, from, from};
+  return v;
 }
 
 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   {
-  int EIGEN_ALIGN16 ai[4];
-  ai[0] = from;
-  Packet4i vc = vec_ld(0, ai);
-  vc = vec_splat(vc, 0);
-  return vc;
+  Packet4i v = {from, from, from, from};
+  return v;
+}
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = pload<Packet4f>(a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
+}
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4i>(const int *a,
+                      Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3)
+{
+  a3 = pload<Packet4i>(a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
 }
 
-template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a) { return vec_add(pset1<Packet4f>(a), p4f_COUNTDOWN); }
-template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a)     { return vec_add(pset1<Packet4i>(a), p4i_COUNTDOWN); }
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+  float EIGEN_ALIGN16 af[4];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  af[2] = from[2*stride];
+  af[3] = from[3*stride];
+ return pload<Packet4f>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return pload<Packet4i>(ai);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  float EIGEN_ALIGN16 af[4];
+  pstore<float>(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+  to[2*stride] = af[2];
+  to[3*stride] = af[3];
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  pstore<int>((int *)ai, from);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
 
-template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_add(a,b); }
-template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_add(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a) { return pset1<Packet4f>(a) + p4f_COUNTDOWN; }
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)   { return pset1<Packet4i>(a) + p4i_COUNTDOWN; }
 
-template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_sub(a,b); }
-template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_sub(a,b); }
+template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return a + b; }
+template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return a + b; }
 
-template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a) { return psub<Packet4f>(p4f_ZERO, a); }
-template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return psub<Packet4i>(p4i_ZERO, a); }
+template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b) { return a - b; }
+template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return a - b; }
+
+template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a) { return p4f_ZERO - a; }
+template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return p4i_ZERO - a; }
 
 template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
 template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
 
-template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_madd(a,b,p4f_ZERO); }
-/* Commented out: it's actually slower than processing it scalar
- *
-template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b)
-{
-  // Detailed in: http://freevec.org/content/32bit_signed_integer_multiplication_altivec
-  //Set up constants, variables
-  Packet4i a1, b1, bswap, low_prod, high_prod, prod, prod_, v1sel;
+template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_madd(a,b, p4f_MZERO); }
+template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return a * b; }
 
-  // Get the absolute values
-  a1  = vec_abs(a);
-  b1  = vec_abs(b);
-
-  // Get the signs using xor
-  Packet4bi sgn = (Packet4bi) vec_cmplt(vec_xor(a, b), p4i_ZERO);
-
-  // Do the multiplication for the asbolute values.
-  bswap = (Packet4i) vec_rl((Packet4ui) b1, (Packet4ui) p4i_MINUS16 );
-  low_prod = vec_mulo((Packet8i) a1, (Packet8i)b1);
-  high_prod = vec_msum((Packet8i) a1, (Packet8i) bswap, p4i_ZERO);
-  high_prod = (Packet4i) vec_sl((Packet4ui) high_prod, (Packet4ui) p4i_MINUS16);
-  prod = vec_add( low_prod, high_prod );
-
-  // NOR the product and select only the negative elements according to the sign mask
-  prod_ = vec_nor(prod, prod);
-  prod_ = vec_sel(p4i_ZERO, prod_, sgn);
-
-  // Add 1 to the result to get the negative numbers
-  v1sel = vec_sel(p4i_ZERO, p4i_ONE, sgn);
-  prod_ = vec_add(prod_, v1sel);
-
-  // Merge the results back to the final vector.
-  prod = vec_sel(prod, prod_, sgn);
-
-  return prod;
-}
-*/
 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
 {
-  Packet4f t, y_0, y_1, res;
+#ifndef __VSX__  // VSX actually provides a div instruction
+  Packet4f t, y_0, y_1;
 
   // Altivec does not offer a divide instruction, we have to do a reciprocal approximation
   y_0 = vec_re(b);
@@ -224,8 +373,10 @@
   t   = vec_nmsub(y_0, b, p4f_ONE);
   y_1 = vec_madd(y_0, t, y_0);
 
-  res = vec_madd(a, y_1, p4f_ZERO);
-  return res;
+  return vec_madd(a, y_1, p4f_MZERO);
+#else
+  return vec_div(a, b);
+#endif
 }
 
 template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& /*a*/, const Packet4i& /*b*/)
@@ -234,16 +385,33 @@
 }
 
 // for some weird raisons, it has to be overloaded for packet of integers
-template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vec_madd(a, b, c); }
-template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd(pmul(a,b), c); }
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vec_madd(a,b,c); }
+template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return a*b + c; }
 
-template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_min(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  #ifdef __VSX__
+  Packet4f ret;
+  __asm__ ("xvcmpgesp %x0,%x1,%x2\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+  #else
+  return vec_min(a, b);
+  #endif
+}
 template<> EIGEN_STRONG_INLINE Packet4i pmin<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_min(a, b); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_max(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  #ifdef __VSX__
+  Packet4f ret;
+  __asm__ ("xvcmpgtsp %x0,%x2,%x1\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+  #else
+  return vec_max(a, b);
+  #endif
+}
 template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); }
 
-// Logical Operations are not supported for float, so we have to reinterpret casts using NEON intrinsics
 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, b); }
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); }
 
@@ -256,13 +424,14 @@
 template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b) { return vec_and(a, vec_nor(b, b)); }
 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, vec_nor(b, b)); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) { EIGEN_DEBUG_ALIGNED_LOAD return vec_ld(0, from); }
-template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return vec_ld(0, from); }
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const  Packet4f& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) { return vec_floor(a); }
 
+#ifdef _BIG_ENDIAN
 template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
 {
   EIGEN_DEBUG_ALIGNED_LOAD
-  // Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
   Packet16uc MSQ, LSQ;
   Packet16uc mask;
   MSQ = vec_ld(0, (unsigned char *)from);          // most significant quadword
@@ -282,25 +451,36 @@
   mask = vec_lvsl(0, from);                        // create the permute mask
   return (Packet4i) vec_perm(MSQ, LSQ, mask);    // align the data
 }
+#else
+// We also need ot redefine little endian loading of Packet4i/Packet4f using VSX
+template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
+{
+  EIGEN_DEBUG_UNALIGNED_LOAD
+  return (Packet4i) vec_vsx_ld((long)from & 15, (const int*) _EIGEN_ALIGNED_PTR(from));
+}
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
+{
+  EIGEN_DEBUG_UNALIGNED_LOAD
+  return (Packet4f) vec_vsx_ld((long)from & 15, (const float*) _EIGEN_ALIGNED_PTR(from));
+}
+#endif
 
 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
 {
   Packet4f p;
-  if((ptrdiff_t(&from) % 16) == 0)  p = pload<Packet4f>(from);
-  else                              p = ploadu<Packet4f>(from);
-  return vec_perm(p, p, p16uc_DUPLICATE);
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet4f>(from);
+  else                                  p = ploadu<Packet4f>(from);
+  return vec_perm(p, p, p16uc_DUPLICATE32_HI);
 }
 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
 {
   Packet4i p;
-  if((ptrdiff_t(&from) % 16) == 0)  p = pload<Packet4i>(from);
-  else                              p = ploadu<Packet4i>(from);
-  return vec_perm(p, p, p16uc_DUPLICATE);
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet4i>(from);
+  else                                  p = ploadu<Packet4i>(from);
+  return vec_perm(p, p, p16uc_DUPLICATE32_HI);
 }
 
-template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vec_st(from, 0, to); }
-template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vec_st(from, 0, to); }
-
+#ifdef _BIG_ENDIAN
 template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const Packet4f& from)
 {
   EIGEN_DEBUG_UNALIGNED_STORE
@@ -337,15 +517,33 @@
   vec_st( LSQ, 15, (unsigned char *)to );                   // Store the LSQ part first
   vec_st( MSQ, 0, (unsigned char *)to );                    // Store the MSQ part
 }
+#else
+// We also need ot redefine little endian loading of Packet4i/Packet4f using VSX
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*       to, const Packet4i& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_vsx_st(from, (long)to & 15, (int*) _EIGEN_ALIGNED_PTR(to));
+}
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*   to, const Packet4f& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_vsx_st(from, (long)to & 15, (float*) _EIGEN_ALIGNED_PTR(to));
+}
+#endif
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { vec_dstt(addr, DST_CTRL(2,2,32), DST_CHAN); }
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr) { vec_dstt(addr, DST_CTRL(2,2,32), DST_CHAN); }
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr)    { EIGEN_PPC_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr)    { EIGEN_PPC_PREFETCH(addr); }
 
-template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; vec_st(a, 0, x); return x[0]; }
-template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int   EIGEN_ALIGN16 x[4]; vec_st(a, 0, x); return x[0]; }
+template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x; vec_ste(a, 0, &x); return x; }
+template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int   EIGEN_ALIGN16 x; vec_ste(a, 0, &x); return x; }
 
-template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) { return (Packet4f)vec_perm((Packet16uc)a,(Packet16uc)a, p16uc_REVERSE); }
-template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a) { return (Packet4i)vec_perm((Packet16uc)a,(Packet16uc)a, p16uc_REVERSE); }
+template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a)
+{
+  return reinterpret_cast<Packet4f>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32));
+}
+template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
+{
+  return reinterpret_cast<Packet4i>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32)); }
 
 template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a) { return vec_abs(a); }
 template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vec_abs(a); }
@@ -353,10 +551,10 @@
 template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
 {
   Packet4f b, sum;
-  b   = (Packet4f) vec_sld(a, a, 8);
-  sum = vec_add(a, b);
-  b   = (Packet4f) vec_sld(sum, sum, 4);
-  sum = vec_add(sum, b);
+  b   = vec_sld(a, a, 8);
+  sum = a + b;
+  b   = vec_sld(sum, sum, 4);
+  sum += b;
   return pfirst(sum);
 }
 
@@ -379,11 +577,11 @@
 
   // Now do the summation:
   // Lines 0+1
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
   // Lines 2+3
-  sum[1] = vec_add(sum[2], sum[3]);
+  sum[1] = sum[2] + sum[3];
   // Add the results
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
 
   return sum[0];
 }
@@ -392,7 +590,11 @@
 {
   Packet4i sum;
   sum = vec_sums(a, p4i_ZERO);
+#ifdef _BIG_ENDIAN
   sum = vec_sld(sum, p4i_ZERO, 12);
+#else
+  sum = vec_sld(p4i_ZERO, sum, 4);
+#endif
   return pfirst(sum);
 }
 
@@ -415,11 +617,11 @@
 
   // Now do the summation:
   // Lines 0+1
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
   // Lines 2+3
-  sum[1] = vec_add(sum[2], sum[3]);
+  sum[1] = sum[2] + sum[3];
   // Add the results
-  sum[0] = vec_add(sum[0], sum[1]);
+  sum[0] = sum[0] + sum[1];
 
   return sum[0];
 }
@@ -429,8 +631,8 @@
 template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
 {
   Packet4f prod;
-  prod = pmul(a, (Packet4f)vec_sld(a, a, 8));
-  return pfirst(pmul(prod, (Packet4f)vec_sld(prod, prod, 4)));
+  prod = pmul(a, vec_sld(a, a, 8));
+  return pfirst(pmul(prod, vec_sld(prod, prod, 4)));
 }
 
 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
@@ -479,8 +681,25 @@
 {
   static EIGEN_STRONG_INLINE void run(Packet4f& first, const Packet4f& second)
   {
-    if (Offset!=0)
-      first = vec_sld(first, second, Offset*4);
+#ifdef _BIG_ENDIAN
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(first, second, 4); break;
+    case 2:
+      first = vec_sld(first, second, 8); break;
+    case 3:
+      first = vec_sld(first, second, 12); break;
+    }
+#else
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(second, first, 12); break;
+    case 2:
+      first = vec_sld(second, first, 8); break;
+    case 3:
+      first = vec_sld(second, first, 4); break;
+    }
+#endif
   }
 };
 
@@ -489,11 +708,352 @@
 {
   static EIGEN_STRONG_INLINE void run(Packet4i& first, const Packet4i& second)
   {
-    if (Offset!=0)
-      first = vec_sld(first, second, Offset*4);
+#ifdef _BIG_ENDIAN
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(first, second, 4); break;
+    case 2:
+      first = vec_sld(first, second, 8); break;
+    case 3:
+      first = vec_sld(first, second, 12); break;
+    }
+#else
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(second, first, 12); break;
+    case 2:
+      first = vec_sld(second, first, 8); break;
+    case 3:
+      first = vec_sld(second, first, 4); break;
+    }
+#endif
   }
 };
 
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  Packet4f t0, t1, t2, t3;
+  t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
+  t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
+  t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
+  t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
+  kernel.packet[0] = vec_mergeh(t0, t2);
+  kernel.packet[1] = vec_mergel(t0, t2);
+  kernel.packet[2] = vec_mergeh(t1, t3);
+  kernel.packet[3] = vec_mergel(t1, t3);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  Packet4i t0, t1, t2, t3;
+  t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
+  t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
+  t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
+  t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
+  kernel.packet[0] = vec_mergeh(t0, t2);
+  kernel.packet[1] = vec_mergel(t0, t2);
+  kernel.packet[2] = vec_mergeh(t1, t3);
+  kernel.packet[3] = vec_mergel(t1, t3);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) {
+  Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] };
+  Packet4ui mask = reinterpret_cast<Packet4ui>(vec_cmpeq(reinterpret_cast<Packet4ui>(select), reinterpret_cast<Packet4ui>(p4i_ONE)));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) {
+  Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] };
+  Packet4ui mask = reinterpret_cast<Packet4ui>(vec_cmpeq(reinterpret_cast<Packet4ui>(select), reinterpret_cast<Packet4ui>(p4i_ONE)));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+
+//---------- double ----------
+#ifdef __VSX__
+typedef __vector double              Packet2d;
+typedef __vector unsigned long long  Packet2ul;
+typedef __vector long long           Packet2l;
+#if EIGEN_COMP_CLANG
+typedef Packet2ul                    Packet2bl;
+#else
+typedef __vector __bool long         Packet2bl;
+#endif
+
+static Packet2l  p2l_ONE  = { 1, 1 };
+static Packet2l  p2l_ZERO = reinterpret_cast<Packet2l>(p4i_ZERO);
+static Packet2d  p2d_ONE  = { 1.0, 1.0 };
+static Packet2d  p2d_ZERO = reinterpret_cast<Packet2d>(p4f_ZERO);
+static Packet2d  p2d_MZERO = { -0.0, -0.0 };
+
+#ifdef _BIG_ENDIAN
+static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(p2d_ZERO), reinterpret_cast<Packet4f>(p2d_ONE), 8));
+#else
+static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(p2d_ONE), reinterpret_cast<Packet4f>(p2d_ZERO), 8));
+#endif
+
+template<int index> Packet2d vec_splat_dbl(Packet2d& a);
+
+template<> EIGEN_STRONG_INLINE Packet2d vec_splat_dbl<0>(Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(a, a, p16uc_PSET64_HI));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d vec_splat_dbl<1>(Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(a, a, p16uc_PSET64_LO));
+}
+
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet2d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 1,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2l & v)
+{
+  union {
+    Packet2l   v;
+    int64_t n[2];
+  } vt;
+  vt.v = v;
+  s << vt.n[0] << ", " << vt.n[1];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2d & v)
+{
+  union {
+    Packet2d   v;
+    double n[2];
+  } vt;
+  vt.v = v;
+  s << vt.n[0] << ", " << vt.n[1];
+  return s;
+}
+
+// Need to define them first or we get specialization after instantiation errors
+template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from)
+{
+  EIGEN_DEBUG_ALIGNED_LOAD
+#ifdef __VSX__
+  return vec_vsx_ld(0, from);
+#else
+  return vec_ld(0, from);
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+#ifdef __VSX__
+  vec_vsx_st(from, 0, to);
+#else
+  vec_st(from, 0, to);
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double&  from) {
+  Packet2d v = {from, from};
+  return v;
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet2d>(const double *a,
+                      Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3)
+{
+  a1 = pload<Packet2d>(a);
+  a0 = vec_splat_dbl<0>(a1);
+  a1 = vec_splat_dbl<1>(a1);
+  a3 = pload<Packet2d>(a+2);
+  a2 = vec_splat_dbl<0>(a3);
+  a3 = vec_splat_dbl<1>(a3);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+ return pload<Packet2d>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  pstore<double>(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return pset1<Packet2d>(a) + p2d_COUNTDOWN; }
+
+template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return a + b; }
+
+template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return a - b; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return p2d_ZERO - a; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_madd(a,b,p2d_MZERO); }
+template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_div(a,b); }
+
+// for some weird raisons, it has to be overloaded for packet of integers
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  Packet2d ret;
+  __asm__ ("xvcmpgedp %x0,%x1,%x2\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+ }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  Packet2d ret;
+  __asm__ ("xvcmpgtdp %x0,%x2,%x1\n\txxsel %x0,%x1,%x2,%x0" : "=&wa" (ret) : "wa" (a), "wa" (b));
+  return ret;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_or(a, b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_xor(a, b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, vec_nor(b, b)); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const  Packet2d& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); }
+
+template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
+{
+  EIGEN_DEBUG_ALIGNED_LOAD
+  return (Packet2d) vec_vsx_ld((long)from & 15, (const double*) _EIGEN_ALIGNED_PTR(from));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
+{
+  Packet2d p;
+  if((std::ptrdiff_t(from) % 16) == 0)  p = pload<Packet2d>(from);
+  else                                  p = ploadu<Packet2d>(from);
+  return vec_splat_dbl<0>(p);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d& from)
+{
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_vsx_st((Packet4f)from, (long)to & 15, (float*) _EIGEN_ALIGNED_PTR(to));
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_PPC_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE double  pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore<double>(x, a); return x[0]; }
+
+template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE64));
+}
+template<> EIGEN_STRONG_INLINE Packet2d pabs(const Packet2d& a) { return vec_abs(a); }
+
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
+{
+  Packet2d b, sum;
+  b   = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(a), reinterpret_cast<Packet4f>(a), 8));
+  sum = a + b;
+  return pfirst<Packet2d>(sum);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
+{
+  Packet2d v[2], sum;
+  v[0] = vecs[0] + reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(vecs[0]), reinterpret_cast<Packet4f>(vecs[0]), 8));
+  v[1] = vecs[1] + reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(vecs[1]), reinterpret_cast<Packet4f>(vecs[1]), 8));
+
+#ifdef _BIG_ENDIAN
+  sum = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(v[0]), reinterpret_cast<Packet4f>(v[1]), 8));
+#else
+  sum = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4f>(v[1]), reinterpret_cast<Packet4f>(v[0]), 8));
+#endif
+
+  return sum;
+}
+// Other reduction functions:
+// mul
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmul(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(a), reinterpret_cast<Packet4ui>(a), 8))));
+}
+
+// min
+template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmin(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(a), reinterpret_cast<Packet4ui>(a), 8))));
+}
+
+// max
+template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmax(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(a), reinterpret_cast<Packet4ui>(a), 8))));
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet2d>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2d& first, const Packet2d& second)
+  {
+    if (Offset == 1)
+#ifdef _BIG_ENDIAN
+      first = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(first), reinterpret_cast<Packet4ui>(second), 8));
+#else
+      first = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(second), reinterpret_cast<Packet4ui>(first), 8));
+#endif
+  }
+};
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  Packet2d t0, t1;
+  t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI);
+  t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO);
+  kernel.packet[0] = t0;
+  kernel.packet[1] = t1;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) {
+  Packet2l select = { ifPacket.select[0], ifPacket.select[1] };
+  Packet2bl mask = reinterpret_cast<Packet2bl>( vec_cmpeq(reinterpret_cast<Packet2d>(select), reinterpret_cast<Packet2d>(p2l_ONE)) );
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+#endif // __VSX__
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/arch/CMakeLists.txt b/Eigen/src/Core/arch/CMakeLists.txt
deleted file mode 100644
index 8456dec..0000000
--- a/Eigen/src/Core/arch/CMakeLists.txt
+++ /dev/null
@@ -1,4 +0,0 @@
-ADD_SUBDIRECTORY(SSE)
-ADD_SUBDIRECTORY(AltiVec)
-ADD_SUBDIRECTORY(NEON)
-ADD_SUBDIRECTORY(Default)
diff --git a/Eigen/src/Core/arch/CUDA/Complex.h b/Eigen/src/Core/arch/CUDA/Complex.h
new file mode 100644
index 0000000..9c25365
--- /dev/null
+++ b/Eigen/src/Core/arch/CUDA/Complex.h
@@ -0,0 +1,103 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLEX_CUDA_H
+#define EIGEN_COMPLEX_CUDA_H
+
+// clang-format off
+
+namespace Eigen {
+
+namespace internal {
+
+#if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
+
+// Many std::complex methods such as operator+, operator-, operator* and
+// operator/ are not constexpr. Due to this, clang does not treat them as device
+// functions and thus Eigen functors making use of these operators fail to
+// compile. Here, we manually specialize these functors for complex types when
+// building for CUDA to avoid non-constexpr methods.
+
+// Sum
+template<typename T> struct scalar_sum_op<const std::complex<T>, const std::complex<T> > : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sum_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    return std::complex<T>(numext::real(a) + numext::real(b),
+                           numext::imag(a) + numext::imag(b));
+  }
+};
+
+template<typename T> struct scalar_sum_op<std::complex<T>, std::complex<T> > : scalar_sum_op<const std::complex<T>, const std::complex<T> > {};
+
+
+// Difference
+template<typename T> struct scalar_difference_op<const std::complex<T>, const std::complex<T> >  : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_difference_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    return std::complex<T>(numext::real(a) - numext::real(b),
+                           numext::imag(a) - numext::imag(b));
+  }
+};
+
+template<typename T> struct scalar_difference_op<std::complex<T>, std::complex<T> > : scalar_difference_op<const std::complex<T>, const std::complex<T> > {};
+
+
+// Product
+template<typename T> struct scalar_product_op<const std::complex<T>, const std::complex<T> >  : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  enum {
+    Vectorizable = packet_traits<std::complex<T>>::HasMul
+  };
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_product_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    const T a_real = numext::real(a);
+    const T a_imag = numext::imag(a);
+    const T b_real = numext::real(b);
+    const T b_imag = numext::imag(b);
+    return std::complex<T>(a_real * b_real - a_imag * b_imag,
+                           a_real * b_imag + a_imag * b_real);
+  }
+};
+
+template<typename T> struct scalar_product_op<std::complex<T>, std::complex<T> > : scalar_product_op<const std::complex<T>, const std::complex<T> > {};
+
+
+// Quotient
+template<typename T> struct scalar_quotient_op<const std::complex<T>, const std::complex<T> > : binary_op_base<const std::complex<T>, const std::complex<T> > {
+  enum {
+    Vectorizable = packet_traits<std::complex<T>>::HasDiv
+  };
+  typedef typename std::complex<T> result_type;
+
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::complex<T> operator() (const std::complex<T>& a, const std::complex<T>& b) const {
+    const T a_real = numext::real(a);
+    const T a_imag = numext::imag(a);
+    const T b_real = numext::real(b);
+    const T b_imag = numext::imag(b);
+    const T norm = T(1) / (b_real * b_real + b_imag * b_imag);
+    return std::complex<T>((a_real * b_real + a_imag * b_imag) * norm,
+                           (a_imag * b_real - a_real * b_imag) * norm);
+  }
+};
+
+template<typename T> struct scalar_quotient_op<std::complex<T>, std::complex<T> > : scalar_quotient_op<const std::complex<T>, const std::complex<T> > {};
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COMPLEX_CUDA_H
diff --git a/Eigen/src/Core/arch/CUDA/Half.h b/Eigen/src/Core/arch/CUDA/Half.h
new file mode 100644
index 0000000..755e620
--- /dev/null
+++ b/Eigen/src/Core/arch/CUDA/Half.h
@@ -0,0 +1,674 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+//
+// The conversion routines are Copyright (c) Fabian Giesen, 2016.
+// The original license follows:
+//
+// Copyright (c) Fabian Giesen, 2016
+// All rights reserved.
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted.
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+
+// Standard 16-bit float type, mostly useful for GPUs. Defines a new
+// type Eigen::half (inheriting from CUDA's __half struct) with
+// operator overloads such that it behaves basically as an arithmetic
+// type. It will be quite slow on CPUs (so it is recommended to stay
+// in float32_bits for CPUs, except for simple parameter conversions, I/O
+// to disk and the likes), but fast on GPUs.
+
+
+#ifndef EIGEN_HALF_CUDA_H
+#define EIGEN_HALF_CUDA_H
+
+#if __cplusplus > 199711L
+#define EIGEN_EXPLICIT_CAST(tgt_type) explicit operator tgt_type()
+#else
+#define EIGEN_EXPLICIT_CAST(tgt_type) operator tgt_type()
+#endif
+
+
+namespace Eigen {
+
+struct half;
+
+namespace half_impl {
+
+#if !defined(EIGEN_HAS_CUDA_FP16)
+// Make our own __half_raw definition that is similar to CUDA's.
+struct __half_raw {
+  EIGEN_DEVICE_FUNC __half_raw() : x(0) {}
+  explicit EIGEN_DEVICE_FUNC __half_raw(unsigned short raw) : x(raw) {}
+  unsigned short x;
+};
+#elif defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER < 90000
+// In CUDA < 9.0, __half is the equivalent of CUDA 9's __half_raw
+typedef __half __half_raw;
+#endif
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw raw_uint16_to_half(unsigned short x);
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw float_to_half_rtne(float ff);
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float half_to_float(__half_raw h);
+
+struct half_base : public __half_raw {
+  EIGEN_DEVICE_FUNC half_base() {}
+  EIGEN_DEVICE_FUNC half_base(const half_base& h) : __half_raw(h) {}
+  EIGEN_DEVICE_FUNC half_base(const __half_raw& h) : __half_raw(h) {}
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER >= 90000
+  EIGEN_DEVICE_FUNC half_base(const __half& h) : __half_raw(*(__half_raw*)&h) {}
+#endif
+};
+
+} // namespace half_impl
+
+// Class definition.
+struct half : public half_impl::half_base {
+  #if !defined(EIGEN_HAS_CUDA_FP16) || (defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER < 90000)
+    typedef half_impl::__half_raw __half_raw;
+  #endif
+
+  EIGEN_DEVICE_FUNC half() {}
+
+  EIGEN_DEVICE_FUNC half(const __half_raw& h) : half_impl::half_base(h) {}
+  EIGEN_DEVICE_FUNC half(const half& h) : half_impl::half_base(h) {}
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDACC_VER) && EIGEN_CUDACC_VER >= 90000
+  EIGEN_DEVICE_FUNC half(const __half& h) : half_impl::half_base(h) {}
+#endif
+
+  explicit EIGEN_DEVICE_FUNC half(bool b)
+      : half_impl::half_base(half_impl::raw_uint16_to_half(b ? 0x3c00 : 0)) {}
+  template<class T>
+  explicit EIGEN_DEVICE_FUNC half(const T& val)
+      : half_impl::half_base(half_impl::float_to_half_rtne(static_cast<float>(val))) {}
+  explicit EIGEN_DEVICE_FUNC half(float f)
+      : half_impl::half_base(half_impl::float_to_half_rtne(f)) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(bool) const {
+    // +0.0 and -0.0 become false, everything else becomes true.
+    return (x & 0x7fff) != 0;
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(signed char) const {
+    return static_cast<signed char>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned char) const {
+    return static_cast<unsigned char>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(short) const {
+    return static_cast<short>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned short) const {
+    return static_cast<unsigned short>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(int) const {
+    return static_cast<int>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned int) const {
+    return static_cast<unsigned int>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(long) const {
+    return static_cast<long>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned long) const {
+    return static_cast<unsigned long>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(long long) const {
+    return static_cast<long long>(half_impl::half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(unsigned long long) const {
+    return static_cast<unsigned long long>(half_to_float(*this));
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(float) const {
+    return half_impl::half_to_float(*this);
+  }
+  EIGEN_DEVICE_FUNC EIGEN_EXPLICIT_CAST(double) const {
+    return static_cast<double>(half_impl::half_to_float(*this));
+  }
+
+  EIGEN_DEVICE_FUNC half& operator=(const half& other) {
+    x = other.x;
+    return *this;
+  }
+};
+
+} // end namespace Eigen
+
+namespace std {
+template<>
+struct numeric_limits<Eigen::half> {
+  static const bool is_specialized = true;
+  static const bool is_signed = true;
+  static const bool is_integer = false;
+  static const bool is_exact = false;
+  static const bool has_infinity = true;
+  static const bool has_quiet_NaN = true;
+  static const bool has_signaling_NaN = true;
+  static const float_denorm_style has_denorm = denorm_present;
+  static const bool has_denorm_loss = false;
+  static const std::float_round_style round_style = std::round_to_nearest;
+  static const bool is_iec559 = false;
+  static const bool is_bounded = false;
+  static const bool is_modulo = false;
+  static const int digits = 11;
+  static const int digits10 = 3;      // according to http://half.sourceforge.net/structstd_1_1numeric__limits_3_01half__float_1_1half_01_4.html
+  static const int max_digits10 = 5;  // according to http://half.sourceforge.net/structstd_1_1numeric__limits_3_01half__float_1_1half_01_4.html
+  static const int radix = 2;
+  static const int min_exponent = -13;
+  static const int min_exponent10 = -4;
+  static const int max_exponent = 16;
+  static const int max_exponent10 = 4;
+  static const bool traps = true;
+  static const bool tinyness_before = false;
+
+  static Eigen::half (min)() { return Eigen::half_impl::raw_uint16_to_half(0x400); }
+  static Eigen::half lowest() { return Eigen::half_impl::raw_uint16_to_half(0xfbff); }
+  static Eigen::half (max)() { return Eigen::half_impl::raw_uint16_to_half(0x7bff); }
+  static Eigen::half epsilon() { return Eigen::half_impl::raw_uint16_to_half(0x0800); }
+  static Eigen::half round_error() { return Eigen::half(0.5); }
+  static Eigen::half infinity() { return Eigen::half_impl::raw_uint16_to_half(0x7c00); }
+  static Eigen::half quiet_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7e00); }
+  static Eigen::half signaling_NaN() { return Eigen::half_impl::raw_uint16_to_half(0x7e00); }
+  static Eigen::half denorm_min() { return Eigen::half_impl::raw_uint16_to_half(0x1); }
+};
+
+// If std::numeric_limits<T> is specialized, should also specialize
+// std::numeric_limits<const T>, std::numeric_limits<volatile T>, and
+// std::numeric_limits<const volatile T>
+// https://stackoverflow.com/a/16519653/
+template<>
+struct numeric_limits<const Eigen::half> : numeric_limits<Eigen::half> {};
+template<>
+struct numeric_limits<volatile Eigen::half> : numeric_limits<Eigen::half> {};
+template<>
+struct numeric_limits<const volatile Eigen::half> : numeric_limits<Eigen::half> {};
+} // end namespace std
+
+namespace Eigen {
+
+namespace half_impl {
+
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+
+// Intrinsics for native fp16 support. Note that on current hardware,
+// these are no faster than float32_bits arithmetic (you need to use the half2
+// versions to get the ALU speed increased), but you do save the
+// conversion steps back and forth.
+
+EIGEN_STRONG_INLINE __device__ half operator + (const half& a, const half& b) {
+  return __hadd(a, b);
+}
+EIGEN_STRONG_INLINE __device__ half operator * (const half& a, const half& b) {
+  return __hmul(a, b);
+}
+EIGEN_STRONG_INLINE __device__ half operator - (const half& a, const half& b) {
+  return __hsub(a, b);
+}
+EIGEN_STRONG_INLINE __device__ half operator / (const half& a, const half& b) {
+  float num = __half2float(a);
+  float denom = __half2float(b);
+  return __float2half(num / denom);
+}
+EIGEN_STRONG_INLINE __device__ half operator - (const half& a) {
+  return __hneg(a);
+}
+EIGEN_STRONG_INLINE __device__ half& operator += (half& a, const half& b) {
+  a = a + b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ half& operator *= (half& a, const half& b) {
+  a = a * b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ half& operator -= (half& a, const half& b) {
+  a = a - b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ half& operator /= (half& a, const half& b) {
+  a = a / b;
+  return a;
+}
+EIGEN_STRONG_INLINE __device__ bool operator == (const half& a, const half& b) {
+  return __heq(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator != (const half& a, const half& b) {
+  return __hne(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator < (const half& a, const half& b) {
+  return __hlt(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator <= (const half& a, const half& b) {
+  return __hle(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator > (const half& a, const half& b) {
+  return __hgt(a, b);
+}
+EIGEN_STRONG_INLINE __device__ bool operator >= (const half& a, const half& b) {
+  return __hge(a, b);
+}
+
+#else  // Emulate support for half floats
+
+// Definitions for CPUs and older CUDA, mostly working through conversion
+// to/from float32_bits.
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator + (const half& a, const half& b) {
+  return half(float(a) + float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator * (const half& a, const half& b) {
+  return half(float(a) * float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a, const half& b) {
+  return half(float(a) - float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, const half& b) {
+  return half(float(a) / float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator - (const half& a) {
+  half result;
+  result.x = a.x ^ 0x8000;
+  return result;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator += (half& a, const half& b) {
+  a = half(float(a) + float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator *= (half& a, const half& b) {
+  a = half(float(a) * float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator -= (half& a, const half& b) {
+  a = half(float(a) - float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half& operator /= (half& a, const half& b) {
+  a = half(float(a) / float(b));
+  return a;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator == (const half& a, const half& b) {
+  return numext::equal_strict(float(a),float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator != (const half& a, const half& b) {
+  return numext::not_equal_strict(float(a), float(b));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator < (const half& a, const half& b) {
+  return float(a) < float(b);
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator <= (const half& a, const half& b) {
+  return float(a) <= float(b);
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator > (const half& a, const half& b) {
+  return float(a) > float(b);
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool operator >= (const half& a, const half& b) {
+  return float(a) >= float(b);
+}
+
+#endif  // Emulate support for half floats
+
+// Division by an index. Do it in full float precision to avoid accuracy
+// issues in converting the denominator to half.
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half operator / (const half& a, Index b) {
+  return half(static_cast<float>(a) / static_cast<float>(b));
+}
+
+// Conversion routines, including fallbacks for the host or older CUDA.
+// Note that newer Intel CPUs (Haswell or newer) have vectorized versions of
+// these in hardware. If we need more performance on older/other CPUs, they are
+// also possible to vectorize directly.
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw raw_uint16_to_half(unsigned short x) {
+  __half_raw h;
+  h.x = x;
+  return h;
+}
+
+union float32_bits {
+  unsigned int u;
+  float f;
+};
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC __half_raw float_to_half_rtne(float ff) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300
+  __half tmp_ff = __float2half(ff);
+  return *(__half_raw*)&tmp_ff;
+
+#elif defined(EIGEN_HAS_FP16_C)
+  __half_raw h;
+  h.x = _cvtss_sh(ff, 0);
+  return h;
+
+#else
+  float32_bits f; f.f = ff;
+
+  const float32_bits f32infty = { 255 << 23 };
+  const float32_bits f16max = { (127 + 16) << 23 };
+  const float32_bits denorm_magic = { ((127 - 15) + (23 - 10) + 1) << 23 };
+  unsigned int sign_mask = 0x80000000u;
+  __half_raw o;
+  o.x = static_cast<unsigned short>(0x0u);
+
+  unsigned int sign = f.u & sign_mask;
+  f.u ^= sign;
+
+  // NOTE all the integer compares in this function can be safely
+  // compiled into signed compares since all operands are below
+  // 0x80000000. Important if you want fast straight SSE2 code
+  // (since there's no unsigned PCMPGTD).
+
+  if (f.u >= f16max.u) {  // result is Inf or NaN (all exponent bits set)
+    o.x = (f.u > f32infty.u) ? 0x7e00 : 0x7c00; // NaN->qNaN and Inf->Inf
+  } else {  // (De)normalized number or zero
+    if (f.u < (113 << 23)) {  // resulting FP16 is subnormal or zero
+      // use a magic value to align our 10 mantissa bits at the bottom of
+      // the float. as long as FP addition is round-to-nearest-even this
+      // just works.
+      f.f += denorm_magic.f;
+
+      // and one integer subtract of the bias later, we have our final float!
+      o.x = static_cast<unsigned short>(f.u - denorm_magic.u);
+    } else {
+      unsigned int mant_odd = (f.u >> 13) & 1; // resulting mantissa is odd
+
+      // update exponent, rounding bias part 1
+      f.u += ((unsigned int)(15 - 127) << 23) + 0xfff;
+      // rounding bias part 2
+      f.u += mant_odd;
+      // take the bits!
+      o.x = static_cast<unsigned short>(f.u >> 13);
+    }
+  }
+
+  o.x |= static_cast<unsigned short>(sign >> 16);
+  return o;
+#endif
+}
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC float half_to_float(__half_raw h) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300
+  return __half2float(h);
+
+#elif defined(EIGEN_HAS_FP16_C)
+  return _cvtsh_ss(h.x);
+
+#else
+  const float32_bits magic = { 113 << 23 };
+  const unsigned int shifted_exp = 0x7c00 << 13; // exponent mask after shift
+  float32_bits o;
+
+  o.u = (h.x & 0x7fff) << 13;             // exponent/mantissa bits
+  unsigned int exp = shifted_exp & o.u;   // just the exponent
+  o.u += (127 - 15) << 23;                // exponent adjust
+
+  // handle exponent special cases
+  if (exp == shifted_exp) {     // Inf/NaN?
+    o.u += (128 - 16) << 23;    // extra exp adjust
+  } else if (exp == 0) {        // Zero/Denormal?
+    o.u += 1 << 23;             // extra exp adjust
+    o.f -= magic.f;             // renormalize
+  }
+
+  o.u |= (h.x & 0x8000) << 16;    // sign bit
+  return o.f;
+#endif
+}
+
+// --- standard functions ---
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isinf)(const half& a) {
+  return (a.x & 0x7fff) == 0x7c00;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isnan)(const half& a) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return __hisnan(a);
+#else
+  return (a.x & 0x7fff) > 0x7c00;
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC bool (isfinite)(const half& a) {
+  return !(isinf EIGEN_NOT_A_MACRO (a)) && !(isnan EIGEN_NOT_A_MACRO (a));
+}
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half abs(const half& a) {
+  half result;
+  result.x = a.x & 0x7FFF;
+  return result;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half exp(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+  return half(hexp(a));
+#else
+   return half(::expf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log(const half& a) {
+#if defined(EIGEN_HAS_CUDA_FP16) && EIGEN_CUDACC_VER >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return half(::hlog(a));
+#else
+  return half(::logf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log1p(const half& a) {
+  return half(numext::log1p(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half log10(const half& a) {
+  return half(::log10f(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half sqrt(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+  return half(hsqrt(a));
+#else
+    return half(::sqrtf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half pow(const half& a, const half& b) {
+  return half(::powf(float(a), float(b)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half sin(const half& a) {
+  return half(::sinf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half cos(const half& a) {
+  return half(::cosf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half tan(const half& a) {
+  return half(::tanf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half tanh(const half& a) {
+  return half(::tanhf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half floor(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
+  return half(hfloor(a));
+#else
+  return half(::floorf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half ceil(const half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 300
+  return half(hceil(a));
+#else
+  return half(::ceilf(float(a)));
+#endif
+}
+
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half (min)(const half& a, const half& b) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return __hlt(b, a) ? b : a;
+#else
+  const float f1 = static_cast<float>(a);
+  const float f2 = static_cast<float>(b);
+  return f2 < f1 ? b : a;
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC half (max)(const half& a, const half& b) {
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return __hlt(a, b) ? b : a;
+#else
+  const float f1 = static_cast<float>(a);
+  const float f2 = static_cast<float>(b);
+  return f1 < f2 ? b : a;
+#endif
+}
+
+EIGEN_ALWAYS_INLINE std::ostream& operator << (std::ostream& os, const half& v) {
+  os << static_cast<float>(v);
+  return os;
+}
+
+} // end namespace half_impl
+
+// import Eigen::half_impl::half into Eigen namespace
+// using half_impl::half;
+
+namespace internal {
+
+template<>
+struct random_default_impl<half, false, false>
+{
+  static inline half run(const half& x, const half& y)
+  {
+    return x + (y-x) * half(float(std::rand()) / float(RAND_MAX));
+  }
+  static inline half run()
+  {
+    return run(half(-1.f), half(1.f));
+  }
+};
+
+template<> struct is_arithmetic<half> { enum { value = true }; };
+
+} // end namespace internal
+
+template<> struct NumTraits<Eigen::half>
+    : GenericNumTraits<Eigen::half>
+{
+  enum {
+    IsSigned = true,
+    IsInteger = false,
+    IsComplex = false,
+    RequireInitialization = false
+  };
+
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half epsilon() {
+    return half_impl::raw_uint16_to_half(0x0800);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half dummy_precision() { return Eigen::half(1e-2f); }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half highest() {
+    return half_impl::raw_uint16_to_half(0x7bff);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half lowest() {
+    return half_impl::raw_uint16_to_half(0xfbff);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half infinity() {
+    return half_impl::raw_uint16_to_half(0x7c00);
+  }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Eigen::half quiet_NaN() {
+    return half_impl::raw_uint16_to_half(0x7c01);
+  }
+};
+
+} // end namespace Eigen
+
+// C-like standard mathematical functions and trancendentals.
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half fabsh(const Eigen::half& a) {
+  Eigen::half result;
+  result.x = a.x & 0x7FFF;
+  return result;
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half exph(const Eigen::half& a) {
+  return Eigen::half(::expf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half logh(const Eigen::half& a) {
+#if EIGEN_CUDACC_VER >= 80000 && defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 530
+  return Eigen::half(::hlog(a));
+#else
+  return Eigen::half(::logf(float(a)));
+#endif
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half sqrth(const Eigen::half& a) {
+  return Eigen::half(::sqrtf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half powh(const Eigen::half& a, const Eigen::half& b) {
+  return Eigen::half(::powf(float(a), float(b)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half floorh(const Eigen::half& a) {
+  return Eigen::half(::floorf(float(a)));
+}
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half ceilh(const Eigen::half& a) {
+  return Eigen::half(::ceilf(float(a)));
+}
+
+namespace std {
+
+#if __cplusplus > 199711L
+template <>
+struct hash<Eigen::half> {
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE std::size_t operator()(const Eigen::half& a) const {
+    return static_cast<std::size_t>(a.x);
+  }
+};
+#endif
+
+} // end namespace std
+
+
+// Add the missing shfl_xor intrinsic
+#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 300
+__device__ EIGEN_STRONG_INLINE Eigen::half __shfl_xor(Eigen::half var, int laneMask, int width=warpSize) {
+  #if EIGEN_CUDACC_VER < 90000
+  return static_cast<Eigen::half>(__shfl_xor(static_cast<float>(var), laneMask, width));
+  #else
+  return static_cast<Eigen::half>(__shfl_xor_sync(0xFFFFFFFF, static_cast<float>(var), laneMask, width));
+  #endif
+}
+#endif
+
+// ldg() has an overload for __half_raw, but we also need one for Eigen::half.
+#if defined(EIGEN_CUDA_ARCH) && EIGEN_CUDA_ARCH >= 350
+EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Eigen::half __ldg(const Eigen::half* ptr) {
+  return Eigen::half_impl::raw_uint16_to_half(
+      __ldg(reinterpret_cast<const unsigned short*>(ptr)));
+}
+#endif
+
+
+#if defined(EIGEN_CUDA_ARCH)
+namespace Eigen {
+namespace numext {
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+bool (isnan)(const Eigen::half& h) {
+  return (half_impl::isnan)(h);
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+bool (isinf)(const Eigen::half& h) {
+  return (half_impl::isinf)(h);
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+bool (isfinite)(const Eigen::half& h) {
+  return (half_impl::isfinite)(h);
+}
+
+} // namespace Eigen
+}  // namespace numext
+#endif
+
+#endif // EIGEN_HALF_CUDA_H
diff --git a/Eigen/src/Core/arch/CUDA/MathFunctions.h b/Eigen/src/Core/arch/CUDA/MathFunctions.h
new file mode 100644
index 0000000..0348b41
--- /dev/null
+++ b/Eigen/src/Core/arch/CUDA/MathFunctions.h
@@ -0,0 +1,91 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_MATH_FUNCTIONS_CUDA_H
+#define EIGEN_MATH_FUNCTIONS_CUDA_H
+
+namespace Eigen {
+
+namespace internal {
+
+// Make sure this is only available when targeting a GPU: we don't want to
+// introduce conflicts between these packet_traits definitions and the ones
+// we'll use on the host side (SSE, AVX, ...)
+#if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 plog<float4>(const float4& a)
+{
+  return make_float4(logf(a.x), logf(a.y), logf(a.z), logf(a.w));
+}
+
+template<>  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 plog<double2>(const double2& a)
+{
+  using ::log;
+  return make_double2(log(a.x), log(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 plog1p<float4>(const float4& a)
+{
+  return make_float4(log1pf(a.x), log1pf(a.y), log1pf(a.z), log1pf(a.w));
+}
+
+template<>  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 plog1p<double2>(const double2& a)
+{
+  return make_double2(log1p(a.x), log1p(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 pexp<float4>(const float4& a)
+{
+  return make_float4(expf(a.x), expf(a.y), expf(a.z), expf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 pexp<double2>(const double2& a)
+{
+  using ::exp;
+  return make_double2(exp(a.x), exp(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 psqrt<float4>(const float4& a)
+{
+  return make_float4(sqrtf(a.x), sqrtf(a.y), sqrtf(a.z), sqrtf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 psqrt<double2>(const double2& a)
+{
+  using ::sqrt;
+  return make_double2(sqrt(a.x), sqrt(a.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+float4 prsqrt<float4>(const float4& a)
+{
+  return make_float4(rsqrtf(a.x), rsqrtf(a.y), rsqrtf(a.z), rsqrtf(a.w));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+double2 prsqrt<double2>(const double2& a)
+{
+  return make_double2(rsqrt(a.x), rsqrt(a.y));
+}
+
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_MATH_FUNCTIONS_CUDA_H
diff --git a/Eigen/src/Core/arch/CUDA/PacketMath.h b/Eigen/src/Core/arch/CUDA/PacketMath.h
new file mode 100644
index 0000000..4dda631
--- /dev/null
+++ b/Eigen/src/Core/arch/CUDA/PacketMath.h
@@ -0,0 +1,333 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_CUDA_H
+#define EIGEN_PACKET_MATH_CUDA_H
+
+namespace Eigen {
+
+namespace internal {
+
+// Make sure this is only available when targeting a GPU: we don't want to
+// introduce conflicts between these packet_traits definitions and the ones
+// we'll use on the host side (SSE, AVX, ...)
+#if defined(__CUDACC__) && defined(EIGEN_USE_GPU)
+template<> struct is_arithmetic<float4>  { enum { value = true }; };
+template<> struct is_arithmetic<double2> { enum { value = true }; };
+
+template<> struct packet_traits<float> : default_packet_traits
+{
+  typedef float4 type;
+  typedef float4 half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=4,
+    HasHalfPacket = 0,
+
+    HasDiv  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasLGamma = 1,
+    HasDiGamma = 1,
+    HasZeta = 1,
+    HasPolygamma = 1,
+    HasErf = 1,
+    HasErfc = 1,
+    HasIGamma = 1,
+    HasIGammac = 1,
+    HasBetaInc = 1,
+
+    HasBlend = 0,
+  };
+};
+
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef double2 type;
+  typedef double2 half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 0,
+
+    HasDiv  = 1,
+    HasLog  = 1,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasLGamma = 1,
+    HasDiGamma = 1,
+    HasZeta = 1,
+    HasPolygamma = 1,
+    HasErf = 1,
+    HasErfc = 1,
+    HasIGamma = 1,
+    HasIGammac = 1,
+    HasBetaInc = 1,
+
+    HasBlend = 0,
+  };
+};
+
+
+template<> struct unpacket_traits<float4>  { typedef float  type; enum {size=4, alignment=Aligned16}; typedef float4 half; };
+template<> struct unpacket_traits<double2> { typedef double type; enum {size=2, alignment=Aligned16}; typedef double2 half; };
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pset1<float4>(const float&  from) {
+  return make_float4(from, from, from, from);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pset1<double2>(const double& from) {
+  return make_double2(from, from);
+}
+
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 plset<float4>(const float& a) {
+  return make_float4(a, a+1, a+2, a+3);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 plset<double2>(const double& a) {
+  return make_double2(a, a+1);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 padd<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x+b.x, a.y+b.y, a.z+b.z, a.w+b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 padd<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x+b.x, a.y+b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 psub<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x-b.x, a.y-b.y, a.z-b.z, a.w-b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 psub<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x-b.x, a.y-b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pnegate(const float4& a) {
+  return make_float4(-a.x, -a.y, -a.z, -a.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pnegate(const double2& a) {
+  return make_double2(-a.x, -a.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pconj(const float4& a) { return a; }
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pconj(const double2& a) { return a; }
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pmul<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x*b.x, a.y*b.y, a.z*b.z, a.w*b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pmul<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x*b.x, a.y*b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pdiv<float4>(const float4& a, const float4& b) {
+  return make_float4(a.x/b.x, a.y/b.y, a.z/b.z, a.w/b.w);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pdiv<double2>(const double2& a, const double2& b) {
+  return make_double2(a.x/b.x, a.y/b.y);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pmin<float4>(const float4& a, const float4& b) {
+  return make_float4(fminf(a.x, b.x), fminf(a.y, b.y), fminf(a.z, b.z), fminf(a.w, b.w));
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pmin<double2>(const double2& a, const double2& b) {
+  return make_double2(fmin(a.x, b.x), fmin(a.y, b.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pmax<float4>(const float4& a, const float4& b) {
+  return make_float4(fmaxf(a.x, b.x), fmaxf(a.y, b.y), fmaxf(a.z, b.z), fmaxf(a.w, b.w));
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pmax<double2>(const double2& a, const double2& b) {
+  return make_double2(fmax(a.x, b.x), fmax(a.y, b.y));
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pload<float4>(const float* from) {
+  return *reinterpret_cast<const float4*>(from);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 pload<double2>(const double* from) {
+  return *reinterpret_cast<const double2*>(from);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 ploadu<float4>(const float* from) {
+  return make_float4(from[0], from[1], from[2], from[3]);
+}
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE double2 ploadu<double2>(const double* from) {
+  return make_double2(from[0], from[1]);
+}
+
+template<> EIGEN_STRONG_INLINE float4 ploaddup<float4>(const float*   from) {
+  return make_float4(from[0], from[0], from[1], from[1]);
+}
+template<> EIGEN_STRONG_INLINE double2 ploaddup<double2>(const double*  from) {
+  return make_double2(from[0], from[0]);
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstore<float>(float*   to, const float4& from) {
+  *reinterpret_cast<float4*>(to) = from;
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstore<double>(double* to, const double2& from) {
+  *reinterpret_cast<double2*>(to) = from;
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const float4& from) {
+  to[0] = from.x;
+  to[1] = from.y;
+  to[2] = from.z;
+  to[3] = from.w;
+}
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const double2& from) {
+  to[0] = from.x;
+  to[1] = from.y;
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Aligned>(const float* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return __ldg((const float4*)from);
+#else
+  return make_float4(from[0], from[1], from[2], from[3]);
+#endif
+}
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE double2 ploadt_ro<double2, Aligned>(const double* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return __ldg((const double2*)from);
+#else
+  return make_double2(from[0], from[1]);
+#endif
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE float4 ploadt_ro<float4, Unaligned>(const float* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return make_float4(__ldg(from+0), __ldg(from+1), __ldg(from+2), __ldg(from+3));
+#else
+  return make_float4(from[0], from[1], from[2], from[3]);
+#endif
+}
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE double2 ploadt_ro<double2, Unaligned>(const double* from) {
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 350
+  return make_double2(__ldg(from+0), __ldg(from+1));
+#else
+  return make_double2(from[0], from[1]);
+#endif
+}
+
+template<> EIGEN_DEVICE_FUNC inline float4 pgather<float, float4>(const float* from, Index stride) {
+  return make_float4(from[0*stride], from[1*stride], from[2*stride], from[3*stride]);
+}
+
+template<> EIGEN_DEVICE_FUNC inline double2 pgather<double, double2>(const double* from, Index stride) {
+  return make_double2(from[0*stride], from[1*stride]);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, float4>(float* to, const float4& from, Index stride) {
+  to[stride*0] = from.x;
+  to[stride*1] = from.y;
+  to[stride*2] = from.z;
+  to[stride*3] = from.w;
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, double2>(double* to, const double2& from, Index stride) {
+  to[stride*0] = from.x;
+  to[stride*1] = from.y;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  pfirst<float4>(const float4& a) {
+  return a.x;
+}
+template<> EIGEN_DEVICE_FUNC inline double pfirst<double2>(const double2& a) {
+  return a.x;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux<float4>(const float4& a) {
+  return a.x + a.y + a.z + a.w;
+}
+template<> EIGEN_DEVICE_FUNC inline double predux<double2>(const double2& a) {
+  return a.x + a.y;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux_max<float4>(const float4& a) {
+  return fmaxf(fmaxf(a.x, a.y), fmaxf(a.z, a.w));
+}
+template<> EIGEN_DEVICE_FUNC inline double predux_max<double2>(const double2& a) {
+  return fmax(a.x, a.y);
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux_min<float4>(const float4& a) {
+  return fminf(fminf(a.x, a.y), fminf(a.z, a.w));
+}
+template<> EIGEN_DEVICE_FUNC inline double predux_min<double2>(const double2& a) {
+  return fmin(a.x, a.y);
+}
+
+template<> EIGEN_DEVICE_FUNC inline float  predux_mul<float4>(const float4& a) {
+  return a.x * a.y * a.z * a.w;
+}
+template<> EIGEN_DEVICE_FUNC inline double predux_mul<double2>(const double2& a) {
+  return a.x * a.y;
+}
+
+template<> EIGEN_DEVICE_FUNC inline float4  pabs<float4>(const float4& a) {
+  return make_float4(fabsf(a.x), fabsf(a.y), fabsf(a.z), fabsf(a.w));
+}
+template<> EIGEN_DEVICE_FUNC inline double2 pabs<double2>(const double2& a) {
+  return make_double2(fabs(a.x), fabs(a.y));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<float4,4>& kernel) {
+  float tmp = kernel.packet[0].y;
+  kernel.packet[0].y = kernel.packet[1].x;
+  kernel.packet[1].x = tmp;
+
+  tmp = kernel.packet[0].z;
+  kernel.packet[0].z = kernel.packet[2].x;
+  kernel.packet[2].x = tmp;
+
+  tmp = kernel.packet[0].w;
+  kernel.packet[0].w = kernel.packet[3].x;
+  kernel.packet[3].x = tmp;
+
+  tmp = kernel.packet[1].z;
+  kernel.packet[1].z = kernel.packet[2].y;
+  kernel.packet[2].y = tmp;
+
+  tmp = kernel.packet[1].w;
+  kernel.packet[1].w = kernel.packet[3].y;
+  kernel.packet[3].y = tmp;
+
+  tmp = kernel.packet[2].w;
+  kernel.packet[2].w = kernel.packet[3].z;
+  kernel.packet[3].z = tmp;
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<double2,2>& kernel) {
+  double tmp = kernel.packet[0].y;
+  kernel.packet[0].y = kernel.packet[1].x;
+  kernel.packet[1].x = tmp;
+}
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+
+#endif // EIGEN_PACKET_MATH_CUDA_H
diff --git a/Eigen/src/Core/arch/CUDA/PacketMathHalf.h b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
new file mode 100644
index 0000000..c66d384
--- /dev/null
+++ b/Eigen/src/Core/arch/CUDA/PacketMathHalf.h
@@ -0,0 +1,1124 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_HALF_CUDA_H
+#define EIGEN_PACKET_MATH_HALF_CUDA_H
+
+
+namespace Eigen {
+namespace internal {
+
+// Most of the following operations require arch >= 3.0
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDACC__) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+
+template<> struct is_arithmetic<half2> { enum { value = true }; };
+
+template<> struct packet_traits<Eigen::half> : default_packet_traits
+{
+  typedef half2 type;
+  typedef half2 half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 0,
+    HasAdd    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasSqrt   = 1,
+    HasRsqrt  = 1,
+    HasExp    = 1,
+    HasLog    = 1,
+    HasLog1p  = 1
+  };
+};
+
+template<> struct unpacket_traits<half2> { typedef Eigen::half type; enum {size=2, alignment=Aligned16}; typedef half2 half; };
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pset1<half2>(const Eigen::half& from) {
+  return __half2half2(from);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pload<half2>(const Eigen::half* from) {
+  return *reinterpret_cast<const half2*>(from);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 ploadu<half2>(const Eigen::half* from) {
+  return __halves2half2(from[0], from[1]);
+}
+
+template<> EIGEN_STRONG_INLINE half2 ploaddup<half2>(const Eigen::half*  from) {
+  return __halves2half2(from[0], from[0]);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE void pstore<Eigen::half>(Eigen::half* to, const half2& from) {
+  *reinterpret_cast<half2*>(to) = from;
+}
+
+template<> __device__ EIGEN_STRONG_INLINE void pstoreu<Eigen::half>(Eigen::half* to, const half2& from) {
+  to[0] = __low2half(from);
+  to[1] = __high2half(from);
+}
+
+template<>
+ __device__ EIGEN_ALWAYS_INLINE half2 ploadt_ro<half2, Aligned>(const Eigen::half* from) {
+#if __CUDA_ARCH__ >= 350
+   return __ldg((const half2*)from);
+#else
+  return __halves2half2(*(from+0), *(from+1));
+#endif
+}
+
+template<>
+__device__ EIGEN_ALWAYS_INLINE half2 ploadt_ro<half2, Unaligned>(const Eigen::half* from) {
+#if __CUDA_ARCH__ >= 350
+   return __halves2half2(__ldg(from+0), __ldg(from+1));
+#else
+  return __halves2half2(*(from+0), *(from+1));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pgather<Eigen::half, half2>(const Eigen::half* from, Index stride) {
+  return __halves2half2(from[0*stride], from[1*stride]);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE void pscatter<Eigen::half, half2>(Eigen::half* to, const half2& from, Index stride) {
+  to[stride*0] = __low2half(from);
+  to[stride*1] = __high2half(from);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half pfirst<half2>(const half2& a) {
+  return __low2half(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pabs<half2>(const half2& a) {
+  half2 result;
+  unsigned temp = *(reinterpret_cast<const unsigned*>(&(a)));
+  *(reinterpret_cast<unsigned*>(&(result))) = temp & 0x7FFF7FFF;
+  return result;
+}
+
+
+__device__ EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<half2,2>& kernel) {
+  __half a1 = __low2half(kernel.packet[0]);
+  __half a2 = __high2half(kernel.packet[0]);
+  __half b1 = __low2half(kernel.packet[1]);
+  __half b2 = __high2half(kernel.packet[1]);
+  kernel.packet[0] = __halves2half2(a1, b1);
+  kernel.packet[1] = __halves2half2(a2, b2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 plset<half2>(const Eigen::half& a) {
+#if __CUDA_ARCH__ >= 530
+  return __halves2half2(a, __hadd(a, __float2half(1.0f)));
+#else
+  float f = __half2float(a) + 1.0f;
+  return __halves2half2(a, __float2half(f));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 padd<half2>(const half2& a, const half2& b) {
+#if __CUDA_ARCH__ >= 530
+  return __hadd2(a, b);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 + b1;
+  float r2 = a2 + b2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 psub<half2>(const half2& a, const half2& b) {
+#if __CUDA_ARCH__ >= 530
+  return __hsub2(a, b);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 - b1;
+  float r2 = a2 - b2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pnegate(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  return __hneg2(a);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return __floats2half2_rn(-a1, -a2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pconj(const half2& a) { return a; }
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmul<half2>(const half2& a, const half2& b) {
+#if __CUDA_ARCH__ >= 530
+  return __hmul2(a, b);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 * b1;
+  float r2 = a2 * b2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmadd<half2>(const half2& a, const half2& b, const half2& c) {
+#if __CUDA_ARCH__ >= 530
+   return __hfma2(a, b, c);
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float c1 = __low2float(c);
+  float c2 = __high2float(c);
+  float r1 = a1 * b1 + c1;
+  float r2 = a2 * b2 + c2;
+  return __floats2half2_rn(r1, r2);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pdiv<half2>(const half2& a, const half2& b) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  float r1 = a1 / b1;
+  float r2 = a2 / b2;
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmin<half2>(const half2& a, const half2& b) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  __half r1 = a1 < b1 ? __low2half(a) : __low2half(b);
+  __half r2 = a2 < b2 ? __high2half(a) : __high2half(b);
+  return __halves2half2(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pmax<half2>(const half2& a, const half2& b) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float b1 = __low2float(b);
+  float b2 = __high2float(b);
+  __half r1 = a1 > b1 ? __low2half(a) : __low2half(b);
+  __half r2 = a2 > b2 ? __high2half(a) : __high2half(b);
+  return __halves2half2(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  return __hadd(__low2half(a), __high2half(a));
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return Eigen::half(half_impl::raw_uint16_to_half(__float2half_rn(a1 + a2)));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux_max<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  __half first = __low2half(a);
+  __half second = __high2half(a);
+  return __hgt(first, second) ? first : second;
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return a1 > a2 ? __low2half(a) : __high2half(a);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux_min<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  __half first = __low2half(a);
+  __half second = __high2half(a);
+  return __hlt(first, second) ? first : second;
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return a1 < a2 ? __low2half(a) : __high2half(a);
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE Eigen::half predux_mul<half2>(const half2& a) {
+#if __CUDA_ARCH__ >= 530
+  return __hmul(__low2half(a), __high2half(a));
+#else
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  return Eigen::half(half_impl::raw_uint16_to_half(__float2half_rn(a1 * a2)));
+#endif
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 plog1p<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = log1pf(a1);
+  float r2 = log1pf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+#if EIGEN_CUDACC_VER >= 80000 && defined EIGEN_CUDA_ARCH && EIGEN_CUDA_ARCH >= 530
+
+template<>  __device__ EIGEN_STRONG_INLINE
+half2 plog<half2>(const half2& a) {
+  return h2log(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE
+half2 pexp<half2>(const half2& a) {
+  return h2exp(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE
+half2 psqrt<half2>(const half2& a) {
+  return h2sqrt(a);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE
+half2 prsqrt<half2>(const half2& a) {
+  return h2rsqrt(a);
+}
+
+#else
+
+template<> __device__ EIGEN_STRONG_INLINE half2 plog<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = logf(a1);
+  float r2 = logf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 pexp<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = expf(a1);
+  float r2 = expf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 psqrt<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = sqrtf(a1);
+  float r2 = sqrtf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+template<> __device__ EIGEN_STRONG_INLINE half2 prsqrt<half2>(const half2& a) {
+  float a1 = __low2float(a);
+  float a2 = __high2float(a);
+  float r1 = rsqrtf(a1);
+  float r2 = rsqrtf(a2);
+  return __floats2half2_rn(r1, r2);
+}
+
+#endif
+
+#elif defined EIGEN_VECTORIZE_AVX512
+
+typedef struct {
+  __m256i x;
+} Packet16h;
+
+
+template<> struct is_arithmetic<Packet16h> { enum { value = true }; };
+
+template <>
+struct packet_traits<half> : default_packet_traits {
+  typedef Packet16h type;
+  // There is no half-size packet for Packet16h.
+  typedef Packet16h half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 16,
+    HasHalfPacket = 0,
+    HasAdd    = 0,
+    HasSub    = 0,
+    HasMul    = 0,
+    HasNegate = 0,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasConj   = 0,
+    HasSetLinear = 0,
+    HasDiv = 0,
+    HasSqrt = 0,
+    HasRsqrt = 0,
+    HasExp = 0,
+    HasLog = 0,
+    HasBlend = 0
+  };
+};
+
+
+template<> struct unpacket_traits<Packet16h> { typedef Eigen::half type; enum {size=16, alignment=Aligned32}; typedef Packet16h half; };
+
+template<> EIGEN_STRONG_INLINE Packet16h pset1<Packet16h>(const Eigen::half& from) {
+  Packet16h result;
+  result.x = _mm256_set1_epi16(from.x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half pfirst<Packet16h>(const Packet16h& from) {
+  return half_impl::raw_uint16_to_half(static_cast<unsigned short>(_mm256_extract_epi16(from.x, 0)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h pload<Packet16h>(const Eigen::half* from) {
+  Packet16h result;
+  result.x = _mm256_load_si256(reinterpret_cast<const __m256i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h ploadu<Packet16h>(const Eigen::half* from) {
+  Packet16h result;
+  result.x = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<half>(Eigen::half* to, const Packet16h& from) {
+  _mm256_store_si256((__m256i*)to, from.x);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<half>(Eigen::half* to, const Packet16h& from) {
+  _mm256_storeu_si256((__m256i*)to, from.x);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h
+ploadquad(const Eigen::half* from) {
+  Packet16h result;
+  unsigned short a = from[0].x;
+  unsigned short b = from[1].x;
+  unsigned short c = from[2].x;
+  unsigned short d = from[3].x;
+  result.x = _mm256_set_epi16(d, d, d, d, c, c, c, c, b, b, b, b, a, a, a, a);
+  return result;
+}
+
+EIGEN_STRONG_INLINE Packet16f half2float(const Packet16h& a) {
+#ifdef EIGEN_HAS_FP16_C
+  return _mm512_cvtph_ps(a.x);
+#else
+  EIGEN_ALIGN64 half aux[16];
+  pstore(aux, a);
+  float f0(aux[0]);
+  float f1(aux[1]);
+  float f2(aux[2]);
+  float f3(aux[3]);
+  float f4(aux[4]);
+  float f5(aux[5]);
+  float f6(aux[6]);
+  float f7(aux[7]);
+  float f8(aux[8]);
+  float f9(aux[9]);
+  float fa(aux[10]);
+  float fb(aux[11]);
+  float fc(aux[12]);
+  float fd(aux[13]);
+  float fe(aux[14]);
+  float ff(aux[15]);
+
+  return _mm512_set_ps(
+      ff, fe, fd, fc, fb, fa, f9, f8, f7, f6, f5, f4, f3, f2, f1, f0);
+#endif
+}
+
+EIGEN_STRONG_INLINE Packet16h float2half(const Packet16f& a) {
+#ifdef EIGEN_HAS_FP16_C
+  Packet16h result;
+  result.x = _mm512_cvtps_ph(a, _MM_FROUND_TO_NEAREST_INT|_MM_FROUND_NO_EXC);
+  return result;
+#else
+  EIGEN_ALIGN64 float aux[16];
+  pstore(aux, a);
+  half h0(aux[0]);
+  half h1(aux[1]);
+  half h2(aux[2]);
+  half h3(aux[3]);
+  half h4(aux[4]);
+  half h5(aux[5]);
+  half h6(aux[6]);
+  half h7(aux[7]);
+  half h8(aux[8]);
+  half h9(aux[9]);
+  half ha(aux[10]);
+  half hb(aux[11]);
+  half hc(aux[12]);
+  half hd(aux[13]);
+  half he(aux[14]);
+  half hf(aux[15]);
+
+  Packet16h result;
+  result.x = _mm256_set_epi16(
+      hf.x, he.x, hd.x, hc.x, hb.x, ha.x, h9.x, h8.x,
+      h7.x, h6.x, h5.x, h4.x, h3.x, h2.x, h1.x, h0.x);
+  return result;
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h padd<Packet16h>(const Packet16h& a, const Packet16h& b) {
+  Packet16f af = half2float(a);
+  Packet16f bf = half2float(b);
+  Packet16f rf = padd(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h pmul<Packet16h>(const Packet16h& a, const Packet16h& b) {
+  Packet16f af = half2float(a);
+  Packet16f bf = half2float(b);
+  Packet16f rf = pmul(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE half predux<Packet16h>(const Packet16h& from) {
+  Packet16f from_float = half2float(from);
+  return half(predux(from_float));
+}
+
+template<> EIGEN_STRONG_INLINE Packet16h pgather<Eigen::half, Packet16h>(const Eigen::half* from, Index stride)
+{
+  Packet16h result;
+  result.x = _mm256_set_epi16(
+      from[15*stride].x, from[14*stride].x, from[13*stride].x, from[12*stride].x,
+      from[11*stride].x, from[10*stride].x, from[9*stride].x, from[8*stride].x,
+      from[7*stride].x, from[6*stride].x, from[5*stride].x, from[4*stride].x,
+      from[3*stride].x, from[2*stride].x, from[1*stride].x, from[0*stride].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pscatter<half, Packet16h>(half* to, const Packet16h& from, Index stride)
+{
+  EIGEN_ALIGN64 half aux[16];
+  pstore(aux, from);
+  to[stride*0].x = aux[0].x;
+  to[stride*1].x = aux[1].x;
+  to[stride*2].x = aux[2].x;
+  to[stride*3].x = aux[3].x;
+  to[stride*4].x = aux[4].x;
+  to[stride*5].x = aux[5].x;
+  to[stride*6].x = aux[6].x;
+  to[stride*7].x = aux[7].x;
+  to[stride*8].x = aux[8].x;
+  to[stride*9].x = aux[9].x;
+  to[stride*10].x = aux[10].x;
+  to[stride*11].x = aux[11].x;
+  to[stride*12].x = aux[12].x;
+  to[stride*13].x = aux[13].x;
+  to[stride*14].x = aux[14].x;
+  to[stride*15].x = aux[15].x;
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet16h,16>& kernel) {
+  __m256i a = kernel.packet[0].x;
+  __m256i b = kernel.packet[1].x;
+  __m256i c = kernel.packet[2].x;
+  __m256i d = kernel.packet[3].x;
+  __m256i e = kernel.packet[4].x;
+  __m256i f = kernel.packet[5].x;
+  __m256i g = kernel.packet[6].x;
+  __m256i h = kernel.packet[7].x;
+  __m256i i = kernel.packet[8].x;
+  __m256i j = kernel.packet[9].x;
+  __m256i k = kernel.packet[10].x;
+  __m256i l = kernel.packet[11].x;
+  __m256i m = kernel.packet[12].x;
+  __m256i n = kernel.packet[13].x;
+  __m256i o = kernel.packet[14].x;
+  __m256i p = kernel.packet[15].x;
+
+  __m256i ab_07 = _mm256_unpacklo_epi16(a, b);
+  __m256i cd_07 = _mm256_unpacklo_epi16(c, d);
+  __m256i ef_07 = _mm256_unpacklo_epi16(e, f);
+  __m256i gh_07 = _mm256_unpacklo_epi16(g, h);
+  __m256i ij_07 = _mm256_unpacklo_epi16(i, j);
+  __m256i kl_07 = _mm256_unpacklo_epi16(k, l);
+  __m256i mn_07 = _mm256_unpacklo_epi16(m, n);
+  __m256i op_07 = _mm256_unpacklo_epi16(o, p);
+
+  __m256i ab_8f = _mm256_unpackhi_epi16(a, b);
+  __m256i cd_8f = _mm256_unpackhi_epi16(c, d);
+  __m256i ef_8f = _mm256_unpackhi_epi16(e, f);
+  __m256i gh_8f = _mm256_unpackhi_epi16(g, h);
+  __m256i ij_8f = _mm256_unpackhi_epi16(i, j);
+  __m256i kl_8f = _mm256_unpackhi_epi16(k, l);
+  __m256i mn_8f = _mm256_unpackhi_epi16(m, n);
+  __m256i op_8f = _mm256_unpackhi_epi16(o, p);
+
+  __m256i abcd_03 = _mm256_unpacklo_epi32(ab_07, cd_07);
+  __m256i abcd_47 = _mm256_unpackhi_epi32(ab_07, cd_07);
+  __m256i efgh_03 = _mm256_unpacklo_epi32(ef_07, gh_07);
+  __m256i efgh_47 = _mm256_unpackhi_epi32(ef_07, gh_07);
+  __m256i ijkl_03 = _mm256_unpacklo_epi32(ij_07, kl_07);
+  __m256i ijkl_47 = _mm256_unpackhi_epi32(ij_07, kl_07);
+  __m256i mnop_03 = _mm256_unpacklo_epi32(mn_07, op_07);
+  __m256i mnop_47 = _mm256_unpackhi_epi32(mn_07, op_07);
+
+  __m256i abcd_8b = _mm256_unpacklo_epi32(ab_8f, cd_8f);
+  __m256i abcd_cf = _mm256_unpackhi_epi32(ab_8f, cd_8f);
+  __m256i efgh_8b = _mm256_unpacklo_epi32(ef_8f, gh_8f);
+  __m256i efgh_cf = _mm256_unpackhi_epi32(ef_8f, gh_8f);
+  __m256i ijkl_8b = _mm256_unpacklo_epi32(ij_8f, kl_8f);
+  __m256i ijkl_cf = _mm256_unpackhi_epi32(ij_8f, kl_8f);
+  __m256i mnop_8b = _mm256_unpacklo_epi32(mn_8f, op_8f);
+  __m256i mnop_cf = _mm256_unpackhi_epi32(mn_8f, op_8f);
+
+  __m256i abcdefgh_01 = _mm256_unpacklo_epi64(abcd_03, efgh_03);
+  __m256i abcdefgh_23 = _mm256_unpackhi_epi64(abcd_03, efgh_03);
+  __m256i ijklmnop_01 = _mm256_unpacklo_epi64(ijkl_03, mnop_03);
+  __m256i ijklmnop_23 = _mm256_unpackhi_epi64(ijkl_03, mnop_03);
+  __m256i abcdefgh_45 = _mm256_unpacklo_epi64(abcd_47, efgh_47);
+  __m256i abcdefgh_67 = _mm256_unpackhi_epi64(abcd_47, efgh_47);
+  __m256i ijklmnop_45 = _mm256_unpacklo_epi64(ijkl_47, mnop_47);
+  __m256i ijklmnop_67 = _mm256_unpackhi_epi64(ijkl_47, mnop_47);
+  __m256i abcdefgh_89 = _mm256_unpacklo_epi64(abcd_8b, efgh_8b);
+  __m256i abcdefgh_ab = _mm256_unpackhi_epi64(abcd_8b, efgh_8b);
+  __m256i ijklmnop_89 = _mm256_unpacklo_epi64(ijkl_8b, mnop_8b);
+  __m256i ijklmnop_ab = _mm256_unpackhi_epi64(ijkl_8b, mnop_8b);
+  __m256i abcdefgh_cd = _mm256_unpacklo_epi64(abcd_cf, efgh_cf);
+  __m256i abcdefgh_ef = _mm256_unpackhi_epi64(abcd_cf, efgh_cf);
+  __m256i ijklmnop_cd = _mm256_unpacklo_epi64(ijkl_cf, mnop_cf);
+  __m256i ijklmnop_ef = _mm256_unpackhi_epi64(ijkl_cf, mnop_cf);
+
+  // NOTE: no unpacklo/hi instr in this case, so using permute instr.
+  __m256i a_p_0 = _mm256_permute2x128_si256(abcdefgh_01, ijklmnop_01, 0x20);
+  __m256i a_p_1 = _mm256_permute2x128_si256(abcdefgh_01, ijklmnop_01, 0x31);
+  __m256i a_p_2 = _mm256_permute2x128_si256(abcdefgh_23, ijklmnop_23, 0x20);
+  __m256i a_p_3 = _mm256_permute2x128_si256(abcdefgh_23, ijklmnop_23, 0x31);
+  __m256i a_p_4 = _mm256_permute2x128_si256(abcdefgh_45, ijklmnop_45, 0x20);
+  __m256i a_p_5 = _mm256_permute2x128_si256(abcdefgh_45, ijklmnop_45, 0x31);
+  __m256i a_p_6 = _mm256_permute2x128_si256(abcdefgh_67, ijklmnop_67, 0x20);
+  __m256i a_p_7 = _mm256_permute2x128_si256(abcdefgh_67, ijklmnop_67, 0x31);
+  __m256i a_p_8 = _mm256_permute2x128_si256(abcdefgh_89, ijklmnop_89, 0x20);
+  __m256i a_p_9 = _mm256_permute2x128_si256(abcdefgh_89, ijklmnop_89, 0x31);
+  __m256i a_p_a = _mm256_permute2x128_si256(abcdefgh_ab, ijklmnop_ab, 0x20);
+  __m256i a_p_b = _mm256_permute2x128_si256(abcdefgh_ab, ijklmnop_ab, 0x31);
+  __m256i a_p_c = _mm256_permute2x128_si256(abcdefgh_cd, ijklmnop_cd, 0x20);
+  __m256i a_p_d = _mm256_permute2x128_si256(abcdefgh_cd, ijklmnop_cd, 0x31);
+  __m256i a_p_e = _mm256_permute2x128_si256(abcdefgh_ef, ijklmnop_ef, 0x20);
+  __m256i a_p_f = _mm256_permute2x128_si256(abcdefgh_ef, ijklmnop_ef, 0x31);
+
+  kernel.packet[0].x = a_p_0;
+  kernel.packet[1].x = a_p_1;
+  kernel.packet[2].x = a_p_2;
+  kernel.packet[3].x = a_p_3;
+  kernel.packet[4].x = a_p_4;
+  kernel.packet[5].x = a_p_5;
+  kernel.packet[6].x = a_p_6;
+  kernel.packet[7].x = a_p_7;
+  kernel.packet[8].x = a_p_8;
+  kernel.packet[9].x = a_p_9;
+  kernel.packet[10].x = a_p_a;
+  kernel.packet[11].x = a_p_b;
+  kernel.packet[12].x = a_p_c;
+  kernel.packet[13].x = a_p_d;
+  kernel.packet[14].x = a_p_e;
+  kernel.packet[15].x = a_p_f;
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet16h,8>& kernel) {
+  EIGEN_ALIGN64 half in[8][16];
+  pstore<half>(in[0], kernel.packet[0]);
+  pstore<half>(in[1], kernel.packet[1]);
+  pstore<half>(in[2], kernel.packet[2]);
+  pstore<half>(in[3], kernel.packet[3]);
+  pstore<half>(in[4], kernel.packet[4]);
+  pstore<half>(in[5], kernel.packet[5]);
+  pstore<half>(in[6], kernel.packet[6]);
+  pstore<half>(in[7], kernel.packet[7]);
+
+  EIGEN_ALIGN64 half out[8][16];
+
+  for (int i = 0; i < 8; ++i) {
+    for (int j = 0; j < 8; ++j) {
+      out[i][j] = in[j][2*i];
+    }
+    for (int j = 0; j < 8; ++j) {
+      out[i][j+8] = in[j][2*i+1];
+    }
+  }
+
+  kernel.packet[0] = pload<Packet16h>(out[0]);
+  kernel.packet[1] = pload<Packet16h>(out[1]);
+  kernel.packet[2] = pload<Packet16h>(out[2]);
+  kernel.packet[3] = pload<Packet16h>(out[3]);
+  kernel.packet[4] = pload<Packet16h>(out[4]);
+  kernel.packet[5] = pload<Packet16h>(out[5]);
+  kernel.packet[6] = pload<Packet16h>(out[6]);
+  kernel.packet[7] = pload<Packet16h>(out[7]);
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet16h,4>& kernel) {
+  EIGEN_ALIGN64 half in[4][16];
+  pstore<half>(in[0], kernel.packet[0]);
+  pstore<half>(in[1], kernel.packet[1]);
+  pstore<half>(in[2], kernel.packet[2]);
+  pstore<half>(in[3], kernel.packet[3]);
+
+  EIGEN_ALIGN64 half out[4][16];
+
+  for (int i = 0; i < 4; ++i) {
+    for (int j = 0; j < 4; ++j) {
+      out[i][j] = in[j][4*i];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+4] = in[j][4*i+1];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+8] = in[j][4*i+2];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+12] = in[j][4*i+3];
+    }
+  }
+
+  kernel.packet[0] = pload<Packet16h>(out[0]);
+  kernel.packet[1] = pload<Packet16h>(out[1]);
+  kernel.packet[2] = pload<Packet16h>(out[2]);
+  kernel.packet[3] = pload<Packet16h>(out[3]);
+}
+
+
+#elif defined EIGEN_VECTORIZE_AVX
+
+typedef struct {
+  __m128i x;
+} Packet8h;
+
+
+template<> struct is_arithmetic<Packet8h> { enum { value = true }; };
+
+template <>
+struct packet_traits<Eigen::half> : default_packet_traits {
+  typedef Packet8h type;
+  // There is no half-size packet for Packet8h.
+  typedef Packet8h half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 8,
+    HasHalfPacket = 0,
+    HasAdd    = 0,
+    HasSub    = 0,
+    HasMul    = 0,
+    HasNegate = 0,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasConj   = 0,
+    HasSetLinear = 0,
+    HasDiv = 0,
+    HasSqrt = 0,
+    HasRsqrt = 0,
+    HasExp = 0,
+    HasLog = 0,
+    HasBlend = 0
+  };
+};
+
+
+template<> struct unpacket_traits<Packet8h> { typedef Eigen::half type; enum {size=8, alignment=Aligned16}; typedef Packet8h half; };
+
+template<> EIGEN_STRONG_INLINE Packet8h pset1<Packet8h>(const Eigen::half& from) {
+  Packet8h result;
+  result.x = _mm_set1_epi16(from.x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half pfirst<Packet8h>(const Packet8h& from) {
+  return half_impl::raw_uint16_to_half(static_cast<unsigned short>(_mm_extract_epi16(from.x, 0)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pload<Packet8h>(const Eigen::half* from) {
+  Packet8h result;
+  result.x = _mm_load_si128(reinterpret_cast<const __m128i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h ploadu<Packet8h>(const Eigen::half* from) {
+  Packet8h result;
+  result.x = _mm_loadu_si128(reinterpret_cast<const __m128i*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<Eigen::half>(Eigen::half* to, const Packet8h& from) {
+  _mm_store_si128(reinterpret_cast<__m128i*>(to), from.x);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<Eigen::half>(Eigen::half* to, const Packet8h& from) {
+  _mm_storeu_si128(reinterpret_cast<__m128i*>(to), from.x);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h
+ploadquad<Packet8h>(const Eigen::half* from) {
+  Packet8h result;
+  unsigned short a = from[0].x;
+  unsigned short b = from[1].x;
+  result.x = _mm_set_epi16(b, b, b, b, a, a, a, a);
+  return result;
+}
+
+EIGEN_STRONG_INLINE Packet8f half2float(const Packet8h& a) {
+#ifdef EIGEN_HAS_FP16_C
+  return _mm256_cvtph_ps(a.x);
+#else
+  EIGEN_ALIGN32 Eigen::half aux[8];
+  pstore(aux, a);
+  float f0(aux[0]);
+  float f1(aux[1]);
+  float f2(aux[2]);
+  float f3(aux[3]);
+  float f4(aux[4]);
+  float f5(aux[5]);
+  float f6(aux[6]);
+  float f7(aux[7]);
+
+  return _mm256_set_ps(f7, f6, f5, f4, f3, f2, f1, f0);
+#endif
+}
+
+EIGEN_STRONG_INLINE Packet8h float2half(const Packet8f& a) {
+#ifdef EIGEN_HAS_FP16_C
+  Packet8h result;
+  result.x = _mm256_cvtps_ph(a, _MM_FROUND_TO_NEAREST_INT|_MM_FROUND_NO_EXC);
+  return result;
+#else
+  EIGEN_ALIGN32 float aux[8];
+  pstore(aux, a);
+  Eigen::half h0(aux[0]);
+  Eigen::half h1(aux[1]);
+  Eigen::half h2(aux[2]);
+  Eigen::half h3(aux[3]);
+  Eigen::half h4(aux[4]);
+  Eigen::half h5(aux[5]);
+  Eigen::half h6(aux[6]);
+  Eigen::half h7(aux[7]);
+
+  Packet8h result;
+  result.x = _mm_set_epi16(h7.x, h6.x, h5.x, h4.x, h3.x, h2.x, h1.x, h0.x);
+  return result;
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pconj(const Packet8h& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet8h padd<Packet8h>(const Packet8h& a, const Packet8h& b) {
+  Packet8f af = half2float(a);
+  Packet8f bf = half2float(b);
+  Packet8f rf = padd(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pmul<Packet8h>(const Packet8h& a, const Packet8h& b) {
+  Packet8f af = half2float(a);
+  Packet8f bf = half2float(b);
+  Packet8f rf = pmul(af, bf);
+  return float2half(rf);
+}
+
+template<> EIGEN_STRONG_INLINE Packet8h pgather<Eigen::half, Packet8h>(const Eigen::half* from, Index stride)
+{
+  Packet8h result;
+  result.x = _mm_set_epi16(from[7*stride].x, from[6*stride].x, from[5*stride].x, from[4*stride].x, from[3*stride].x, from[2*stride].x, from[1*stride].x, from[0*stride].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pscatter<Eigen::half, Packet8h>(Eigen::half* to, const Packet8h& from, Index stride)
+{
+  EIGEN_ALIGN32 Eigen::half aux[8];
+  pstore(aux, from);
+  to[stride*0].x = aux[0].x;
+  to[stride*1].x = aux[1].x;
+  to[stride*2].x = aux[2].x;
+  to[stride*3].x = aux[3].x;
+  to[stride*4].x = aux[4].x;
+  to[stride*5].x = aux[5].x;
+  to[stride*6].x = aux[6].x;
+  to[stride*7].x = aux[7].x;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux_max<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux_max<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux_min<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux_min<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half predux_mul<Packet8h>(const Packet8h& a) {
+  Packet8f af = half2float(a);
+  float reduced = predux_mul<Packet8f>(af);
+  return Eigen::half(reduced);
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet8h,8>& kernel) {
+  __m128i a = kernel.packet[0].x;
+  __m128i b = kernel.packet[1].x;
+  __m128i c = kernel.packet[2].x;
+  __m128i d = kernel.packet[3].x;
+  __m128i e = kernel.packet[4].x;
+  __m128i f = kernel.packet[5].x;
+  __m128i g = kernel.packet[6].x;
+  __m128i h = kernel.packet[7].x;
+
+  __m128i a03b03 = _mm_unpacklo_epi16(a, b);
+  __m128i c03d03 = _mm_unpacklo_epi16(c, d);
+  __m128i e03f03 = _mm_unpacklo_epi16(e, f);
+  __m128i g03h03 = _mm_unpacklo_epi16(g, h);
+  __m128i a47b47 = _mm_unpackhi_epi16(a, b);
+  __m128i c47d47 = _mm_unpackhi_epi16(c, d);
+  __m128i e47f47 = _mm_unpackhi_epi16(e, f);
+  __m128i g47h47 = _mm_unpackhi_epi16(g, h);
+
+  __m128i a01b01c01d01 = _mm_unpacklo_epi32(a03b03, c03d03);
+  __m128i a23b23c23d23 = _mm_unpackhi_epi32(a03b03, c03d03);
+  __m128i e01f01g01h01 = _mm_unpacklo_epi32(e03f03, g03h03);
+  __m128i e23f23g23h23 = _mm_unpackhi_epi32(e03f03, g03h03);
+  __m128i a45b45c45d45 = _mm_unpacklo_epi32(a47b47, c47d47);
+  __m128i a67b67c67d67 = _mm_unpackhi_epi32(a47b47, c47d47);
+  __m128i e45f45g45h45 = _mm_unpacklo_epi32(e47f47, g47h47);
+  __m128i e67f67g67h67 = _mm_unpackhi_epi32(e47f47, g47h47);
+
+  __m128i a0b0c0d0e0f0g0h0 = _mm_unpacklo_epi64(a01b01c01d01, e01f01g01h01);
+  __m128i a1b1c1d1e1f1g1h1 = _mm_unpackhi_epi64(a01b01c01d01, e01f01g01h01);
+  __m128i a2b2c2d2e2f2g2h2 = _mm_unpacklo_epi64(a23b23c23d23, e23f23g23h23);
+  __m128i a3b3c3d3e3f3g3h3 = _mm_unpackhi_epi64(a23b23c23d23, e23f23g23h23);
+  __m128i a4b4c4d4e4f4g4h4 = _mm_unpacklo_epi64(a45b45c45d45, e45f45g45h45);
+  __m128i a5b5c5d5e5f5g5h5 = _mm_unpackhi_epi64(a45b45c45d45, e45f45g45h45);
+  __m128i a6b6c6d6e6f6g6h6 = _mm_unpacklo_epi64(a67b67c67d67, e67f67g67h67);
+  __m128i a7b7c7d7e7f7g7h7 = _mm_unpackhi_epi64(a67b67c67d67, e67f67g67h67);
+
+  kernel.packet[0].x = a0b0c0d0e0f0g0h0;
+  kernel.packet[1].x = a1b1c1d1e1f1g1h1;
+  kernel.packet[2].x = a2b2c2d2e2f2g2h2;
+  kernel.packet[3].x = a3b3c3d3e3f3g3h3;
+  kernel.packet[4].x = a4b4c4d4e4f4g4h4;
+  kernel.packet[5].x = a5b5c5d5e5f5g5h5;
+  kernel.packet[6].x = a6b6c6d6e6f6g6h6;
+  kernel.packet[7].x = a7b7c7d7e7f7g7h7;
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet8h,4>& kernel) {
+  EIGEN_ALIGN32 Eigen::half in[4][8];
+  pstore<Eigen::half>(in[0], kernel.packet[0]);
+  pstore<Eigen::half>(in[1], kernel.packet[1]);
+  pstore<Eigen::half>(in[2], kernel.packet[2]);
+  pstore<Eigen::half>(in[3], kernel.packet[3]);
+
+  EIGEN_ALIGN32 Eigen::half out[4][8];
+
+  for (int i = 0; i < 4; ++i) {
+    for (int j = 0; j < 4; ++j) {
+      out[i][j] = in[j][2*i];
+    }
+    for (int j = 0; j < 4; ++j) {
+      out[i][j+4] = in[j][2*i+1];
+    }
+  }
+
+  kernel.packet[0] = pload<Packet8h>(out[0]);
+  kernel.packet[1] = pload<Packet8h>(out[1]);
+  kernel.packet[2] = pload<Packet8h>(out[2]);
+  kernel.packet[3] = pload<Packet8h>(out[3]);
+}
+
+
+// Disable the following code since it's broken on too many platforms / compilers.
+//#elif defined(EIGEN_VECTORIZE_SSE) && (!EIGEN_ARCH_x86_64) && (!EIGEN_COMP_MSVC)
+#elif 0
+
+typedef struct {
+  __m64 x;
+} Packet4h;
+
+
+template<> struct is_arithmetic<Packet4h> { enum { value = true }; };
+
+template <>
+struct packet_traits<Eigen::half> : default_packet_traits {
+  typedef Packet4h type;
+  // There is no half-size packet for Packet4h.
+  typedef Packet4h half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 4,
+    HasHalfPacket = 0,
+    HasAdd    = 0,
+    HasSub    = 0,
+    HasMul    = 0,
+    HasNegate = 0,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasConj   = 0,
+    HasSetLinear = 0,
+    HasDiv = 0,
+    HasSqrt = 0,
+    HasRsqrt = 0,
+    HasExp = 0,
+    HasLog = 0,
+    HasBlend = 0
+  };
+};
+
+
+template<> struct unpacket_traits<Packet4h> { typedef Eigen::half type; enum {size=4, alignment=Aligned16}; typedef Packet4h half; };
+
+template<> EIGEN_STRONG_INLINE Packet4h pset1<Packet4h>(const Eigen::half& from) {
+  Packet4h result;
+  result.x = _mm_set1_pi16(from.x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Eigen::half pfirst<Packet4h>(const Packet4h& from) {
+  return half_impl::raw_uint16_to_half(static_cast<unsigned short>(_mm_cvtsi64_si32(from.x)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pconj(const Packet4h& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet4h padd<Packet4h>(const Packet4h& a, const Packet4h& b) {
+  __int64_t a64 = _mm_cvtm64_si64(a.x);
+  __int64_t b64 = _mm_cvtm64_si64(b.x);
+
+  Eigen::half h[4];
+
+  Eigen::half ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64));
+  Eigen::half hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64));
+  h[0] = ha + hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 16));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 16));
+  h[1] = ha + hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 32));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 32));
+  h[2] = ha + hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 48));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 48));
+  h[3] = ha + hb;
+  Packet4h result;
+  result.x = _mm_set_pi16(h[3].x, h[2].x, h[1].x, h[0].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pmul<Packet4h>(const Packet4h& a, const Packet4h& b) {
+  __int64_t a64 = _mm_cvtm64_si64(a.x);
+  __int64_t b64 = _mm_cvtm64_si64(b.x);
+
+  Eigen::half h[4];
+
+  Eigen::half ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64));
+  Eigen::half hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64));
+  h[0] = ha * hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 16));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 16));
+  h[1] = ha * hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 32));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 32));
+  h[2] = ha * hb;
+  ha = half_impl::raw_uint16_to_half(static_cast<unsigned short>(a64 >> 48));
+  hb = half_impl::raw_uint16_to_half(static_cast<unsigned short>(b64 >> 48));
+  h[3] = ha * hb;
+  Packet4h result;
+  result.x = _mm_set_pi16(h[3].x, h[2].x, h[1].x, h[0].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pload<Packet4h>(const Eigen::half* from) {
+  Packet4h result;
+  result.x = _mm_cvtsi64_m64(*reinterpret_cast<const __int64_t*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h ploadu<Packet4h>(const Eigen::half* from) {
+  Packet4h result;
+  result.x = _mm_cvtsi64_m64(*reinterpret_cast<const __int64_t*>(from));
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<Eigen::half>(Eigen::half* to, const Packet4h& from) {
+  __int64_t r = _mm_cvtm64_si64(from.x);
+  *(reinterpret_cast<__int64_t*>(to)) = r;
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<Eigen::half>(Eigen::half* to, const Packet4h& from) {
+  __int64_t r = _mm_cvtm64_si64(from.x);
+  *(reinterpret_cast<__int64_t*>(to)) = r;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h
+ploadquad<Packet4h>(const Eigen::half* from) {
+  return pset1<Packet4h>(*from);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4h pgather<Eigen::half, Packet4h>(const Eigen::half* from, Index stride)
+{
+  Packet4h result;
+  result.x = _mm_set_pi16(from[3*stride].x, from[2*stride].x, from[1*stride].x, from[0*stride].x);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE void pscatter<Eigen::half, Packet4h>(Eigen::half* to, const Packet4h& from, Index stride)
+{
+  __int64_t a = _mm_cvtm64_si64(from.x);
+  to[stride*0].x = static_cast<unsigned short>(a);
+  to[stride*1].x = static_cast<unsigned short>(a >> 16);
+  to[stride*2].x = static_cast<unsigned short>(a >> 32);
+  to[stride*3].x = static_cast<unsigned short>(a >> 48);
+}
+
+EIGEN_STRONG_INLINE void
+ptranspose(PacketBlock<Packet4h,4>& kernel) {
+  __m64 T0 = _mm_unpacklo_pi16(kernel.packet[0].x, kernel.packet[1].x);
+  __m64 T1 = _mm_unpacklo_pi16(kernel.packet[2].x, kernel.packet[3].x);
+  __m64 T2 = _mm_unpackhi_pi16(kernel.packet[0].x, kernel.packet[1].x);
+  __m64 T3 = _mm_unpackhi_pi16(kernel.packet[2].x, kernel.packet[3].x);
+
+  kernel.packet[0].x = _mm_unpacklo_pi32(T0, T1);
+  kernel.packet[1].x = _mm_unpackhi_pi32(T0, T1);
+  kernel.packet[2].x = _mm_unpacklo_pi32(T2, T3);
+  kernel.packet[3].x = _mm_unpackhi_pi32(T2, T3);
+}
+
+#endif
+
+}
+}
+
+#endif // EIGEN_PACKET_MATH_HALF_CUDA_H
diff --git a/Eigen/src/Core/arch/CUDA/TypeCasting.h b/Eigen/src/Core/arch/CUDA/TypeCasting.h
new file mode 100644
index 0000000..aa5fbce
--- /dev/null
+++ b/Eigen/src/Core/arch/CUDA/TypeCasting.h
@@ -0,0 +1,212 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TYPE_CASTING_CUDA_H
+#define EIGEN_TYPE_CASTING_CUDA_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<>
+struct scalar_cast_op<float, Eigen::half> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef Eigen::half result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const float& a) const {
+    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+      return __float2half(a);
+    #else
+      return Eigen::half(a);
+    #endif
+  }
+};
+
+template<>
+struct functor_traits<scalar_cast_op<float, Eigen::half> >
+{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
+
+
+template<>
+struct scalar_cast_op<int, Eigen::half> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef Eigen::half result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Eigen::half operator() (const int& a) const {
+    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+      return __float2half(static_cast<float>(a));
+    #else
+      return Eigen::half(static_cast<float>(a));
+    #endif
+  }
+};
+
+template<>
+struct functor_traits<scalar_cast_op<int, Eigen::half> >
+{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
+
+
+template<>
+struct scalar_cast_op<Eigen::half, float> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef float result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float operator() (const Eigen::half& a) const {
+    #if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+      return __half2float(a);
+    #else
+      return static_cast<float>(a);
+    #endif
+  }
+};
+
+template<>
+struct functor_traits<scalar_cast_op<Eigen::half, float> >
+{ enum { Cost = NumTraits<float>::AddCost, PacketAccess = false }; };
+
+
+
+#if defined(EIGEN_HAS_CUDA_FP16) && defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 300
+
+template <>
+struct type_casting_traits<Eigen::half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 2,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE float4 pcast<half2, float4>(const half2& a, const half2& b) {
+  float2 r1 = __half22float2(a);
+  float2 r2 = __half22float2(b);
+  return make_float4(r1.x, r1.y, r2.x, r2.y);
+}
+
+template <>
+struct type_casting_traits<float, Eigen::half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 2
+  };
+};
+
+template<> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE half2 pcast<float4, half2>(const float4& a) {
+  // Simply discard the second half of the input
+  return __floats2half2_rn(a.x, a.y);
+}
+
+#elif defined EIGEN_VECTORIZE_AVX512
+template <>
+struct type_casting_traits<half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet16f pcast<Packet16h, Packet16f>(const Packet16h& a) {
+  return half2float(a);
+}
+
+template <>
+struct type_casting_traits<float, half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet16h pcast<Packet16f, Packet16h>(const Packet16f& a) {
+  return float2half(a);
+}
+
+#elif defined EIGEN_VECTORIZE_AVX
+
+template <>
+struct type_casting_traits<Eigen::half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet8f pcast<Packet8h, Packet8f>(const Packet8h& a) {
+  return half2float(a);
+}
+
+template <>
+struct type_casting_traits<float, Eigen::half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f& a) {
+  return float2half(a);
+}
+
+// Disable the following code since it's broken on too many platforms / compilers.
+//#elif defined(EIGEN_VECTORIZE_SSE) && (!EIGEN_ARCH_x86_64) && (!EIGEN_COMP_MSVC)
+#elif 0
+
+template <>
+struct type_casting_traits<Eigen::half, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4h, Packet4f>(const Packet4h& a) {
+  __int64_t a64 = _mm_cvtm64_si64(a.x);
+  Eigen::half h = raw_uint16_to_half(static_cast<unsigned short>(a64));
+  float f1 = static_cast<float>(h);
+  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 16));
+  float f2 = static_cast<float>(h);
+  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 32));
+  float f3 = static_cast<float>(h);
+  h = raw_uint16_to_half(static_cast<unsigned short>(a64 >> 48));
+  float f4 = static_cast<float>(h);
+  return _mm_set_ps(f4, f3, f2, f1);
+}
+
+template <>
+struct type_casting_traits<float, Eigen::half> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template<> EIGEN_STRONG_INLINE Packet4h pcast<Packet4f, Packet4h>(const Packet4f& a) {
+  EIGEN_ALIGN16 float aux[4];
+  pstore(aux, a);
+  Eigen::half h0(aux[0]);
+  Eigen::half h1(aux[1]);
+  Eigen::half h2(aux[2]);
+  Eigen::half h3(aux[3]);
+
+  Packet4h result;
+  result.x = _mm_set_pi16(h3.x, h2.x, h1.x, h0.x);
+  return result;
+}
+
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TYPE_CASTING_CUDA_H
diff --git a/Eigen/src/Core/arch/Default/CMakeLists.txt b/Eigen/src/Core/arch/Default/CMakeLists.txt
deleted file mode 100644
index 339c091..0000000
--- a/Eigen/src/Core/arch/Default/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Core_arch_Default_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Core_arch_Default_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core/arch/Default COMPONENT Devel
-)
diff --git a/Eigen/src/Core/arch/Default/ConjHelper.h b/Eigen/src/Core/arch/Default/ConjHelper.h
new file mode 100644
index 0000000..4cfe34e
--- /dev/null
+++ b/Eigen/src/Core/arch/Default/ConjHelper.h
@@ -0,0 +1,29 @@
+
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2017 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ARCH_CONJ_HELPER_H
+#define EIGEN_ARCH_CONJ_HELPER_H
+
+#define EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(PACKET_CPLX, PACKET_REAL)                                                          \
+  template<> struct conj_helper<PACKET_REAL, PACKET_CPLX, false,false> {                                          \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmadd(const PACKET_REAL& x, const PACKET_CPLX& y, const PACKET_CPLX& c) const \
+    { return padd(c, pmul(x,y)); }                                                                                \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmul(const PACKET_REAL& x, const PACKET_CPLX& y) const                        \
+    { return PACKET_CPLX(Eigen::internal::pmul<PACKET_REAL>(x, y.v)); }                                           \
+  };                                                                                                              \
+                                                                                                                  \
+  template<> struct conj_helper<PACKET_CPLX, PACKET_REAL, false,false> {                                          \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmadd(const PACKET_CPLX& x, const PACKET_REAL& y, const PACKET_CPLX& c) const \
+    { return padd(c, pmul(x,y)); }                                                                                \
+    EIGEN_STRONG_INLINE PACKET_CPLX pmul(const PACKET_CPLX& x, const PACKET_REAL& y) const                        \
+    { return PACKET_CPLX(Eigen::internal::pmul<PACKET_REAL>(x.v, y)); }                                           \
+  };
+
+#endif // EIGEN_ARCH_CONJ_HELPER_H
diff --git a/Eigen/src/Core/arch/NEON/CMakeLists.txt b/Eigen/src/Core/arch/NEON/CMakeLists.txt
deleted file mode 100644
index fd4d4af..0000000
--- a/Eigen/src/Core/arch/NEON/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Core_arch_NEON_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Core_arch_NEON_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core/arch/NEON COMPONENT Devel
-)
diff --git a/Eigen/src/Core/arch/NEON/Complex.h b/Eigen/src/Core/arch/NEON/Complex.h
index 8d9255e..306a309 100644
--- a/Eigen/src/Core/arch/NEON/Complex.h
+++ b/Eigen/src/Core/arch/NEON/Complex.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2010 Konstantinos Margaritis <markos@freevec.org>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -14,8 +15,21 @@
 
 namespace internal {
 
-static uint32x4_t p4ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET4(0x00000000, 0x80000000, 0x00000000, 0x80000000);
-static uint32x2_t p2ui_CONJ_XOR = EIGEN_INIT_NEON_PACKET2(0x00000000, 0x80000000);
+inline uint32x4_t p4ui_CONJ_XOR() {
+// See bug 1325, clang fails to call vld1q_u64.
+#if EIGEN_COMP_CLANG
+  uint32x4_t ret = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+  return ret;
+#else
+  static const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000, 0x00000000, 0x80000000 };
+  return vld1q_u32( conj_XOR_DATA );
+#endif
+}
+
+inline uint32x2_t p2ui_CONJ_XOR() {
+  static const uint32_t conj_XOR_DATA[] = { 0x00000000, 0x80000000 };
+  return vld1_u32( conj_XOR_DATA );
+}
 
 //---------- float ----------
 struct Packet2cf
@@ -28,10 +42,12 @@
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet2cf type;
+  typedef Packet2cf half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 2,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -46,12 +62,12 @@
   };
 };
 
-template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
 
 template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
 {
   float32x2_t r64;
-  r64 = vld1_f32((float *)&from);
+  r64 = vld1_f32((const float *)&from);
 
   return Packet2cf(vcombine_f32(r64, r64));
 }
@@ -62,7 +78,7 @@
 template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
 {
   Packet4ui b = vreinterpretq_u32_f32(a.v);
-  return Packet2cf(vreinterpretq_f32_u32(veorq_u32(b, p4ui_CONJ_XOR)));
+  return Packet2cf(vreinterpretq_f32_u32(veorq_u32(b, p4ui_CONJ_XOR())));
 }
 
 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
@@ -71,14 +87,14 @@
 
   // Get the real values of a | a1_re | a1_re | a2_re | a2_re |
   v1 = vcombine_f32(vdup_lane_f32(vget_low_f32(a.v), 0), vdup_lane_f32(vget_high_f32(a.v), 0));
-  // Get the real values of a | a1_im | a1_im | a2_im | a2_im |
+  // Get the imag values of a | a1_im | a1_im | a2_im | a2_im |
   v2 = vcombine_f32(vdup_lane_f32(vget_low_f32(a.v), 1), vdup_lane_f32(vget_high_f32(a.v), 1));
   // Multiply the real a with b
   v1 = vmulq_f32(v1, b.v);
   // Multiply the imag a with b
   v2 = vmulq_f32(v2, b.v);
   // Conjugate v2 
-  v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), p4ui_CONJ_XOR));
+  v2 = vreinterpretq_f32_u32(veorq_u32(vreinterpretq_u32_f32(v2), p4ui_CONJ_XOR()));
   // Swap real/imag elements in v2.
   v2 = vrev64q_f32(v2);
   // Add and return the result
@@ -87,7 +103,7 @@
 
 template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
-  return Packet2cf(vreinterpretq_f32_u32(vorrq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
+  return Packet2cf(vreinterpretq_f32_u32(vandq_u32(vreinterpretq_u32_f32(a.v),vreinterpretq_u32_f32(b.v))));
 }
 template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
@@ -110,7 +126,23 @@
 template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
 template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { EIGEN_ARM_PREFETCH((float *)addr); }
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  Packet4f res = pset1<Packet4f>(0.f);
+  res = vsetq_lane_f32(std::real(from[0*stride]), res, 0);
+  res = vsetq_lane_f32(std::imag(from[0*stride]), res, 1);
+  res = vsetq_lane_f32(std::real(from[1*stride]), res, 2);
+  res = vsetq_lane_f32(std::imag(from[1*stride]), res, 3);
+  return Packet2cf(res);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  to[stride*0] = std::complex<float>(vgetq_lane_f32(from.v, 0), vgetq_lane_f32(from.v, 1));
+  to[stride*1] = std::complex<float>(vgetq_lane_f32(from.v, 2), vgetq_lane_f32(from.v, 3));
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { EIGEN_ARM_PREFETCH((const float *)addr); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
 {
@@ -177,7 +209,7 @@
   // Multiply the imag a with b
   v2 = vmul_f32(v2, a2);
   // Conjugate v2 
-  v2 = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(v2), p2ui_CONJ_XOR));
+  v2 = vreinterpret_f32_u32(veor_u32(vreinterpret_u32_f32(v2), p2ui_CONJ_XOR()));
   // Swap real/imag elements in v2.
   v2 = vrev64_f32(v2);
   // Add v1, v2
@@ -233,9 +265,11 @@
   }
 };
 
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
+
 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
-  // TODO optimize it for AltiVec
+  // TODO optimize it for NEON
   Packet2cf res = conj_helper<Packet2cf,Packet2cf,false,true>().pmul(a,b);
   Packet4f s, rev_s;
 
@@ -243,9 +277,212 @@
   s = vmulq_f32(b.v, b.v);
   rev_s = vrev64q_f32(s);
 
-  return Packet2cf(pdiv(res.v, vaddq_f32(s,rev_s)));
+  return Packet2cf(pdiv<Packet4f>(res.v, vaddq_f32(s,rev_s)));
 }
 
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2cf,2>& kernel) {
+  Packet4f tmp = vcombine_f32(vget_high_f32(kernel.packet[0].v), vget_high_f32(kernel.packet[1].v));
+  kernel.packet[0].v = vcombine_f32(vget_low_f32(kernel.packet[0].v), vget_low_f32(kernel.packet[1].v));
+  kernel.packet[1].v = tmp;
+}
+
+//---------- double ----------
+#if EIGEN_ARCH_ARM64 && !EIGEN_APPLE_DOUBLE_NEON_BUG
+
+// See bug 1325, clang fails to call vld1q_u64.
+#if EIGEN_COMP_CLANG
+  static uint64x2_t p2ul_CONJ_XOR = {0x0, 0x8000000000000000};
+#else
+  const uint64_t  p2ul_conj_XOR_DATA[] = { 0x0, 0x8000000000000000 };
+  static uint64x2_t p2ul_CONJ_XOR = vld1q_u64( p2ul_conj_XOR_DATA );
+#endif
+
+struct Packet1cd
+{
+  EIGEN_STRONG_INLINE Packet1cd() {}
+  EIGEN_STRONG_INLINE explicit Packet1cd(const Packet2d& a) : v(a) {}
+  Packet2d v;
+};
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet1cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 0,
+    size = 1,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
+
+template<> EIGEN_STRONG_INLINE Packet1cd pload<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
+{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(padd<Packet2d>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(psub<Packet2d>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate<Packet2d>(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd(vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a.v), p2ul_CONJ_XOR))); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  Packet2d v1, v2;
+
+  // Get the real values of a 
+  v1 = vdupq_lane_f64(vget_low_f64(a.v), 0);
+  // Get the imag values of a
+  v2 = vdupq_lane_f64(vget_high_f64(a.v), 0);
+  // Multiply the real a with b
+  v1 = vmulq_f64(v1, b.v);
+  // Multiply the imag a with b
+  v2 = vmulq_f64(v2, b.v);
+  // Conjugate v2 
+  v2 = vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(v2), p2ul_CONJ_XOR));
+  // Swap real/imag elements in v2.
+  v2 = preverse<Packet2d>(v2);
+  // Add and return the result
+  return Packet1cd(vaddq_f64(v1, v2));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(vandq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(vorrq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  return Packet1cd(vreinterpretq_f64_u64(vbicq_u64(vreinterpretq_u64_f64(a.v),vreinterpretq_u64_f64(b.v))));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
+
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { EIGEN_ARM_PREFETCH((const double *)addr); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride)
+{
+  Packet2d res = pset1<Packet2d>(0.0);
+  res = vsetq_lane_f64(std::real(from[0*stride]), res, 0);
+  res = vsetq_lane_f64(std::imag(from[0*stride]), res, 1);
+  return Packet1cd(res);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride)
+{
+  to[stride*0] = std::complex<double>(vgetq_lane_f64(from.v, 0), vgetq_lane_f64(from.v, 1));
+}
+
+
+template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
+{
+  std::complex<double> EIGEN_ALIGN16 res;
+  pstore<std::complex<double> >(&res, a);
+
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs) { return vecs[0]; }
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a) { return pfirst(a); }
+
+template<int Offset>
+struct palign_impl<Offset,Packet1cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/)
+  {
+    // FIXME is it sure we never have to align a Packet1cd?
+    // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary...
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
+
+template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  // TODO optimize it for NEON
+  Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
+  Packet2d s = pmul<Packet2d>(b.v, b.v);
+  Packet2d rev_s = preverse<Packet2d>(s);
+
+  return Packet1cd(pdiv(res.v, padd<Packet2d>(s,rev_s)));
+}
+
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+{
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
+{
+  Packet2d tmp = vcombine_f64(vget_high_f64(kernel.packet[0].v), vget_high_f64(kernel.packet[1].v));
+  kernel.packet[0].v = vcombine_f64(vget_low_f64(kernel.packet[0].v), vget_low_f64(kernel.packet[1].v));
+  kernel.packet[1].v = tmp;
+}
+#endif // EIGEN_ARCH_ARM64
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/arch/NEON/MathFunctions.h b/Eigen/src/Core/arch/NEON/MathFunctions.h
new file mode 100644
index 0000000..6bb05bb
--- /dev/null
+++ b/Eigen/src/Core/arch/NEON/MathFunctions.h
@@ -0,0 +1,91 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+/* The sin, cos, exp, and log functions of this file come from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+#ifndef EIGEN_MATH_FUNCTIONS_NEON_H
+#define EIGEN_MATH_FUNCTIONS_NEON_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pexp<Packet4f>(const Packet4f& _x)
+{
+  Packet4f x = _x;
+  Packet4f tmp, fx;
+
+  _EIGEN_DECLARE_CONST_Packet4f(1 , 1.0f);
+  _EIGEN_DECLARE_CONST_Packet4f(half, 0.5f);
+  _EIGEN_DECLARE_CONST_Packet4i(0x7f, 0x7f);
+  _EIGEN_DECLARE_CONST_Packet4f(exp_hi,  88.3762626647950f);
+  _EIGEN_DECLARE_CONST_Packet4f(exp_lo, -88.3762626647949f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_LOG2EF, 1.44269504088896341f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C1, 0.693359375f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_C2, -2.12194440e-4f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p0, 1.9875691500E-4f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p1, 1.3981999507E-3f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p2, 8.3334519073E-3f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p3, 4.1665795894E-2f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p4, 1.6666665459E-1f);
+  _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p5, 5.0000001201E-1f);
+
+  x = vminq_f32(x, p4f_exp_hi);
+  x = vmaxq_f32(x, p4f_exp_lo);
+
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = vmlaq_f32(p4f_half, x, p4f_cephes_LOG2EF);
+
+  /* perform a floorf */
+  tmp = vcvtq_f32_s32(vcvtq_s32_f32(fx));
+
+  /* if greater, substract 1 */
+  Packet4ui mask = vcgtq_f32(tmp, fx);
+  mask = vandq_u32(mask, vreinterpretq_u32_f32(p4f_1));
+
+  fx = vsubq_f32(tmp, vreinterpretq_f32_u32(mask));
+
+  tmp = vmulq_f32(fx, p4f_cephes_exp_C1);
+  Packet4f z = vmulq_f32(fx, p4f_cephes_exp_C2);
+  x = vsubq_f32(x, tmp);
+  x = vsubq_f32(x, z);
+
+  Packet4f y = vmulq_f32(p4f_cephes_exp_p0, x);
+  z = vmulq_f32(x, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p1);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p2);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p3);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p4);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, p4f_cephes_exp_p5);
+
+  y = vmulq_f32(y, z);
+  y = vaddq_f32(y, x);
+  y = vaddq_f32(y, p4f_1);
+
+  /* build 2^n */
+  int32x4_t mm;
+  mm = vcvtq_s32_f32(fx);
+  mm = vaddq_s32(mm, p4i_0x7f);
+  mm = vshlq_n_s32(mm, 23);
+  Packet4f pow2n = vreinterpretq_f32_s32(mm);
+
+  y = vmulq_f32(y, pow2n);
+  return y;
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_MATH_FUNCTIONS_NEON_H
diff --git a/Eigen/src/Core/arch/NEON/PacketMath.h b/Eigen/src/Core/arch/NEON/PacketMath.h
index 94dfab3..3d5ed0d 100644
--- a/Eigen/src/Core/arch/NEON/PacketMath.h
+++ b/Eigen/src/Core/arch/NEON/PacketMath.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2010 Konstantinos Margaritis <markos@codex.gr>
+// Copyright (C) 2010 Konstantinos Margaritis <markos@freevec.org>
 // Heavily based on Gael's SSE version.
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -20,43 +20,79 @@
 #define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 8
 #endif
 
-// FIXME NEON has 16 quad registers, but since the current register allocator
-// is so bad, it is much better to reduce it to 8
-#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
-#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 8
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
 #endif
 
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#endif
+
+#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#if EIGEN_ARCH_ARM64
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 32
+#else
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS 16 
+#endif
+#endif
+
+#if EIGEN_COMP_MSVC
+
+// In MSVC's arm_neon.h header file, all NEON vector types
+// are aliases to the same underlying type __n128.
+// We thus have to wrap them to make them different C++ types.
+// (See also bug 1428)
+
+template<typename T,int unique_id>
+struct eigen_packet_wrapper
+{
+  operator T&() { return m_val; }
+  operator const T&() const { return m_val; }
+  eigen_packet_wrapper() {}
+  eigen_packet_wrapper(const T &v) : m_val(v) {}
+  eigen_packet_wrapper& operator=(const T &v) {
+    m_val = v;
+    return *this;
+  }
+
+  T m_val;
+};
+typedef eigen_packet_wrapper<float32x2_t,0> Packet2f;
+typedef eigen_packet_wrapper<float32x4_t,1> Packet4f;
+typedef eigen_packet_wrapper<int32x4_t  ,2> Packet4i;
+typedef eigen_packet_wrapper<int32x2_t  ,3> Packet2i;
+typedef eigen_packet_wrapper<uint32x4_t ,4> Packet4ui;
+
+#else
+
+typedef float32x2_t Packet2f;
 typedef float32x4_t Packet4f;
 typedef int32x4_t   Packet4i;
+typedef int32x2_t   Packet2i;
 typedef uint32x4_t  Packet4ui;
 
+#endif // EIGEN_COMP_MSVC
+
 #define _EIGEN_DECLARE_CONST_Packet4f(NAME,X) \
   const Packet4f p4f_##NAME = pset1<Packet4f>(X)
 
 #define _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(NAME,X) \
-  const Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int>(X))
+  const Packet4f p4f_##NAME = vreinterpretq_f32_u32(pset1<int32_t>(X))
 
 #define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
   const Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
-#if defined(__llvm__) && !defined(__clang__)
-  //Special treatment for Apple's llvm-gcc, its NEON packet types are unions
-  #define EIGEN_INIT_NEON_PACKET2(X, Y)       {{X, Y}}
-  #define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {{X, Y, Z, W}}
-#else
-  //Default initializer for packets
-  #define EIGEN_INIT_NEON_PACKET2(X, Y)       {X, Y}
-  #define EIGEN_INIT_NEON_PACKET4(X, Y, Z, W) {X, Y, Z, W}
-#endif
-
-// arm64 does have the pld instruction. If available, let's trust the __builtin_prefetch built-in function
-// which available on LLVM and GCC (at least)
-#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || defined(__GNUC__)
+#if EIGEN_ARCH_ARM64
+  // __builtin_prefetch tends to do nothing on ARM64 compilers because the
+  // prefetch instructions there are too detailed for __builtin_prefetch to map
+  // meaningfully to them.
+  #define EIGEN_ARM_PREFETCH(ADDR)  __asm__ __volatile__("prfm pldl1keep, [%[addr]]\n" ::[addr] "r"(ADDR) : );
+#elif EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
   #define EIGEN_ARM_PREFETCH(ADDR) __builtin_prefetch(ADDR);
 #elif defined __pld
   #define EIGEN_ARM_PREFETCH(ADDR) __pld(ADDR)
-#elif !defined(__aarch64__)
-  #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ( "   pld [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
+#elif EIGEN_ARCH_ARM32
+  #define EIGEN_ARM_PREFETCH(ADDR) __asm__ __volatile__ ("pld [%[addr]]\n" :: [addr] "r" (ADDR) : );
 #else
   // by default no explicit prefetching
   #define EIGEN_ARM_PREFETCH(ADDR)
@@ -65,53 +101,60 @@
 template<> struct packet_traits<float>  : default_packet_traits
 {
   typedef Packet4f type;
+  typedef Packet4f half; // Packet2f intrinsics not implemented yet
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 4,
+    HasHalfPacket=0, // Packet2f intrinsics not implemented yet
    
     HasDiv  = 1,
     // FIXME check the Has*
     HasSin  = 0,
     HasCos  = 0,
     HasLog  = 0,
-    HasExp  = 0,
+    HasExp  = 1,
     HasSqrt = 0
   };
 };
-template<> struct packet_traits<int>    : default_packet_traits
+template<> struct packet_traits<int32_t>    : default_packet_traits
 {
   typedef Packet4i type;
+  typedef Packet4i half; // Packet2i intrinsics not implemented yet
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4
+    size=4,
+    HasHalfPacket=0 // Packet2i intrinsics not implemented yet
     // FIXME check the Has*
   };
 };
 
-#if EIGEN_GNUC_AT_MOST(4,4) && !defined(__llvm__)
+#if EIGEN_GNUC_AT_MOST(4,4) && !EIGEN_COMP_LLVM
 // workaround gcc 4.2, 4.3 and 4.4 compilatin issue
 EIGEN_STRONG_INLINE float32x4_t vld1q_f32(const float* x) { return ::vld1q_f32((const float32_t*)x); }
 EIGEN_STRONG_INLINE float32x2_t vld1_f32 (const float* x) { return ::vld1_f32 ((const float32_t*)x); }
+EIGEN_STRONG_INLINE float32x2_t vld1_dup_f32 (const float* x) { return ::vld1_dup_f32 ((const float32_t*)x); }
 EIGEN_STRONG_INLINE void        vst1q_f32(float* to, float32x4_t from) { ::vst1q_f32((float32_t*)to,from); }
 EIGEN_STRONG_INLINE void        vst1_f32 (float* to, float32x2_t from) { ::vst1_f32 ((float32_t*)to,from); }
 #endif
 
-template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4}; };
-template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4}; };
+template<> struct unpacket_traits<Packet4f> { typedef float   type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet4i> { typedef int32_t type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
 
 template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { return vdupq_n_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)   { return vdupq_n_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int32_t&    from)   { return vdupq_n_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a)
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a)
 {
-  Packet4f countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
+  const float f[] = {0, 1, 2, 3};
+  Packet4f countdown = vld1q_f32(f);
   return vaddq_f32(pset1<Packet4f>(a), countdown);
 }
-template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a)
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int32_t& a)
 {
-  Packet4i countdown = EIGEN_INIT_NEON_PACKET4(0, 1, 2, 3);
+  const int32_t i[] = {0, 1, 2, 3};
+  Packet4i countdown = vld1q_s32(i);
   return vaddq_s32(pset1<Packet4i>(a), countdown);
 }
 
@@ -132,6 +175,9 @@
 
 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
 {
+#if EIGEN_ARCH_ARM64
+  return vdivq_f32(a,b);
+#else
   Packet4f inv, restep, div;
 
   // NEON does not offer a divide instruction, we have to do a reciprocal approximation
@@ -150,14 +196,51 @@
   div = vmulq_f32(a, inv);
 
   return div;
+#endif
 }
+
 template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& /*a*/, const Packet4i& /*b*/)
 { eigen_assert(false && "packet integer division are not supported by NEON");
   return pset1<Packet4i>(0);
 }
 
-// for some weird raisons, it has to be overloaded for packet of integers
-template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vmlaq_f32(c,a,b); }
+// Clang/ARM wrongly advertises __ARM_FEATURE_FMA even when it's not available,
+// then implements a slow software scalar fallback calling fmaf()!
+// Filed LLVM bug:
+//     https://llvm.org/bugs/show_bug.cgi?id=27216
+#if (defined __ARM_FEATURE_FMA) && !(EIGEN_COMP_CLANG && EIGEN_ARCH_ARM)
+// See bug 936.
+// FMA is available on VFPv4 i.e. when compiling with -mfpu=neon-vfpv4.
+// FMA is a true fused multiply-add i.e. only 1 rounding at the end, no intermediate rounding.
+// MLA is not fused i.e. does 2 roundings.
+// In addition to giving better accuracy, FMA also gives better performance here on a Krait (Nexus 4):
+// MLA: 10 GFlop/s ; FMA: 12 GFlops/s.
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return vfmaq_f32(c,a,b); }
+#else
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) {
+#if EIGEN_COMP_CLANG && EIGEN_ARCH_ARM
+  // Clang/ARM will replace VMLA by VMUL+VADD at least for some values of -mcpu,
+  // at least -mcpu=cortex-a8 and -mcpu=cortex-a7. Since the former is the default on
+  // -march=armv7-a, that is a very common case.
+  // See e.g. this thread:
+  //     http://lists.llvm.org/pipermail/llvm-dev/2013-December/068806.html
+  // Filed LLVM bug:
+  //     https://llvm.org/bugs/show_bug.cgi?id=27219
+  Packet4f r = c;
+  asm volatile(
+    "vmla.f32 %q[r], %q[a], %q[b]"
+    : [r] "+w" (r)
+    : [a] "w" (a),
+      [b] "w" (b)
+    : );
+  return r;
+#else
+  return vmlaq_f32(c,a,b);
+#endif
+}
+#endif
+
+// No FMA instruction for int, so use MLA unconditionally.
 template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return vmlaq_s32(c,a,b); }
 
 template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) { return vminq_f32(a,b); }
@@ -191,20 +274,20 @@
 }
 template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return vbicq_s32(a,b); }
 
-template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float* from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*   from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*    from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int32_t*  from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f32(from); }
-template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)   { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_s32(from); }
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*   from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f32(from); }
+template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int32_t* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_s32(from); }
 
-template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
+template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float* from)
 {
   float32x2_t lo, hi;
   lo = vld1_dup_f32(from);
   hi = vld1_dup_f32(from+1);
   return vcombine_f32(lo, hi);
 }
-template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
+template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int32_t* from)
 {
   int32x2_t lo, hi;
   lo = vld1_dup_s32(from);
@@ -212,18 +295,52 @@
   return vcombine_s32(lo, hi);
 }
 
-template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f32(to, from); }
-template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_s32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<float>  (float*    to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstore<int32_t>(int32_t*  to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_s32(to, from); }
 
-template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); }
-template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*      to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>  (float*   to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f32(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<int32_t>(int32_t* to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_s32(to, from); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float* addr) { EIGEN_ARM_PREFETCH(addr); }
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*     addr) { EIGEN_ARM_PREFETCH(addr); }
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+  Packet4f res = pset1<Packet4f>(0.f);
+  res = vsetq_lane_f32(from[0*stride], res, 0);
+  res = vsetq_lane_f32(from[1*stride], res, 1);
+  res = vsetq_lane_f32(from[2*stride], res, 2);
+  res = vsetq_lane_f32(from[3*stride], res, 3);
+  return res;
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int32_t, Packet4i>(const int32_t* from, Index stride)
+{
+  Packet4i res = pset1<Packet4i>(0);
+  res = vsetq_lane_s32(from[0*stride], res, 0);
+  res = vsetq_lane_s32(from[1*stride], res, 1);
+  res = vsetq_lane_s32(from[2*stride], res, 2);
+  res = vsetq_lane_s32(from[3*stride], res, 3);
+  return res;
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  to[stride*0] = vgetq_lane_f32(from, 0);
+  to[stride*1] = vgetq_lane_f32(from, 1);
+  to[stride*2] = vgetq_lane_f32(from, 2);
+  to[stride*3] = vgetq_lane_f32(from, 3);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int32_t, Packet4i>(int32_t* to, const Packet4i& from, Index stride)
+{
+  to[stride*0] = vgetq_lane_s32(from, 0);
+  to[stride*1] = vgetq_lane_s32(from, 1);
+  to[stride*2] = vgetq_lane_s32(from, 2);
+  to[stride*3] = vgetq_lane_s32(from, 3);
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<float>  (const float*    addr) { EIGEN_ARM_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<int32_t>(const int32_t*  addr) { EIGEN_ARM_PREFETCH(addr); }
 
 // FIXME only store the 2 first elements ?
-template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float EIGEN_ALIGN16 x[4]; vst1q_f32(x, a); return x[0]; }
-template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int   EIGEN_ALIGN16 x[4]; vst1q_s32(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE float   pfirst<Packet4f>(const Packet4f& a) { float   EIGEN_ALIGN16 x[4]; vst1q_f32(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE int32_t pfirst<Packet4i>(const Packet4i& a) { int32_t EIGEN_ALIGN16 x[4]; vst1q_s32(x, a); return x[0]; }
 
 template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a) {
   float32x2_t a_lo, a_hi;
@@ -243,6 +360,7 @@
   a_hi = vget_high_s32(a_r64);
   return vcombine_s32(a_hi, a_lo);
 }
+
 template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a) { return vabsq_f32(a); }
 template<> EIGEN_STRONG_INLINE Packet4i pabs(const Packet4i& a) { return vabsq_s32(a); }
 
@@ -277,7 +395,7 @@
   return sum;
 }
 
-template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, sum;
 
@@ -324,7 +442,7 @@
 
   return vget_lane_f32(prod, 0);
 }
-template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_mul<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, prod;
 
@@ -352,7 +470,7 @@
   return vget_lane_f32(min, 0);
 }
 
-template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_min<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, min;
 
@@ -377,13 +495,14 @@
   return vget_lane_f32(max, 0);
 }
 
-template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
+template<> EIGEN_STRONG_INLINE int32_t predux_max<Packet4i>(const Packet4i& a)
 {
   int32x2_t a_lo, a_hi, max;
 
   a_lo = vget_low_s32(a);
   a_hi = vget_high_s32(a);
   max = vpmax_s32(a_lo, a_hi);
+  max = vpmax_s32(max, max);
 
   return vget_lane_s32(max, 0);
 }
@@ -409,9 +528,231 @@
 PALIGN_NEON(1,Packet4i,vextq_s32)
 PALIGN_NEON(2,Packet4i,vextq_s32)
 PALIGN_NEON(3,Packet4i,vextq_s32)
-    
+
 #undef PALIGN_NEON
 
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  float32x4x2_t tmp1 = vzipq_f32(kernel.packet[0], kernel.packet[1]);
+  float32x4x2_t tmp2 = vzipq_f32(kernel.packet[2], kernel.packet[3]);
+
+  kernel.packet[0] = vcombine_f32(vget_low_f32(tmp1.val[0]), vget_low_f32(tmp2.val[0]));
+  kernel.packet[1] = vcombine_f32(vget_high_f32(tmp1.val[0]), vget_high_f32(tmp2.val[0]));
+  kernel.packet[2] = vcombine_f32(vget_low_f32(tmp1.val[1]), vget_low_f32(tmp2.val[1]));
+  kernel.packet[3] = vcombine_f32(vget_high_f32(tmp1.val[1]), vget_high_f32(tmp2.val[1]));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  int32x4x2_t tmp1 = vzipq_s32(kernel.packet[0], kernel.packet[1]);
+  int32x4x2_t tmp2 = vzipq_s32(kernel.packet[2], kernel.packet[3]);
+  kernel.packet[0] = vcombine_s32(vget_low_s32(tmp1.val[0]), vget_low_s32(tmp2.val[0]));
+  kernel.packet[1] = vcombine_s32(vget_high_s32(tmp1.val[0]), vget_high_s32(tmp2.val[0]));
+  kernel.packet[2] = vcombine_s32(vget_low_s32(tmp1.val[1]), vget_low_s32(tmp2.val[1]));
+  kernel.packet[3] = vcombine_s32(vget_high_s32(tmp1.val[1]), vget_high_s32(tmp2.val[1]));
+}
+
+//---------- double ----------
+
+// Clang 3.5 in the iOS toolchain has an ICE triggered by NEON intrisics for double.
+// Confirmed at least with __apple_build_version__ = 6000054.
+#ifdef __apple_build_version__
+// Let's hope that by the time __apple_build_version__ hits the 601* range, the bug will be fixed.
+// https://gist.github.com/yamaya/2924292 suggests that the 3 first digits are only updated with
+// major toolchain updates.
+#define EIGEN_APPLE_DOUBLE_NEON_BUG (__apple_build_version__ < 6010000)
+#else
+#define EIGEN_APPLE_DOUBLE_NEON_BUG 0
+#endif
+
+#if EIGEN_ARCH_ARM64 && !EIGEN_APPLE_DOUBLE_NEON_BUG
+
+// Bug 907: workaround missing declarations of the following two functions in the ADK
+// Defining these functions as templates ensures that if these intrinsics are
+// already defined in arm_neon.h, then our workaround doesn't cause a conflict
+// and has lower priority in overload resolution.
+template <typename T>
+uint64x2_t vreinterpretq_u64_f64(T a)
+{
+  return (uint64x2_t) a;
+}
+
+template <typename T>
+float64x2_t vreinterpretq_f64_u64(T a)
+{
+  return (float64x2_t) a;
+}
+
+typedef float64x2_t Packet2d;
+typedef float64x1_t Packet1d;
+
+template<> struct packet_traits<double>  : default_packet_traits
+{
+  typedef Packet2d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 2,
+    HasHalfPacket=0,
+   
+    HasDiv  = 1,
+    // FIXME check the Has*
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 0,
+    HasSqrt = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet2d> { typedef double  type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+
+template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double&  from) { return vdupq_n_f64(from); }
+
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a)
+{
+  const double countdown_raw[] = {0.0,1.0};
+  const Packet2d countdown = vld1q_f64(countdown_raw);
+  return vaddq_f64(pset1<Packet2d>(a), countdown);
+}
+template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return vaddq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return vsubq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return vnegq_f64(a); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return vmulq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return vdivq_f64(a,b); }
+
+#ifdef __ARM_FEATURE_FMA
+// See bug 936. See above comment about FMA for float.
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vfmaq_f64(c,a,b); }
+#else
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vmlaq_f64(c,a,b); }
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return vminq_f64(a,b); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b) { return vmaxq_f64(a,b); }
+
+// Logical Operations are not supported for float, so we have to reinterpret casts using NEON intrinsics
+template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(vandq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(vorrq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(veorq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b)
+{
+  return vreinterpretq_f64_u64(vbicq_u64(vreinterpretq_u64_f64(a),vreinterpretq_u64_f64(b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from) { EIGEN_DEBUG_ALIGNED_LOAD return vld1q_f64(from); }
+
+template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return vld1q_f64(from); }
+
+template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
+{
+  return vld1q_dup_f64(from);
+}
+template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d& from) { EIGEN_DEBUG_ALIGNED_STORE vst1q_f64(to, from); }
+
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d& from) { EIGEN_DEBUG_UNALIGNED_STORE vst1q_f64(to, from); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+  Packet2d res = pset1<Packet2d>(0.0);
+  res = vsetq_lane_f64(from[0*stride], res, 0);
+  res = vsetq_lane_f64(from[1*stride], res, 1);
+  return res;
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  to[stride*0] = vgetq_lane_f64(from, 0);
+  to[stride*1] = vgetq_lane_f64(from, 1);
+}
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ARM_PREFETCH(addr); }
+
+// FIXME only store the 2 first elements ?
+template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { return vgetq_lane_f64(a, 0); }
+
+template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a) { return vcombine_f64(vget_high_f64(a), vget_low_f64(a)); }
+
+template<> EIGEN_STRONG_INLINE Packet2d pabs(const Packet2d& a) { return vabsq_f64(a); }
+
+#if EIGEN_COMP_CLANG && defined(__apple_build_version__)
+// workaround ICE, see bug 907
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return (vget_low_f64(a) + vget_high_f64(a))[0]; }
+#else
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return vget_lane_f64(vget_low_f64(a) + vget_high_f64(a), 0); }
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
+{
+  float64x2_t trn1, trn2;
+
+  // NEON zip performs interleaving of the supplied vectors.
+  // We perform two interleaves in a row to acquire the transposed vector
+  trn1 = vzip1q_f64(vecs[0], vecs[1]);
+  trn2 = vzip2q_f64(vecs[0], vecs[1]);
+
+  // Do the addition of the resulting vectors
+  return vaddq_f64(trn1, trn2);
+}
+// Other reduction functions:
+// mul
+#if EIGEN_COMP_CLANG && defined(__apple_build_version__)
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a) { return (vget_low_f64(a) * vget_high_f64(a))[0]; }
+#else
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a) { return vget_lane_f64(vget_low_f64(a) * vget_high_f64(a), 0); }
+#endif
+
+// min
+template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a) { return vgetq_lane_f64(vpminq_f64(a, a), 0); }
+
+// max
+template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a) { return vgetq_lane_f64(vpmaxq_f64(a, a), 0); }
+
+// this PALIGN_NEON business is to work around a bug in LLVM Clang 3.0 causing incorrect compilation errors,
+// see bug 347 and this LLVM bug: http://llvm.org/bugs/show_bug.cgi?id=11074
+#define PALIGN_NEON(Offset,Type,Command) \
+template<>\
+struct palign_impl<Offset,Type>\
+{\
+    EIGEN_STRONG_INLINE static void run(Type& first, const Type& second)\
+    {\
+        if (Offset!=0)\
+            first = Command(first, second, Offset);\
+    }\
+};\
+
+PALIGN_NEON(0,Packet2d,vextq_f64)
+PALIGN_NEON(1,Packet2d,vextq_f64)
+#undef PALIGN_NEON
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  float64x2_t trn1 = vzip1q_f64(kernel.packet[0], kernel.packet[1]);
+  float64x2_t trn2 = vzip2q_f64(kernel.packet[0], kernel.packet[1]);
+
+  kernel.packet[0] = trn1;
+  kernel.packet[1] = trn2;
+}
+#endif // EIGEN_ARCH_ARM64 
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/arch/SSE/CMakeLists.txt b/Eigen/src/Core/arch/SSE/CMakeLists.txt
deleted file mode 100644
index 46ea7cc..0000000
--- a/Eigen/src/Core/arch/SSE/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Core_arch_SSE_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Core_arch_SSE_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core/arch/SSE COMPONENT Devel
-)
diff --git a/Eigen/src/Core/arch/SSE/Complex.h b/Eigen/src/Core/arch/SSE/Complex.h
index 91bba5e..d075043 100644
--- a/Eigen/src/Core/arch/SSE/Complex.h
+++ b/Eigen/src/Core/arch/SSE/Complex.h
@@ -22,13 +22,18 @@
   __m128  v;
 };
 
+// Use the packet_traits defined in AVX/PacketMath.h instead if we're going
+// to leverage AVX instructions.
+#ifndef EIGEN_VECTORIZE_AVX
 template<> struct packet_traits<std::complex<float> >  : default_packet_traits
 {
   typedef Packet2cf type;
+  typedef Packet2cf half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size = 2,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -39,11 +44,13 @@
     HasAbs2   = 0,
     HasMin    = 0,
     HasMax    = 0,
-    HasSetLinear = 0
+    HasSetLinear = 0,
+    HasBlend = 1
   };
 };
+#endif
 
-template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2}; };
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float> type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
 
 template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_add_ps(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(_mm_sub_ps(a.v,b.v)); }
@@ -60,7 +67,6 @@
 
 template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
-  // TODO optimize it for SSE3 and 4
   #ifdef EIGEN_VECTORIZE_SSE3
   return Packet2cf(_mm_addsub_ps(_mm_mul_ps(_mm_moveldup_ps(a.v), b.v),
                                  _mm_mul_ps(_mm_movehdup_ps(a.v),
@@ -104,10 +110,25 @@
 
 template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>* from) { return pset1<Packet2cf>(*from); }
 
-template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), from.v); }
-template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), from.v); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore(&numext::real_ref(*to), Packet4f(from.v)); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *   to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(&numext::real_ref(*to), Packet4f(from.v)); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  return Packet2cf(_mm_set_ps(std::imag(from[1*stride]), std::real(from[1*stride]),
+                              std::imag(from[0*stride]), std::real(from[0*stride])));
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  to[stride*0] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 0)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 1)));
+  to[stride*1] = std::complex<float>(_mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 2)),
+                                     _mm_cvtss_f32(_mm_shuffle_ps(from.v, from.v, 3)));
+}
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
 {
@@ -124,7 +145,7 @@
   #endif
 }
 
-template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a) { return Packet2cf(_mm_castpd_ps(preverse(_mm_castps_pd(a.v)))); }
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a) { return Packet2cf(_mm_castpd_ps(preverse(Packet2d(_mm_castps_pd(a.v))))); }
 
 template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
 {
@@ -208,23 +229,7 @@
   }
 };
 
-template<> struct conj_helper<Packet4f, Packet2cf, false,false>
-{
-  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet4f& x, const Packet2cf& y, const Packet2cf& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet4f& x, const Packet2cf& y) const
-  { return Packet2cf(Eigen::internal::pmul(x, y.v)); }
-};
-
-template<> struct conj_helper<Packet2cf, Packet4f, false,false>
-{
-  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet4f& y, const Packet2cf& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& x, const Packet4f& y) const
-  { return Packet2cf(Eigen::internal::pmul(x.v, y)); }
-};
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
 
 template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
 {
@@ -234,7 +239,7 @@
   return Packet2cf(_mm_div_ps(res.v,_mm_add_ps(s,_mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(s), 0xb1)))));
 }
 
-EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
+EIGEN_STRONG_INLINE Packet2cf pcplxflip/* <Packet2cf> */(const Packet2cf& x)
 {
   return Packet2cf(vec4f_swizzle1(x.v, 1, 0, 3, 2));
 }
@@ -248,13 +253,18 @@
   __m128d  v;
 };
 
+// Use the packet_traits defined in AVX/PacketMath.h instead if we're going
+// to leverage AVX instructions.
+#ifndef EIGEN_VECTORIZE_AVX
 template<> struct packet_traits<std::complex<double> >  : default_packet_traits
 {
   typedef Packet1cd type;
+  typedef Packet1cd half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 0,
     size = 1,
+    HasHalfPacket = 0,
 
     HasAdd    = 1,
     HasSub    = 1,
@@ -268,12 +278,13 @@
     HasSetLinear = 0
   };
 };
+#endif
 
-template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1}; };
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
 
 template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_add_pd(a.v,b.v)); }
 template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(_mm_sub_pd(a.v,b.v)); }
-template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(a.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
 template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a)
 {
   const __m128d mask = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0));
@@ -282,9 +293,8 @@
 
 template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
 {
-  // TODO optimize it for SSE3 and 4
   #ifdef EIGEN_VECTORIZE_SSE3
-  return Packet1cd(_mm_addsub_pd(_mm_mul_pd(vec2d_swizzle1(a.v, 0, 0), b.v),
+  return Packet1cd(_mm_addsub_pd(_mm_mul_pd(_mm_movedup_pd(a.v), b.v),
                                  _mm_mul_pd(vec2d_swizzle1(a.v, 1, 1),
                                             vec2d_swizzle1(b.v, 1, 0))));
   #else
@@ -311,10 +321,10 @@
 template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>* from) { return pset1<Packet1cd>(*from); }
 
 // FIXME force unaligned store, this is a temporary fix
-template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
-template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, Packet2d(from.v)); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, Packet2d(from.v)); }
 
-template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
 
 template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
 {
@@ -404,23 +414,7 @@
   }
 };
 
-template<> struct conj_helper<Packet2d, Packet1cd, false,false>
-{
-  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet2d& x, const Packet1cd& y, const Packet1cd& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet2d& x, const Packet1cd& y) const
-  { return Packet1cd(Eigen::internal::pmul(x, y.v)); }
-};
-
-template<> struct conj_helper<Packet1cd, Packet2d, false,false>
-{
-  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet2d& y, const Packet1cd& c) const
-  { return padd(c, pmul(x,y)); }
-
-  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& x, const Packet2d& y) const
-  { return Packet1cd(Eigen::internal::pmul(x.v, y)); }
-};
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
 
 template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
 {
@@ -430,9 +424,44 @@
   return Packet1cd(_mm_div_pd(res.v, _mm_add_pd(s,_mm_shuffle_pd(s, s, 0x1))));
 }
 
-EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/* <Packet1cd> */(const Packet1cd& x)
 {
-  return Packet1cd(preverse(x.v));
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2cf,2>& kernel) {
+  __m128d w1 = _mm_castps_pd(kernel.packet[0].v);
+  __m128d w2 = _mm_castps_pd(kernel.packet[1].v);
+
+  __m128 tmp = _mm_castpd_ps(_mm_unpackhi_pd(w1, w2));
+  kernel.packet[0].v = _mm_castpd_ps(_mm_unpacklo_pd(w1, w2));
+  kernel.packet[1].v = tmp;
+}
+
+template<>  EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+  __m128d result = pblend<Packet2d>(ifPacket, _mm_castps_pd(thenPacket.v), _mm_castps_pd(elsePacket.v));
+  return Packet2cf(_mm_castpd_ps(result));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pinsertfirst(const Packet2cf& a, std::complex<float> b)
+{
+  return Packet2cf(_mm_loadl_pi(a.v, reinterpret_cast<const __m64*>(&b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pinsertfirst(const Packet1cd&, std::complex<double> b)
+{
+  return pset1<Packet1cd>(b);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pinsertlast(const Packet2cf& a, std::complex<float> b)
+{
+  return Packet2cf(_mm_loadh_pi(a.v, reinterpret_cast<const __m64*>(&b)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pinsertlast(const Packet1cd&, std::complex<double> b)
+{
+  return pset1<Packet1cd>(b);
 }
 
 } // end namespace internal
diff --git a/Eigen/src/Core/arch/SSE/MathFunctions.h b/Eigen/src/Core/arch/SSE/MathFunctions.h
index d16f30b..7b5f948 100644
--- a/Eigen/src/Core/arch/SSE/MathFunctions.h
+++ b/Eigen/src/Core/arch/SSE/MathFunctions.h
@@ -32,7 +32,7 @@
   /* the smallest non denormalized float number */
   _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(min_norm_pos,  0x00800000);
   _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(minus_inf,     0xff800000);//-1.f/0.f);
-  
+
   /* natural logarithm computed for 4 simultaneous float
     return NaN for x <= 0
   */
@@ -63,7 +63,7 @@
   x = _mm_or_ps(x, p4f_half);
 
   emm0 = _mm_sub_epi32(emm0, p4i_0x7f);
-  Packet4f e = padd(_mm_cvtepi32_ps(emm0), p4f_1);
+  Packet4f e = padd(Packet4f(_mm_cvtepi32_ps(emm0)), p4f_1);
 
   /* part2:
      if( x < SQRTHF ) {
@@ -72,9 +72,9 @@
      } else { x = x - 1.0; }
   */
   Packet4f mask = _mm_cmplt_ps(x, p4f_cephes_SQRTHF);
-  Packet4f tmp = _mm_and_ps(x, mask);
+  Packet4f tmp = pand(x, mask);
   x = psub(x, p4f_1);
-  e = psub(e, _mm_and_ps(p4f_1, mask));
+  e = psub(e, pand(p4f_1, mask));
   x = padd(x, tmp);
 
   Packet4f x2 = pmul(x,x);
@@ -126,7 +126,7 @@
   _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p4, 1.6666665459E-1f);
   _EIGEN_DECLARE_CONST_Packet4f(cephes_exp_p5, 5.0000001201E-1f);
 
-  Packet4f tmp = _mm_setzero_ps(), fx;
+  Packet4f tmp, fx;
   Packet4i emm0;
 
   // clamp x
@@ -195,7 +195,7 @@
   _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
   static const __m128i p4i_1023_0 = _mm_setr_epi32(1023, 1023, 0, 0);
 
-  Packet2d tmp = _mm_setzero_pd(), fx;
+  Packet2d tmp, fx;
   Packet4i emm0;
 
   // clamp x
@@ -279,7 +279,7 @@
   _EIGEN_DECLARE_CONST_Packet4f(coscof_p2,  4.166664568298827E-002f);
   _EIGEN_DECLARE_CONST_Packet4f(cephes_FOPI, 1.27323954473516f); // 4 / M_PI
 
-  Packet4f xmm1, xmm2 = _mm_setzero_ps(), xmm3, sign_bit, y;
+  Packet4f xmm1, xmm2, xmm3, sign_bit, y;
 
   Packet4i emm0, emm2;
   sign_bit = x;
@@ -378,7 +378,7 @@
   _EIGEN_DECLARE_CONST_Packet4f(coscof_p2,  4.166664568298827E-002f);
   _EIGEN_DECLARE_CONST_Packet4f(cephes_FOPI, 1.27323954473516f); // 4 / M_PI
 
-  Packet4f xmm1, xmm2 = _mm_setzero_ps(), xmm3, y;
+  Packet4f xmm1, xmm2, xmm3, y;
   Packet4i emm0, emm2;
 
   x = pabs(x);
@@ -444,32 +444,119 @@
 
 #if EIGEN_FAST_MATH
 
-// This is based on Quake3's fast inverse square root.
+// Functions for sqrt.
+// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
+// of Newton's method, at a cost of 1-2 bits of precision as opposed to the
+// exact solution. It does not handle +inf, or denormalized numbers correctly.
+// The main advantage of this approach is not just speed, but also the fact that
+// it can be inlined and pipelined with other computations, further reducing its
+// effective latency. This is similar to Quake3's fast inverse square root.
 // For detail see here: http://www.beyond3d.com/content/articles/8/
-// It lacks 1 (or 2 bits in some rare cases) of precision, and does not handle negative, +inf, or denormalized numbers correctly.
 template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
 Packet4f psqrt<Packet4f>(const Packet4f& _x)
 {
   Packet4f half = pmul(_x, pset1<Packet4f>(.5f));
+  Packet4f denormal_mask = _mm_and_ps(
+      _mm_cmpge_ps(_x, _mm_setzero_ps()),
+      _mm_cmplt_ps(_x, pset1<Packet4f>((std::numeric_limits<float>::min)())));
 
-  /* select only the inverse sqrt of non-zero inputs */
-  Packet4f non_zero_mask = _mm_cmpge_ps(_x, pset1<Packet4f>((std::numeric_limits<float>::min)()));
-  Packet4f x = _mm_and_ps(non_zero_mask, _mm_rsqrt_ps(_x));
-
+  // Compute approximate reciprocal sqrt.
+  Packet4f x = _mm_rsqrt_ps(_x);
+  // Do a single step of Newton's iteration.
   x = pmul(x, psub(pset1<Packet4f>(1.5f), pmul(half, pmul(x,x))));
-  return pmul(_x,x);
+  // Flush results for denormals to zero.
+  return _mm_andnot_ps(denormal_mask, pmul(_x,x));
 }
 
 #else
 
-template<> EIGEN_STRONG_INLINE Packet4f psqrt<Packet4f>(const Packet4f& x) { return _mm_sqrt_ps(x); }
+template<>EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psqrt<Packet4f>(const Packet4f& x) { return _mm_sqrt_ps(x); }
 
 #endif
 
-template<> EIGEN_STRONG_INLINE Packet2d psqrt<Packet2d>(const Packet2d& x) { return _mm_sqrt_pd(x); }
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d psqrt<Packet2d>(const Packet2d& x) { return _mm_sqrt_pd(x); }
+
+#if EIGEN_FAST_MATH
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& _x) {
+  _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(inf, 0x7f800000);
+  _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(nan, 0x7fc00000);
+  _EIGEN_DECLARE_CONST_Packet4f(one_point_five, 1.5f);
+  _EIGEN_DECLARE_CONST_Packet4f(minus_half, -0.5f);
+  _EIGEN_DECLARE_CONST_Packet4f_FROM_INT(flt_min, 0x00800000);
+
+  Packet4f neg_half = pmul(_x, p4f_minus_half);
+
+  // select only the inverse sqrt of positive normal inputs (denormals are
+  // flushed to zero and cause infs as well).
+  Packet4f le_zero_mask = _mm_cmple_ps(_x, p4f_flt_min);
+  Packet4f x = _mm_andnot_ps(le_zero_mask, _mm_rsqrt_ps(_x));
+
+  // Fill in NaNs and Infs for the negative/zero entries.
+  Packet4f neg_mask = _mm_cmplt_ps(_x, _mm_setzero_ps());
+  Packet4f zero_mask = _mm_andnot_ps(neg_mask, le_zero_mask);
+  Packet4f infs_and_nans = _mm_or_ps(_mm_and_ps(neg_mask, p4f_nan),
+                                     _mm_and_ps(zero_mask, p4f_inf));
+
+  // Do a single step of Newton's iteration.
+  x = pmul(x, pmadd(neg_half, pmul(x, x), p4f_one_point_five));
+
+  // Insert NaNs and Infs in all the right places.
+  return _mm_or_ps(x, infs_and_nans);
+}
+
+#else
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& x) {
+  // Unfortunately we can't use the much faster mm_rqsrt_ps since it only provides an approximation.
+  return _mm_div_ps(pset1<Packet4f>(1.0f), _mm_sqrt_ps(x));
+}
+
+#endif
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d prsqrt<Packet2d>(const Packet2d& x) {
+  // Unfortunately we can't use the much faster mm_rqsrt_pd since it only provides an approximation.
+  return _mm_div_pd(pset1<Packet2d>(1.0), _mm_sqrt_pd(x));
+}
+
+// Hyperbolic Tangent function.
+template <>
+EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED Packet4f
+ptanh<Packet4f>(const Packet4f& x) {
+  return internal::generic_fast_tanh_float(x);
+}
 
 } // end namespace internal
 
+namespace numext {
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+float sqrt(const float &x)
+{
+  return internal::pfirst(internal::Packet4f(_mm_sqrt_ss(_mm_set_ss(x))));
+}
+
+template<>
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE
+double sqrt(const double &x)
+{
+#if EIGEN_COMP_GNUC_STRICT
+  // This works around a GCC bug generating poor code for _mm_sqrt_pd
+  // See https://bitbucket.org/eigen/eigen/commits/14f468dba4d350d7c19c9b93072e19f7b3df563b
+  return internal::pfirst(internal::Packet2d(__builtin_ia32_sqrtsd(_mm_set_sd(x))));
+#else
+  return internal::pfirst(internal::Packet2d(_mm_sqrt_pd(_mm_set_sd(x))));
+#endif
+}
+
+} // end namespace numex
+
 } // end namespace Eigen
 
 #endif // EIGEN_MATH_FUNCTIONS_SSE_H
diff --git a/Eigen/src/Core/arch/SSE/PacketMath.h b/Eigen/src/Core/arch/SSE/PacketMath.h
old mode 100644
new mode 100755
index fc8ae50..60e2517
--- a/Eigen/src/Core/arch/SSE/PacketMath.h
+++ b/Eigen/src/Core/arch/SSE/PacketMath.h
@@ -22,9 +22,40 @@
 #define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS (2*sizeof(void*))
 #endif
 
+#ifdef __FMA__
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD 1
+#endif
+#endif
+
+#if ((defined EIGEN_VECTORIZE_AVX) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_MINGW) && (__GXX_ABI_VERSION < 1004)) || EIGEN_OS_QNX
+// With GCC's default ABI version, a __m128 or __m256 are the same types and therefore we cannot
+// have overloads for both types without linking error.
+// One solution is to increase ABI version using -fabi-version=4 (or greater).
+// Otherwise, we workaround this inconvenience by wrapping 128bit types into the following helper
+// structure:
+template<typename T>
+struct eigen_packet_wrapper
+{
+  EIGEN_ALWAYS_INLINE operator T&() { return m_val; }
+  EIGEN_ALWAYS_INLINE operator const T&() const { return m_val; }
+  EIGEN_ALWAYS_INLINE eigen_packet_wrapper() {}
+  EIGEN_ALWAYS_INLINE eigen_packet_wrapper(const T &v) : m_val(v) {}
+  EIGEN_ALWAYS_INLINE eigen_packet_wrapper& operator=(const T &v) {
+    m_val = v;
+    return *this;
+  }
+  
+  T m_val;
+};
+typedef eigen_packet_wrapper<__m128>  Packet4f;
+typedef eigen_packet_wrapper<__m128i> Packet4i;
+typedef eigen_packet_wrapper<__m128d> Packet2d;
+#else
 typedef __m128  Packet4f;
 typedef __m128i Packet4i;
 typedef __m128d Packet2d;
+#endif
 
 template<> struct is_arithmetic<__m128>  { enum { value = true }; };
 template<> struct is_arithmetic<__m128i> { enum { value = true }; };
@@ -58,51 +89,85 @@
   const Packet4i p4i_##NAME = pset1<Packet4i>(X)
 
 
+// Use the packet_traits defined in AVX/PacketMath.h instead if we're going
+// to leverage AVX instructions.
+#ifndef EIGEN_VECTORIZE_AVX
 template<> struct packet_traits<float>  : default_packet_traits
 {
   typedef Packet4f type;
+  typedef Packet4f half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size=4,
+    HasHalfPacket = 0,
 
     HasDiv  = 1,
     HasSin  = EIGEN_FAST_MATH,
     HasCos  = EIGEN_FAST_MATH,
     HasLog  = 1,
     HasExp  = 1,
-    HasSqrt = 1
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasTanh  = EIGEN_FAST_MATH,
+    HasBlend = 1
+
+#ifdef EIGEN_VECTORIZE_SSE4_1
+    ,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+#endif
   };
 };
 template<> struct packet_traits<double> : default_packet_traits
 {
   typedef Packet2d type;
+  typedef Packet2d half;
   enum {
     Vectorizable = 1,
     AlignedOnScalar = 1,
     size=2,
+    HasHalfPacket = 0,
 
     HasDiv  = 1,
     HasExp  = 1,
-    HasSqrt = 1
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasBlend = 1
+
+#ifdef EIGEN_VECTORIZE_SSE4_1
+    ,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1
+#endif
   };
 };
+#endif
 template<> struct packet_traits<int>    : default_packet_traits
 {
   typedef Packet4i type;
+  typedef Packet4i half;
   enum {
-    // FIXME check the Has*
     Vectorizable = 1,
     AlignedOnScalar = 1,
-    size=4
+    size=4,
+
+    HasBlend = 1
   };
 };
 
-template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4}; };
-template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2}; };
-template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4}; };
+template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
 
-#if defined(_MSC_VER) && (_MSC_VER==1500)
+#ifndef EIGEN_VECTORIZE_AVX
+template<> struct scalar_div_cost<float,true> { enum { value = 7 }; };
+template<> struct scalar_div_cost<double,true> { enum { value = 8 }; };
+#endif
+
+#if EIGEN_COMP_MSVC==1500
 // Workaround MSVC 9 internal compiler error.
 // TODO: It has been detected with win64 builds (amd64), so let's check whether it also happens in 32bits+SSE mode
 // TODO: let's check whether there does not exist a better fix, like adding a pset0() function. (it crashed on pset1(0)).
@@ -110,14 +175,25 @@
 template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { return _mm_set_pd(from,from); }
 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from) { return _mm_set_epi32(from,from,from,from); }
 #else
-template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { return _mm_set1_ps(from); }
+template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&  from) { return _mm_set_ps1(from); }
 template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) { return _mm_set1_pd(from); }
 template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from) { return _mm_set1_epi32(from); }
 #endif
 
-template<> EIGEN_STRONG_INLINE Packet4f plset<float>(const float& a) { return _mm_add_ps(pset1<Packet4f>(a), _mm_set_ps(3,2,1,0)); }
-template<> EIGEN_STRONG_INLINE Packet2d plset<double>(const double& a) { return _mm_add_pd(pset1<Packet2d>(a),_mm_set_pd(1,0)); }
-template<> EIGEN_STRONG_INLINE Packet4i plset<int>(const int& a) { return _mm_add_epi32(pset1<Packet4i>(a),_mm_set_epi32(3,2,1,0)); }
+// GCC generates a shufps instruction for _mm_set1_ps/_mm_load1_ps instead of the more efficient pshufd instruction.
+// However, using inrinsics for pset1 makes gcc to generate crappy code in some cases (see bug 203)
+// Using inline assembly is also not an option because then gcc fails to reorder properly the instructions.
+// Therefore, we introduced the pload1 functions to be used in product kernels for which bug 203 does not apply.
+// Also note that with AVX, we want it to generate a vbroadcastss.
+#if EIGEN_COMP_GNUC_STRICT && (!defined __AVX__)
+template<> EIGEN_STRONG_INLINE Packet4f pload1<Packet4f>(const float *from) {
+  return vec4f_swizzle1(_mm_load_ss(from),0,0,0,0);
+}
+#endif
+  
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a) { return _mm_add_ps(pset1<Packet4f>(a), _mm_set_ps(3,2,1,0)); }
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return _mm_add_pd(pset1<Packet2d>(a),_mm_set_pd(1,0)); }
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a) { return _mm_add_epi32(pset1<Packet4i>(a),_mm_set_epi32(3,2,1,0)); }
 
 template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_add_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_add_pd(a,b); }
@@ -139,7 +215,7 @@
 }
 template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a)
 {
-  return psub(_mm_setr_epi32(0,0,0,0), a);
+  return psub(Packet4i(_mm_setr_epi32(0,0,0,0)), a);
 }
 
 template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
@@ -166,13 +242,13 @@
 
 template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_div_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_div_pd(a,b); }
-template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& /*a*/, const Packet4i& /*b*/)
-{ eigen_assert(false && "packet integer division are not supported by SSE");
-  return pset1<Packet4i>(0);
-}
 
 // for some weird raisons, it has to be overloaded for packet of integers
 template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd(pmul(a,b), c); }
+#ifdef __FMA__
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c) { return _mm_fmadd_ps(a,b,c); }
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return _mm_fmadd_pd(a,b,c); }
+#endif
 
 template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_min_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_min_pd(a,b); }
@@ -200,6 +276,17 @@
 #endif
 }
 
+#ifdef EIGEN_VECTORIZE_SSE4_1
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a) { return _mm_round_ps(a, 0); }
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return _mm_round_pd(a, 0); }
+
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const Packet4f& a) { return _mm_ceil_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const Packet2d& a) { return _mm_ceil_pd(a); }
+
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a) { return _mm_floor_ps(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return _mm_floor_pd(a); }
+#endif
+
 template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b) { return _mm_and_ps(a,b); }
 template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return _mm_and_pd(a,b); }
 template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return _mm_and_si128(a,b); }
@@ -218,16 +305,14 @@
 
 template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*   from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_ps(from); }
 template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double*  from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_pd(from); }
-template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_si128(reinterpret_cast<const Packet4i*>(from)); }
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from) { EIGEN_DEBUG_ALIGNED_LOAD return _mm_load_si128(reinterpret_cast<const __m128i*>(from)); }
 
-#if defined(_MSC_VER)
+#if EIGEN_COMP_MSVC
   template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*  from) {
     EIGEN_DEBUG_UNALIGNED_LOAD
-    #if (_MSC_VER==1600)
+    #if (EIGEN_COMP_MSVC==1600)
     // NOTE Some version of MSVC10 generates bad code when using _mm_loadu_ps
     // (i.e., it does not generate an unaligned load!!
-    // TODO On most architectures this version should also be faster than a single _mm_loadu_ps
-    // so we could also enable it for MSVC08 but first we have to make this later does not generate crap when doing so...
     __m128 res = _mm_loadl_pi(_mm_set1_ps(0.0f), (const __m64*)(from));
     res = _mm_loadh_pi(res, (const __m64*)(from+2));
     return res;
@@ -235,63 +320,27 @@
     return _mm_loadu_ps(from);
     #endif
   }
-  template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_pd(from); }
-  template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int*    from) { EIGEN_DEBUG_UNALIGNED_LOAD return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(from)); }
 #else
-// Fast unaligned loads. Note that here we cannot directly use intrinsics: this would
-// require pointer casting to incompatible pointer types and leads to invalid code
-// because of the strict aliasing rule. The "dummy" stuff are required to enforce
-// a correct instruction dependency.
-// TODO: do the same for MSVC (ICC is compatible)
 // NOTE: with the code below, MSVC's compiler crashes!
 
-#if defined(__GNUC__) && defined(__i386__)
-  // bug 195: gcc/i386 emits weird x87 fldl/fstpl instructions for _mm_load_sd
-  #define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 1
-#elif defined(__clang__)
-  // bug 201: Segfaults in __mm_loadh_pd with clang 2.8
-  #define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 1
-#else
-  #define EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS 0
-#endif
-
 template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float* from)
 {
   EIGEN_DEBUG_UNALIGNED_LOAD
-#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
   return _mm_loadu_ps(from);
-#else
-  __m128d res;
-  res =  _mm_load_sd((const double*)(from)) ;
-  res =  _mm_loadh_pd(res, (const double*)(from+2)) ;
-  return _mm_castpd_ps(res);
-#endif
 }
+#endif
+
 template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double* from)
 {
   EIGEN_DEBUG_UNALIGNED_LOAD
-#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
   return _mm_loadu_pd(from);
-#else
-  __m128d res;
-  res = _mm_load_sd(from) ;
-  res = _mm_loadh_pd(res,from+1);
-  return res;
-#endif
 }
 template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int* from)
 {
   EIGEN_DEBUG_UNALIGNED_LOAD
-#if EIGEN_AVOID_CUSTOM_UNALIGNED_LOADS
-  return _mm_loadu_si128(reinterpret_cast<const Packet4i*>(from));
-#else
-  __m128d res;
-  res =  _mm_load_sd((const double*)(from)) ;
-  res =  _mm_loadh_pd(res, (const double*)(from+2)) ;
-  return _mm_castpd_si128(res);
-#endif
+  return _mm_loadu_si128(reinterpret_cast<const __m128i*>(from));
 }
-#endif
+
 
 template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*   from)
 {
@@ -302,46 +351,83 @@
 template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
 {
   Packet4i tmp;
-  tmp = _mm_loadl_epi64(reinterpret_cast<const Packet4i*>(from));
+  tmp = _mm_loadl_epi64(reinterpret_cast<const __m128i*>(from));
   return vec4i_swizzle1(tmp, 0, 0, 1, 1);
 }
 
 template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_ps(to, from); }
 template<> EIGEN_STRONG_INLINE void pstore<double>(double* to, const Packet2d& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_pd(to, from); }
-template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_si128(reinterpret_cast<Packet4i*>(to), from); }
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_ALIGNED_STORE _mm_store_si128(reinterpret_cast<__m128i*>(to), from); }
 
-template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet2d& from) {
-  EIGEN_DEBUG_UNALIGNED_STORE
-  _mm_storel_pd((to), from);
-  _mm_storeh_pd((to+1), from);
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double* to, const Packet2d& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storeu_pd(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*   to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storeu_ps(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*       to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE _mm_storeu_si128(reinterpret_cast<__m128i*>(to), from); }
+
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+ return _mm_set_ps(from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
 }
-template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*  to, const Packet4f& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(reinterpret_cast<double*>(to), _mm_castps_pd(from)); }
-template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*      to, const Packet4i& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu(reinterpret_cast<double*>(to), _mm_castsi128_pd(from)); }
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+ return _mm_set_pd(from[1*stride], from[0*stride]);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
+{
+ return _mm_set_epi32(from[3*stride], from[2*stride], from[1*stride], from[0*stride]);
+ }
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  to[stride*0] = _mm_cvtss_f32(from);
+  to[stride*1] = _mm_cvtss_f32(_mm_shuffle_ps(from, from, 1));
+  to[stride*2] = _mm_cvtss_f32(_mm_shuffle_ps(from, from, 2));
+  to[stride*3] = _mm_cvtss_f32(_mm_shuffle_ps(from, from, 3));
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  to[stride*0] = _mm_cvtsd_f64(from);
+  to[stride*1] = _mm_cvtsd_f64(_mm_shuffle_pd(from, from, 1));
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
+{
+  to[stride*0] = _mm_cvtsi128_si32(from);
+  to[stride*1] = _mm_cvtsi128_si32(_mm_shuffle_epi32(from, 1));
+  to[stride*2] = _mm_cvtsi128_si32(_mm_shuffle_epi32(from, 2));
+  to[stride*3] = _mm_cvtsi128_si32(_mm_shuffle_epi32(from, 3));
+}
 
 // some compilers might be tempted to perform multiple moves instead of using a vector path.
 template<> EIGEN_STRONG_INLINE void pstore1<Packet4f>(float* to, const float& a)
 {
   Packet4f pa = _mm_set_ss(a);
-  pstore(to, vec4f_swizzle1(pa,0,0,0,0));
+  pstore(to, Packet4f(vec4f_swizzle1(pa,0,0,0,0)));
 }
 // some compilers might be tempted to perform multiple moves instead of using a vector path.
 template<> EIGEN_STRONG_INLINE void pstore1<Packet2d>(double* to, const double& a)
 {
   Packet2d pa = _mm_set_sd(a);
-  pstore(to, vec2d_swizzle1(pa,0,0));
+  pstore(to, Packet2d(vec2d_swizzle1(pa,0,0)));
 }
 
-template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
-template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
-template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((const char*)(addr), _MM_HINT_T0); }
+#if EIGEN_COMP_PGI
+typedef const void * SsePrefetchPtrType;
+#else
+typedef const char * SsePrefetchPtrType;
+#endif
 
-#if defined(_MSC_VER) && defined(_WIN64) && !defined(__INTEL_COMPILER)
+#ifndef EIGEN_VECTORIZE_AVX
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { _mm_prefetch((SsePrefetchPtrType)(addr), _MM_HINT_T0); }
+#endif
+
+#if EIGEN_COMP_MSVC_STRICT && EIGEN_OS_WIN64
 // The temporary variable fixes an internal compilation error in vs <= 2008 and a wrong-result bug in vs 2010
 // Direct of the struct members fixed bug #62.
 template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { return a.m128_f32[0]; }
 template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { return a.m128d_f64[0]; }
 template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int x = _mm_cvtsi128_si32(a); return x; }
-#elif defined(_MSC_VER) && !defined(__INTEL_COMPILER)
+#elif EIGEN_COMP_MSVC_STRICT
 // The temporary variable fixes an internal compilation error in vs <= 2008 and a wrong-result bug in vs 2010
 template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float x = _mm_cvtss_f32(a); return x; }
 template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { double x = _mm_cvtsd_f64(a); return x; }
@@ -359,7 +445,6 @@
 template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
 { return _mm_shuffle_epi32(a,0x1B); }
 
-
 template<> EIGEN_STRONG_INLINE Packet4f pabs(const Packet4f& a)
 {
   const Packet4f mask = _mm_castsi128_ps(_mm_setr_epi32(0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF,0x7FFFFFFF));
@@ -380,6 +465,38 @@
   #endif
 }
 
+// with AVX, the default implementations based on pload1 are faster
+#ifndef __AVX__
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = pload<Packet4f>(a);
+  a0 = vec4f_swizzle1(a3, 0,0,0,0);
+  a1 = vec4f_swizzle1(a3, 1,1,1,1);
+  a2 = vec4f_swizzle1(a3, 2,2,2,2);
+  a3 = vec4f_swizzle1(a3, 3,3,3,3);
+}
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet2d>(const double *a,
+                      Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3)
+{
+#ifdef EIGEN_VECTORIZE_SSE3
+  a0 = _mm_loaddup_pd(a+0);
+  a1 = _mm_loaddup_pd(a+1);
+  a2 = _mm_loaddup_pd(a+2);
+  a3 = _mm_loaddup_pd(a+3);
+#else
+  a1 = pload<Packet2d>(a);
+  a0 = vec2d_swizzle1(a1, 0,0);
+  a1 = vec2d_swizzle1(a1, 1,1);
+  a3 = pload<Packet2d>(a+2);
+  a2 = vec2d_swizzle1(a3, 0,0);
+  a3 = vec2d_swizzle1(a3, 1,1);
+#endif
+}
+#endif
+
 EIGEN_STRONG_INLINE void punpackp(Packet4f* vecs)
 {
   vecs[1] = _mm_castsi128_ps(_mm_shuffle_epi32(_mm_castps_si128(vecs[0]), 0x55));
@@ -389,47 +506,17 @@
 }
 
 #ifdef EIGEN_VECTORIZE_SSE3
-// TODO implement SSE2 versions as well as integer versions
 template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
 {
   return _mm_hadd_ps(_mm_hadd_ps(vecs[0], vecs[1]),_mm_hadd_ps(vecs[2], vecs[3]));
 }
+
 template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
 {
   return _mm_hadd_pd(vecs[0], vecs[1]);
 }
-// SSSE3 version:
-// EIGEN_STRONG_INLINE Packet4i preduxp(const Packet4i* vecs)
-// {
-//   return _mm_hadd_epi32(_mm_hadd_epi32(vecs[0], vecs[1]),_mm_hadd_epi32(vecs[2], vecs[3]));
-// }
 
-template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
-{
-  Packet4f tmp0 = _mm_hadd_ps(a,a);
-  return pfirst(_mm_hadd_ps(tmp0, tmp0));
-}
-
-template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a) { return pfirst(_mm_hadd_pd(a, a)); }
-
-// SSSE3 version:
-// EIGEN_STRONG_INLINE float predux(const Packet4i& a)
-// {
-//   Packet4i tmp0 = _mm_hadd_epi32(a,a);
-//   return pfirst(_mm_hadd_epi32(tmp0, tmp0));
-// }
 #else
-// SSE2 versions
-template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
-{
-  Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
-}
-template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
-{
-  return pfirst(_mm_add_sd(a, _mm_unpackhi_pd(a,a)));
-}
-
 template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
 {
   Packet4f tmp0, tmp1, tmp2;
@@ -450,10 +537,45 @@
 }
 #endif  // SSE3
 
+template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
+{
+  // Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures
+  // (from Nehalem to Haswell)
+// #ifdef EIGEN_VECTORIZE_SSE3
+//   Packet4f tmp = _mm_add_ps(a, vec4f_swizzle1(a,2,3,2,3));
+//   return pfirst<Packet4f>(_mm_hadd_ps(tmp, tmp));
+// #else
+  Packet4f tmp = _mm_add_ps(a, _mm_movehl_ps(a,a));
+  return pfirst<Packet4f>(_mm_add_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+// #endif
+}
+
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
+{
+  // Disable SSE3 _mm_hadd_pd that is extremely slow on all existing Intel's architectures
+  // (from Nehalem to Haswell)
+// #ifdef EIGEN_VECTORIZE_SSE3
+//   return pfirst<Packet2d>(_mm_hadd_pd(a, a));
+// #else
+  return pfirst<Packet2d>(_mm_add_sd(a, _mm_unpackhi_pd(a,a)));
+// #endif
+}
+
+#ifdef EIGEN_VECTORIZE_SSSE3
+template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
+{
+  return _mm_hadd_epi32(_mm_hadd_epi32(vecs[0], vecs[1]),_mm_hadd_epi32(vecs[2], vecs[3]));
+}
+template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
+{
+  Packet4i tmp0 = _mm_hadd_epi32(a,a);
+  return pfirst<Packet4i>(_mm_hadd_epi32(tmp0,tmp0));
+}
+#else
 template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
 {
   Packet4i tmp = _mm_add_epi32(a, _mm_unpackhi_epi64(a,a));
-  return pfirst(tmp) + pfirst(_mm_shuffle_epi32(tmp, 1));
+  return pfirst(tmp) + pfirst<Packet4i>(_mm_shuffle_epi32(tmp, 1));
 }
 
 template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
@@ -469,18 +591,18 @@
   tmp0 = _mm_unpackhi_epi64(tmp0, tmp1);
   return _mm_add_epi32(tmp0, tmp2);
 }
-
+#endif
 // Other reduction functions:
 
 // mul
 template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
 {
   Packet4f tmp = _mm_mul_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_mul_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+  return pfirst<Packet4f>(_mm_mul_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
 }
 template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
 {
-  return pfirst(_mm_mul_sd(a, _mm_unpackhi_pd(a,a)));
+  return pfirst<Packet2d>(_mm_mul_sd(a, _mm_unpackhi_pd(a,a)));
 }
 template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
 {
@@ -496,14 +618,18 @@
 template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
 {
   Packet4f tmp = _mm_min_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_min_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+  return pfirst<Packet4f>(_mm_min_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
 }
 template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
 {
-  return pfirst(_mm_min_sd(a, _mm_unpackhi_pd(a,a)));
+  return pfirst<Packet2d>(_mm_min_sd(a, _mm_unpackhi_pd(a,a)));
 }
 template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
 {
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  Packet4i tmp = _mm_min_epi32(a, _mm_shuffle_epi32(a, _MM_SHUFFLE(0,0,3,2)));
+  return pfirst<Packet4i>(_mm_min_epi32(tmp,_mm_shuffle_epi32(tmp, 1)));
+#else
   // after some experiments, it is seems this is the fastest way to implement it
   // for GCC (eg., it does not like using std::min after the pstore !!)
   EIGEN_ALIGN16 int aux[4];
@@ -511,20 +637,25 @@
   int aux0 = aux[0]<aux[1] ? aux[0] : aux[1];
   int aux2 = aux[2]<aux[3] ? aux[2] : aux[3];
   return aux0<aux2 ? aux0 : aux2;
+#endif // EIGEN_VECTORIZE_SSE4_1
 }
 
 // max
 template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
 {
   Packet4f tmp = _mm_max_ps(a, _mm_movehl_ps(a,a));
-  return pfirst(_mm_max_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
+  return pfirst<Packet4f>(_mm_max_ss(tmp, _mm_shuffle_ps(tmp,tmp, 1)));
 }
 template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
 {
-  return pfirst(_mm_max_sd(a, _mm_unpackhi_pd(a,a)));
+  return pfirst<Packet2d>(_mm_max_sd(a, _mm_unpackhi_pd(a,a)));
 }
 template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
 {
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  Packet4i tmp = _mm_max_epi32(a, _mm_shuffle_epi32(a, _MM_SHUFFLE(0,0,3,2)));
+  return pfirst<Packet4i>(_mm_max_epi32(tmp,_mm_shuffle_epi32(tmp, 1)));
+#else
   // after some experiments, it is seems this is the fastest way to implement it
   // for GCC (eg., it does not like using std::min after the pstore !!)
   EIGEN_ALIGN16 int aux[4];
@@ -532,9 +663,10 @@
   int aux0 = aux[0]>aux[1] ? aux[0] : aux[1];
   int aux2 = aux[2]>aux[3] ? aux[2] : aux[3];
   return aux0>aux2 ? aux0 : aux2;
+#endif // EIGEN_VECTORIZE_SSE4_1
 }
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 // template <> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f&  a, const Packet4f&  b, const Packet4f&  c)
 // {
 //   Packet4f res = b;
@@ -642,8 +774,122 @@
 };
 #endif
 
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  _MM_TRANSPOSE4_PS(kernel.packet[0], kernel.packet[1], kernel.packet[2], kernel.packet[3]);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  __m128d tmp = _mm_unpackhi_pd(kernel.packet[0], kernel.packet[1]);
+  kernel.packet[0] = _mm_unpacklo_pd(kernel.packet[0], kernel.packet[1]);
+  kernel.packet[1] = tmp;
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  __m128i T0 = _mm_unpacklo_epi32(kernel.packet[0], kernel.packet[1]);
+  __m128i T1 = _mm_unpacklo_epi32(kernel.packet[2], kernel.packet[3]);
+  __m128i T2 = _mm_unpackhi_epi32(kernel.packet[0], kernel.packet[1]);
+  __m128i T3 = _mm_unpackhi_epi32(kernel.packet[2], kernel.packet[3]);
+
+  kernel.packet[0] = _mm_unpacklo_epi64(T0, T1);
+  kernel.packet[1] = _mm_unpackhi_epi64(T0, T1);
+  kernel.packet[2] = _mm_unpacklo_epi64(T2, T3);
+  kernel.packet[3] = _mm_unpackhi_epi64(T2, T3);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i select = _mm_set_epi32(ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m128i false_mask = _mm_cmpeq_epi32(select, zero);
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blendv_epi8(thenPacket, elsePacket, false_mask);
+#else
+  return _mm_or_si128(_mm_andnot_si128(false_mask, thenPacket), _mm_and_si128(false_mask, elsePacket));
+#endif
+}
+template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) {
+  const __m128 zero = _mm_setzero_ps();
+  const __m128 select = _mm_set_ps(ifPacket.select[3], ifPacket.select[2], ifPacket.select[1], ifPacket.select[0]);
+  __m128 false_mask = _mm_cmpeq_ps(select, zero);
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blendv_ps(thenPacket, elsePacket, false_mask);
+#else
+  return _mm_or_ps(_mm_andnot_ps(false_mask, thenPacket), _mm_and_ps(false_mask, elsePacket));
+#endif
+}
+template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) {
+  const __m128d zero = _mm_setzero_pd();
+  const __m128d select = _mm_set_pd(ifPacket.select[1], ifPacket.select[0]);
+  __m128d false_mask = _mm_cmpeq_pd(select, zero);
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blendv_pd(thenPacket, elsePacket, false_mask);
+#else
+  return _mm_or_pd(_mm_andnot_pd(false_mask, thenPacket), _mm_and_pd(false_mask, elsePacket));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pinsertfirst(const Packet4f& a, float b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_ps(a,pset1<Packet4f>(b),1);
+#else
+  return _mm_move_ss(a, _mm_load_ss(&b));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pinsertfirst(const Packet2d& a, double b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_pd(a,pset1<Packet2d>(b),1);
+#else
+  return _mm_move_sd(a, _mm_load_sd(&b));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pinsertlast(const Packet4f& a, float b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_ps(a,pset1<Packet4f>(b),(1<<3));
+#else
+  const Packet4f mask = _mm_castsi128_ps(_mm_setr_epi32(0x0,0x0,0x0,0xFFFFFFFF));
+  return _mm_or_ps(_mm_andnot_ps(mask, a), _mm_and_ps(mask, pset1<Packet4f>(b)));
+#endif
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pinsertlast(const Packet2d& a, double b)
+{
+#ifdef EIGEN_VECTORIZE_SSE4_1
+  return _mm_blend_pd(a,pset1<Packet2d>(b),(1<<1));
+#else
+  const Packet2d mask = _mm_castsi128_pd(_mm_setr_epi32(0x0,0x0,0xFFFFFFFF,0xFFFFFFFF));
+  return _mm_or_pd(_mm_andnot_pd(mask, a), _mm_and_pd(mask, pset1<Packet2d>(b)));
+#endif
+}
+
+// Scalar path for pmadd with FMA to ensure consistency with vectorized path.
+#ifdef __FMA__
+template<> EIGEN_STRONG_INLINE float pmadd(const float& a, const float& b, const float& c) {
+  return ::fmaf(a,b,c);
+}
+template<> EIGEN_STRONG_INLINE double pmadd(const double& a, const double& b, const double& c) {
+  return ::fma(a,b,c);
+}
+#endif
+
 } // end namespace internal
 
 } // end namespace Eigen
 
+#if EIGEN_COMP_PGI
+// PGI++ does not define the following intrinsics in C++ mode.
+static inline __m128  _mm_castpd_ps   (__m128d x) { return reinterpret_cast<__m128&>(x);  }
+static inline __m128i _mm_castpd_si128(__m128d x) { return reinterpret_cast<__m128i&>(x); }
+static inline __m128d _mm_castps_pd   (__m128  x) { return reinterpret_cast<__m128d&>(x); }
+static inline __m128i _mm_castps_si128(__m128  x) { return reinterpret_cast<__m128i&>(x); }
+static inline __m128  _mm_castsi128_ps(__m128i x) { return reinterpret_cast<__m128&>(x);  }
+static inline __m128d _mm_castsi128_pd(__m128i x) { return reinterpret_cast<__m128d&>(x); }
+#endif
+
 #endif // EIGEN_PACKET_MATH_SSE_H
diff --git a/Eigen/src/Core/arch/SSE/TypeCasting.h b/Eigen/src/Core/arch/SSE/TypeCasting.h
new file mode 100644
index 0000000..c6ca8c7
--- /dev/null
+++ b/Eigen/src/Core/arch/SSE/TypeCasting.h
@@ -0,0 +1,77 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TYPE_CASTING_SSE_H
+#define EIGEN_TYPE_CASTING_SSE_H
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_VECTORIZE_AVX
+template <>
+struct type_casting_traits<float, int> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<int, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<double, float> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 2,
+    TgtCoeffRatio = 1
+  };
+};
+
+template <>
+struct type_casting_traits<float, double> {
+  enum {
+    VectorizedCast = 1,
+    SrcCoeffRatio = 1,
+    TgtCoeffRatio = 2
+  };
+};
+#endif
+
+template<> EIGEN_STRONG_INLINE Packet4i pcast<Packet4f, Packet4i>(const Packet4f& a) {
+  return _mm_cvttps_epi32(a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet4i, Packet4f>(const Packet4i& a) {
+  return _mm_cvtepi32_ps(a);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pcast<Packet2d, Packet4f>(const Packet2d& a, const Packet2d& b) {
+  return _mm_shuffle_ps(_mm_cvtpd_ps(a), _mm_cvtpd_ps(b), (1 << 2) | (1 << 6));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pcast<Packet4f, Packet2d>(const Packet4f& a) {
+  // Simply discard the second half of the input
+  return _mm_cvtps_pd(a);
+}
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TYPE_CASTING_SSE_H
diff --git a/Eigen/src/Core/arch/ZVector/Complex.h b/Eigen/src/Core/arch/ZVector/Complex.h
new file mode 100644
index 0000000..1bfb733
--- /dev/null
+++ b/Eigen/src/Core/arch/ZVector/Complex.h
@@ -0,0 +1,397 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLEX32_ALTIVEC_H
+#define EIGEN_COMPLEX32_ALTIVEC_H
+
+namespace Eigen {
+
+namespace internal {
+
+static Packet2ul  p2ul_CONJ_XOR1 = (Packet2ul) vec_sld((Packet4ui) p2d_ZERO_, (Packet4ui) p2l_ZERO, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+static Packet2ul  p2ul_CONJ_XOR2 = (Packet2ul) vec_sld((Packet4ui) p2l_ZERO,  (Packet4ui) p2d_ZERO_, 8);//{ 0x8000000000000000, 0x0000000000000000 };
+
+struct Packet1cd
+{
+  EIGEN_STRONG_INLINE Packet1cd() {}
+  EIGEN_STRONG_INLINE explicit Packet1cd(const Packet2d& a) : v(a) {}
+  Packet2d v;
+};
+
+struct Packet2cf
+{
+  EIGEN_STRONG_INLINE Packet2cf() {}
+  EIGEN_STRONG_INLINE explicit Packet2cf(const Packet4f& a) : v(a) {}
+  union {
+    Packet4f v;
+    Packet1cd cd[2];
+  };
+};
+
+template<> struct packet_traits<std::complex<float> >  : default_packet_traits
+{
+  typedef Packet2cf type;
+  typedef Packet2cf half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 2,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasBlend  = 1,
+    HasSetLinear = 0
+  };
+};
+
+
+template<> struct packet_traits<std::complex<double> >  : default_packet_traits
+{
+  typedef Packet1cd type;
+  typedef Packet1cd half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 1,
+    HasHalfPacket = 0,
+
+    HasAdd    = 1,
+    HasSub    = 1,
+    HasMul    = 1,
+    HasDiv    = 1,
+    HasNegate = 1,
+    HasAbs    = 0,
+    HasAbs2   = 0,
+    HasMin    = 0,
+    HasMax    = 0,
+    HasSetLinear = 0
+  };
+};
+
+template<> struct unpacket_traits<Packet2cf> { typedef std::complex<float>  type; enum {size=2, alignment=Aligned16}; typedef Packet2cf half; };
+template<> struct unpacket_traits<Packet1cd> { typedef std::complex<double> type; enum {size=1, alignment=Aligned16}; typedef Packet1cd half; };
+
+/* Forward declaration */
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel);
+
+template<> EIGEN_STRONG_INLINE Packet2cf pload <Packet2cf>(const std::complex<float>* from)  { EIGEN_DEBUG_ALIGNED_LOAD return Packet2cf(pload<Packet4f>((const float*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pload <Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_ALIGNED_LOAD return Packet1cd(pload<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploadu<Packet2cf>(const std::complex<float>* from)  { EIGEN_DEBUG_UNALIGNED_LOAD return Packet2cf(ploadu<Packet4f>((const float*)from)); }
+template<> EIGEN_STRONG_INLINE Packet1cd ploadu<Packet1cd>(const std::complex<double>* from) { EIGEN_DEBUG_UNALIGNED_LOAD return Packet1cd(ploadu<Packet2d>((const double*)from)); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<float> >(std::complex<float> *     to, const Packet2cf& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstore <std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_ALIGNED_STORE pstore((double*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<float> >(std::complex<float> *     to, const Packet2cf& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((float*)to, from.v); }
+template<> EIGEN_STRONG_INLINE void pstoreu<std::complex<double> >(std::complex<double> *   to, const Packet1cd& from) { EIGEN_DEBUG_UNALIGNED_STORE pstoreu((double*)to, from.v); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd pset1<Packet1cd>(const std::complex<double>&  from)
+{ /* here we really have to use unaligned loads :( */ return ploadu<Packet1cd>(&from); }
+
+template<> EIGEN_STRONG_INLINE Packet2cf pset1<Packet2cf>(const std::complex<float>&  from)
+{
+  Packet2cf res;
+  res.cd[0] = Packet1cd(vec_ld2f((const float *)&from));
+  res.cd[1] = res.cd[0];
+  return res;
+}
+template<> EIGEN_DEVICE_FUNC inline Packet2cf pgather<std::complex<float>, Packet2cf>(const std::complex<float>* from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+  return pload<Packet2cf>(af);
+}
+template<> EIGEN_DEVICE_FUNC inline Packet1cd pgather<std::complex<double>, Packet1cd>(const std::complex<double>* from, Index stride EIGEN_UNUSED)
+{
+  return pload<Packet1cd>(from);
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<float>, Packet2cf>(std::complex<float>* to, const Packet2cf& from, Index stride)
+{
+  std::complex<float> EIGEN_ALIGN16 af[2];
+  pstore<std::complex<float> >((std::complex<float> *) af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+template<> EIGEN_DEVICE_FUNC inline void pscatter<std::complex<double>, Packet1cd>(std::complex<double>* to, const Packet1cd& from, Index stride EIGEN_UNUSED)
+{
+  pstore<std::complex<double> >(to, from);
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf padd<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(padd<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd padd<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v + b.v); }
+template<> EIGEN_STRONG_INLINE Packet2cf psub<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(psub<Packet4f>(a.v, b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd psub<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(a.v - b.v); }
+template<> EIGEN_STRONG_INLINE Packet1cd pnegate(const Packet1cd& a) { return Packet1cd(pnegate(Packet2d(a.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pnegate(const Packet2cf& a) { return Packet2cf(pnegate(Packet4f(a.v))); }
+template<> EIGEN_STRONG_INLINE Packet1cd pconj(const Packet1cd& a) { return Packet1cd((Packet2d)vec_xor((Packet2d)a.v, (Packet2d)p2ul_CONJ_XOR2)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pconj(const Packet2cf& a)
+{
+  Packet2cf res;
+  res.v.v4f[0] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0]))).v;
+  res.v.v4f[1] = pconj(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1]))).v;
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pmul<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  Packet2d a_re, a_im, v1, v2;
+
+  // Permute and multiply the real parts of a and b
+  a_re = vec_perm(a.v, a.v, p16uc_PSET64_HI);
+  // Get the imaginary parts of a
+  a_im = vec_perm(a.v, a.v, p16uc_PSET64_LO);
+  // multiply a_re * b
+  v1 = vec_madd(a_re, b.v, p2d_ZERO);
+  // multiply a_im * b and get the conjugate result
+  v2 = vec_madd(a_im, b.v, p2d_ZERO);
+  v2 = (Packet2d) vec_sld((Packet4ui)v2, (Packet4ui)v2, 8);
+  v2 = (Packet2d) vec_xor((Packet2d)v2, (Packet2d) p2ul_CONJ_XOR1);
+
+  return Packet1cd(v1 + v2);
+}
+template<> EIGEN_STRONG_INLINE Packet2cf pmul<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+{
+  Packet2cf res;
+  res.v.v4f[0] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[0])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[0]))).v;
+  res.v.v4f[1] = pmul(Packet1cd(reinterpret_cast<Packet2d>(a.v.v4f[1])), Packet1cd(reinterpret_cast<Packet2d>(b.v.v4f[1]))).v;
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd pand   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pand   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pand<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd por    <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_or(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf por    <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(por<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pxor   <Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_xor(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet2cf pxor   <Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pxor<Packet4f>(a.v,b.v)); }
+template<> EIGEN_STRONG_INLINE Packet1cd pandnot<Packet1cd>(const Packet1cd& a, const Packet1cd& b) { return Packet1cd(vec_and(a.v, vec_nor(b.v,b.v))); }
+template<> EIGEN_STRONG_INLINE Packet2cf pandnot<Packet2cf>(const Packet2cf& a, const Packet2cf& b) { return Packet2cf(pandnot<Packet4f>(a.v,b.v)); }
+
+template<> EIGEN_STRONG_INLINE Packet1cd ploaddup<Packet1cd>(const std::complex<double>*     from) {  return pset1<Packet1cd>(*from); }
+template<> EIGEN_STRONG_INLINE Packet2cf ploaddup<Packet2cf>(const std::complex<float>*      from) {  return pset1<Packet2cf>(*from); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<float> >(const std::complex<float> *     addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<std::complex<double> >(const std::complex<double> *   addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE std::complex<double>  pfirst<Packet1cd>(const Packet1cd& a)
+{
+  std::complex<double> EIGEN_ALIGN16 res;
+  pstore<std::complex<double> >(&res, a);
+
+  return res;
+}
+template<> EIGEN_STRONG_INLINE std::complex<float>  pfirst<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> EIGEN_ALIGN16 res[2];
+  pstore<std::complex<float> >(res, a);
+
+  return res[0];
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preverse(const Packet1cd& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet2cf preverse(const Packet2cf& a)
+{
+  Packet2cf res;
+  res.cd[0] = a.cd[1];
+  res.cd[1] = a.cd[0];
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux<Packet1cd>(const Packet1cd& a)
+{
+  return pfirst(a);
+}
+template<> EIGEN_STRONG_INLINE std::complex<float> predux<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> res;
+  Packet1cd b = padd<Packet1cd>(a.cd[0], a.cd[1]);
+  vec_st2f(b.v, (float*)&res);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet1cd preduxp<Packet1cd>(const Packet1cd* vecs)
+{
+  return vecs[0];
+}
+template<> EIGEN_STRONG_INLINE Packet2cf preduxp<Packet2cf>(const Packet2cf* vecs)
+{
+  PacketBlock<Packet2cf,2> transpose;
+  transpose.packet[0] = vecs[0];
+  transpose.packet[1] = vecs[1];
+  ptranspose(transpose);
+
+  return padd<Packet2cf>(transpose.packet[0], transpose.packet[1]);
+} 
+
+template<> EIGEN_STRONG_INLINE std::complex<double> predux_mul<Packet1cd>(const Packet1cd& a)
+{
+  return pfirst(a);
+}
+template<> EIGEN_STRONG_INLINE std::complex<float> predux_mul<Packet2cf>(const Packet2cf& a)
+{
+  std::complex<float> res;
+  Packet1cd b = pmul<Packet1cd>(a.cd[0], a.cd[1]);
+  vec_st2f(b.v, (float*)&res);
+  return res;
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet1cd>
+{
+  static EIGEN_STRONG_INLINE void run(Packet1cd& /*first*/, const Packet1cd& /*second*/)
+  {
+    // FIXME is it sure we never have to align a Packet1cd?
+    // Even though a std::complex<double> has 16 bytes, it is not necessarily aligned on a 16 bytes boundary...
+  }
+};
+
+template<int Offset>
+struct palign_impl<Offset,Packet2cf>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2cf& first, const Packet2cf& second)
+  {
+    if (Offset == 1) {
+      first.cd[0] = first.cd[1];
+      first.cd[1] = second.cd[0];
+    }
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, false,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,false>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet1cd, Packet1cd, true,true>
+{
+  EIGEN_STRONG_INLINE Packet1cd pmadd(const Packet1cd& x, const Packet1cd& y, const Packet1cd& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet1cd pmul(const Packet1cd& a, const Packet1cd& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, false,true>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return internal::pmul(a, pconj(b));
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, true,false>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return internal::pmul(pconj(a), b);
+  }
+};
+
+template<> struct conj_helper<Packet2cf, Packet2cf, true,true>
+{
+  EIGEN_STRONG_INLINE Packet2cf pmadd(const Packet2cf& x, const Packet2cf& y, const Packet2cf& c) const
+  { return padd(pmul(x,y),c); }
+
+  EIGEN_STRONG_INLINE Packet2cf pmul(const Packet2cf& a, const Packet2cf& b) const
+  {
+    return pconj(internal::pmul(a, b));
+  }
+};
+
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet2cf,Packet4f)
+EIGEN_MAKE_CONJ_HELPER_CPLX_REAL(Packet1cd,Packet2d)
+
+template<> EIGEN_STRONG_INLINE Packet1cd pdiv<Packet1cd>(const Packet1cd& a, const Packet1cd& b)
+{
+  // TODO optimize it for AltiVec
+  Packet1cd res = conj_helper<Packet1cd,Packet1cd,false,true>().pmul(a,b);
+  Packet2d s = vec_madd(b.v, b.v, p2d_ZERO_);
+  return Packet1cd(pdiv(res.v, s + vec_perm(s, s, p16uc_REVERSE64)));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pdiv<Packet2cf>(const Packet2cf& a, const Packet2cf& b)
+{
+  // TODO optimize it for AltiVec
+  Packet2cf res;
+  res.cd[0] = pdiv<Packet1cd>(a.cd[0], b.cd[0]);
+  res.cd[1] = pdiv<Packet1cd>(a.cd[1], b.cd[1]);
+  return res;
+}
+
+EIGEN_STRONG_INLINE Packet1cd pcplxflip/*<Packet1cd>*/(const Packet1cd& x)
+{
+  return Packet1cd(preverse(Packet2d(x.v)));
+}
+
+EIGEN_STRONG_INLINE Packet2cf pcplxflip/*<Packet2cf>*/(const Packet2cf& x)
+{
+  Packet2cf res;
+  res.cd[0] = pcplxflip(x.cd[0]);
+  res.cd[1] = pcplxflip(x.cd[1]);
+  return res;
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet1cd,2>& kernel)
+{
+  Packet2d tmp = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_HI);
+  kernel.packet[1].v = vec_perm(kernel.packet[0].v, kernel.packet[1].v, p16uc_TRANSPOSE64_LO);
+  kernel.packet[0].v = tmp;
+}
+
+EIGEN_STRONG_INLINE void ptranspose(PacketBlock<Packet2cf,2>& kernel)
+{
+  Packet1cd tmp = kernel.packet[0].cd[1];
+  kernel.packet[0].cd[1] = kernel.packet[1].cd[0];
+  kernel.packet[1].cd[0] = tmp;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2cf pblend(const Selector<2>& ifPacket, const Packet2cf& thenPacket, const Packet2cf& elsePacket) {
+  Packet2cf result;
+  const Selector<4> ifPacket4 = { ifPacket.select[0], ifPacket.select[0], ifPacket.select[1], ifPacket.select[1] };
+  result.v = pblend<Packet4f>(ifPacket4, thenPacket.v, elsePacket.v);
+  return result;
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_COMPLEX32_ALTIVEC_H
diff --git a/Eigen/src/Core/arch/ZVector/MathFunctions.h b/Eigen/src/Core/arch/ZVector/MathFunctions.h
new file mode 100644
index 0000000..5c7aa72
--- /dev/null
+++ b/Eigen/src/Core/arch/ZVector/MathFunctions.h
@@ -0,0 +1,137 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2007 Julien Pommier
+// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+/* The sin, cos, exp, and log functions of this file come from
+ * Julien Pommier's sse math library: http://gruntthepeon.free.fr/ssemath/
+ */
+
+#ifndef EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+#define EIGEN_MATH_FUNCTIONS_ALTIVEC_H
+
+namespace Eigen {
+
+namespace internal {
+
+static _EIGEN_DECLARE_CONST_Packet2d(1 , 1.0);
+static _EIGEN_DECLARE_CONST_Packet2d(2 , 2.0);
+static _EIGEN_DECLARE_CONST_Packet2d(half, 0.5);
+
+static _EIGEN_DECLARE_CONST_Packet2d(exp_hi,  709.437);
+static _EIGEN_DECLARE_CONST_Packet2d(exp_lo, -709.436139303);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_LOG2EF, 1.4426950408889634073599);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p0, 1.26177193074810590878e-4);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p1, 3.02994407707441961300e-2);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_p2, 9.99999999999999999910e-1);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q0, 3.00198505138664455042e-6);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q1, 2.52448340349684104192e-3);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q2, 2.27265548208155028766e-1);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_q3, 2.00000000000000000009e0);
+
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C1, 0.693145751953125);
+static _EIGEN_DECLARE_CONST_Packet2d(cephes_exp_C2, 1.42860682030941723212e-6);
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d pexp<Packet2d>(const Packet2d& _x)
+{
+  Packet2d x = _x;
+
+  Packet2d tmp, fx;
+  Packet2l emm0;
+
+  // clamp x
+  x = pmax(pmin(x, p2d_exp_hi), p2d_exp_lo);
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = pmadd(p2d_cephes_LOG2EF, x, p2d_half);
+
+  fx = vec_floor(fx);
+
+  tmp = pmul(fx, p2d_cephes_exp_C1);
+  Packet2d z = pmul(fx, p2d_cephes_exp_C2);
+  x = psub(x, tmp);
+  x = psub(x, z);
+
+  Packet2d x2 = pmul(x,x);
+
+  Packet2d px = p2d_cephes_exp_p0;
+  px = pmadd(px, x2, p2d_cephes_exp_p1);
+  px = pmadd(px, x2, p2d_cephes_exp_p2);
+  px = pmul (px, x);
+
+  Packet2d qx = p2d_cephes_exp_q0;
+  qx = pmadd(qx, x2, p2d_cephes_exp_q1);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q2);
+  qx = pmadd(qx, x2, p2d_cephes_exp_q3);
+
+  x = pdiv(px,psub(qx,px));
+  x = pmadd(p2d_2,x,p2d_1);
+
+  // build 2^n
+  emm0 = vec_ctsl(fx, 0);
+
+  static const Packet2l p2l_1023 = { 1023, 1023 };
+  static const Packet2ul p2ul_52 = { 52, 52 };
+
+  emm0 = emm0 + p2l_1023;
+  emm0 = emm0 << reinterpret_cast<Packet2l>(p2ul_52);
+
+  // Altivec's max & min operators just drop silent NaNs. Check NaNs in 
+  // inputs and return them unmodified.
+  Packet2ul isnumber_mask = reinterpret_cast<Packet2ul>(vec_cmpeq(_x, _x));
+  return vec_sel(_x, pmax(pmul(x, reinterpret_cast<Packet2d>(emm0)), _x),
+                 isnumber_mask);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f pexp<Packet4f>(const Packet4f& x)
+{
+  Packet4f res;
+  res.v4f[0] = pexp<Packet2d>(x.v4f[0]);
+  res.v4f[1] = pexp<Packet2d>(x.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d psqrt<Packet2d>(const Packet2d& x)
+{
+  return  __builtin_s390_vfsqdb(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f psqrt<Packet4f>(const Packet4f& x)
+{
+  Packet4f res;
+  res.v4f[0] = psqrt<Packet2d>(x.v4f[0]);
+  res.v4f[1] = psqrt<Packet2d>(x.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet2d prsqrt<Packet2d>(const Packet2d& x) {
+  // Unfortunately we can't use the much faster mm_rqsrt_pd since it only provides an approximation.
+  return pset1<Packet2d>(1.0) / psqrt<Packet2d>(x);
+}
+
+template<> EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_UNUSED
+Packet4f prsqrt<Packet4f>(const Packet4f& x) {
+  Packet4f res;
+  res.v4f[0] = prsqrt<Packet2d>(x.v4f[0]);
+  res.v4f[1] = prsqrt<Packet2d>(x.v4f[1]);
+  return res;
+}
+
+}  // end namespace internal
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_MATH_FUNCTIONS_ALTIVEC_H
diff --git a/Eigen/src/Core/arch/ZVector/PacketMath.h b/Eigen/src/Core/arch/ZVector/PacketMath.h
new file mode 100755
index 0000000..57b01fc
--- /dev/null
+++ b/Eigen/src/Core/arch/ZVector/PacketMath.h
@@ -0,0 +1,945 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Konstantinos Margaritis <markos@freevec.org>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_PACKET_MATH_ZVECTOR_H
+#define EIGEN_PACKET_MATH_ZVECTOR_H
+
+#include <stdint.h>
+
+namespace Eigen {
+
+namespace internal {
+
+#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
+#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 4
+#endif
+
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+#endif
+
+#ifndef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#define EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+#endif
+
+#ifndef EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS
+#define EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS  16
+#endif
+
+typedef __vector int                 Packet4i;
+typedef __vector unsigned int        Packet4ui;
+typedef __vector __bool int          Packet4bi;
+typedef __vector short int           Packet8i;
+typedef __vector unsigned char       Packet16uc;
+typedef __vector double              Packet2d;
+typedef __vector unsigned long long  Packet2ul;
+typedef __vector long long           Packet2l;
+
+typedef struct {
+	Packet2d  v4f[2];
+} Packet4f;
+
+typedef union {
+  int32_t   i[4];
+  uint32_t ui[4];
+  int64_t   l[2];
+  uint64_t ul[2];
+  double    d[2];
+  Packet4i  v4i;
+  Packet4ui v4ui;
+  Packet2l  v2l;
+  Packet2ul v2ul;
+  Packet2d  v2d;
+} Packet;
+
+// We don't want to write the same code all the time, but we need to reuse the constants
+// and it doesn't really work to declare them global, so we define macros instead
+
+#define _EIGEN_DECLARE_CONST_FAST_Packet4i(NAME,X) \
+  Packet4i p4i_##NAME = reinterpret_cast<Packet4i>(vec_splat_s32(X))
+
+#define _EIGEN_DECLARE_CONST_FAST_Packet2d(NAME,X) \
+  Packet2d p2d_##NAME = reinterpret_cast<Packet2d>(vec_splat_s64(X))
+
+#define _EIGEN_DECLARE_CONST_FAST_Packet2l(NAME,X) \
+  Packet2l p2l_##NAME = reinterpret_cast<Packet2l>(vec_splat_s64(X))
+
+#define _EIGEN_DECLARE_CONST_Packet4i(NAME,X) \
+  Packet4i p4i_##NAME = pset1<Packet4i>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet2d(NAME,X) \
+  Packet2d p2d_##NAME = pset1<Packet2d>(X)
+
+#define _EIGEN_DECLARE_CONST_Packet2l(NAME,X) \
+  Packet2l p2l_##NAME = pset1<Packet2l>(X)
+
+// These constants are endian-agnostic
+//static _EIGEN_DECLARE_CONST_FAST_Packet4i(ZERO, 0); //{ 0, 0, 0, 0,}
+static _EIGEN_DECLARE_CONST_FAST_Packet4i(ONE, 1); //{ 1, 1, 1, 1}
+
+static _EIGEN_DECLARE_CONST_FAST_Packet2d(ZERO, 0);
+static _EIGEN_DECLARE_CONST_FAST_Packet2l(ZERO, 0);
+static _EIGEN_DECLARE_CONST_FAST_Packet2l(ONE, 1);
+
+static Packet2d p2d_ONE = { 1.0, 1.0 }; 
+static Packet2d p2d_ZERO_ = { -0.0, -0.0 };
+
+static Packet4i p4i_COUNTDOWN = { 0, 1, 2, 3 };
+static Packet4f p4f_COUNTDOWN = { 0.0, 1.0, 2.0, 3.0 };
+static Packet2d p2d_COUNTDOWN = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet16uc>(p2d_ZERO), reinterpret_cast<Packet16uc>(p2d_ONE), 8));
+
+static Packet16uc p16uc_PSET64_HI = { 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };
+static Packet16uc p16uc_DUPLICATE32_HI = { 0,1,2,3, 0,1,2,3, 4,5,6,7, 4,5,6,7 };
+
+// Mask alignment
+#define _EIGEN_MASK_ALIGNMENT	0xfffffffffffffff0
+
+#define _EIGEN_ALIGNED_PTR(x)	((std::ptrdiff_t)(x) & _EIGEN_MASK_ALIGNMENT)
+
+// Handle endianness properly while loading constants
+// Define global static constants:
+
+static Packet16uc p16uc_FORWARD =   { 0,1,2,3, 4,5,6,7, 8,9,10,11, 12,13,14,15 };
+static Packet16uc p16uc_REVERSE32 = { 12,13,14,15, 8,9,10,11, 4,5,6,7, 0,1,2,3 };
+static Packet16uc p16uc_REVERSE64 = { 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+
+static Packet16uc p16uc_PSET32_WODD   = vec_sld((Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 0), (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 2), 8);//{ 0,1,2,3, 0,1,2,3, 8,9,10,11, 8,9,10,11 };
+static Packet16uc p16uc_PSET32_WEVEN  = vec_sld(p16uc_DUPLICATE32_HI, (Packet16uc) vec_splat((Packet4ui)p16uc_FORWARD, 3), 8);//{ 4,5,6,7, 4,5,6,7, 12,13,14,15, 12,13,14,15 };
+/*static Packet16uc p16uc_HALF64_0_16 = vec_sld((Packet16uc)p4i_ZERO, vec_splat((Packet16uc) vec_abs(p4i_MINUS16), 3), 8);      //{ 0,0,0,0, 0,0,0,0, 16,16,16,16, 16,16,16,16};
+
+static Packet16uc p16uc_PSET64_HI = (Packet16uc) vec_mergeh((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 0,1,2,3, 4,5,6,7, 0,1,2,3, 4,5,6,7 };*/
+static Packet16uc p16uc_PSET64_LO = (Packet16uc) vec_mergel((Packet4ui)p16uc_PSET32_WODD, (Packet4ui)p16uc_PSET32_WEVEN);     //{ 8,9,10,11, 12,13,14,15, 8,9,10,11, 12,13,14,15 };
+/*static Packet16uc p16uc_TRANSPOSE64_HI = vec_add(p16uc_PSET64_HI, p16uc_HALF64_0_16);                                         //{ 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
+static Packet16uc p16uc_TRANSPOSE64_LO = vec_add(p16uc_PSET64_LO, p16uc_HALF64_0_16);                                         //{ 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};*/
+static Packet16uc p16uc_TRANSPOSE64_HI = { 0,1,2,3, 4,5,6,7, 16,17,18,19, 20,21,22,23};
+static Packet16uc p16uc_TRANSPOSE64_LO = { 8,9,10,11, 12,13,14,15, 24,25,26,27, 28,29,30,31};
+
+//static Packet16uc p16uc_COMPLEX32_REV = vec_sld(p16uc_REVERSE32, p16uc_REVERSE32, 8);                                         //{ 4,5,6,7, 0,1,2,3, 12,13,14,15, 8,9,10,11 };
+
+//static Packet16uc p16uc_COMPLEX32_REV2 = vec_sld(p16uc_FORWARD, p16uc_FORWARD, 8);                                            //{ 8,9,10,11, 12,13,14,15, 0,1,2,3, 4,5,6,7 };
+
+
+#if EIGEN_HAS_BUILTIN(__builtin_prefetch) || EIGEN_COMP_GNUC
+  #define EIGEN_ZVECTOR_PREFETCH(ADDR) __builtin_prefetch(ADDR);
+#else
+  #define EIGEN_ZVECTOR_PREFETCH(ADDR) asm( "   pfd [%[addr]]\n" :: [addr] "r" (ADDR) : "cc" );
+#endif
+
+template<> struct packet_traits<int>    : default_packet_traits
+{
+  typedef Packet4i type;
+  typedef Packet4i half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size = 4,
+    HasHalfPacket = 0,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct packet_traits<float> : default_packet_traits
+{
+  typedef Packet4f type;
+  typedef Packet4f half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=4,
+    HasHalfPacket = 0,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct packet_traits<double> : default_packet_traits
+{
+  typedef Packet2d type;
+  typedef Packet2d half;
+  enum {
+    Vectorizable = 1,
+    AlignedOnScalar = 1,
+    size=2,
+    HasHalfPacket = 1,
+
+    HasAdd  = 1,
+    HasSub  = 1,
+    HasMul  = 1,
+    HasDiv  = 1,
+    HasMin  = 1,
+    HasMax  = 1,
+    HasAbs  = 1,
+    HasSin  = 0,
+    HasCos  = 0,
+    HasLog  = 0,
+    HasExp  = 1,
+    HasSqrt = 1,
+    HasRsqrt = 1,
+    HasRound = 1,
+    HasFloor = 1,
+    HasCeil = 1,
+    HasNegate = 1,
+    HasBlend = 1
+  };
+};
+
+template<> struct unpacket_traits<Packet4i> { typedef int    type; enum {size=4, alignment=Aligned16}; typedef Packet4i half; };
+template<> struct unpacket_traits<Packet4f> { typedef float  type; enum {size=4, alignment=Aligned16}; typedef Packet4f half; };
+template<> struct unpacket_traits<Packet2d> { typedef double type; enum {size=2, alignment=Aligned16}; typedef Packet2d half; };
+
+/* Forward declaration */
+EIGEN_DEVICE_FUNC inline void ptranspose(PacketBlock<Packet4f,4>& kernel);
+ 
+inline std::ostream & operator <<(std::ostream & s, const Packet4i & v)
+{
+  Packet vt;
+  vt.v4i = v;
+  s << vt.i[0] << ", " << vt.i[1] << ", " << vt.i[2] << ", " << vt.i[3];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet4ui & v)
+{
+  Packet vt;
+  vt.v4ui = v;
+  s << vt.ui[0] << ", " << vt.ui[1] << ", " << vt.ui[2] << ", " << vt.ui[3];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2l & v)
+{
+  Packet vt;
+  vt.v2l = v;
+  s << vt.l[0] << ", " << vt.l[1];
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2ul & v)
+{
+  Packet vt;
+  vt.v2ul = v;
+  s << vt.ul[0] << ", " << vt.ul[1] ;
+  return s;
+}
+
+inline std::ostream & operator <<(std::ostream & s, const Packet2d & v)
+{
+  Packet vt;
+  vt.v2d = v;
+  s << vt.d[0] << ", " << vt.d[1];
+  return s;
+}
+
+/* Helper function to simulate a vec_splat_packet4f
+ */
+template<int element> EIGEN_STRONG_INLINE Packet4f vec_splat_packet4f(const Packet4f&   from)
+{
+  Packet4f splat;
+  switch (element) {
+  case 0:
+    splat.v4f[0] = vec_splat(from.v4f[0], 0);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 1:
+    splat.v4f[0] = vec_splat(from.v4f[0], 1);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 2:
+    splat.v4f[0] = vec_splat(from.v4f[1], 0);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  case 3:
+    splat.v4f[0] = vec_splat(from.v4f[1], 1);
+    splat.v4f[1] = splat.v4f[0];
+    break;
+  }
+  return splat;
+}
+
+template<int Offset>
+struct palign_impl<Offset,Packet4i>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4i& first, const Packet4i& second)
+  {
+    switch (Offset % 4) {
+    case 1:
+      first = vec_sld(first, second, 4); break;
+    case 2:
+      first = vec_sld(first, second, 8); break;
+    case 3:
+      first = vec_sld(first, second, 12); break;
+    }
+  }
+};
+
+/* This is a tricky one, we have to translate float alignment to vector elements of sizeof double
+ */
+template<int Offset>
+struct palign_impl<Offset,Packet4f>
+{
+  static EIGEN_STRONG_INLINE void run(Packet4f& first, const Packet4f& second)
+  {
+    switch (Offset % 4) {
+    case 1:
+      first.v4f[0] = vec_sld(first.v4f[0], first.v4f[1], 8);
+      first.v4f[1] = vec_sld(first.v4f[1], second.v4f[0], 8);
+      break;
+    case 2:
+      first.v4f[0] = first.v4f[1];
+      first.v4f[1] = second.v4f[0];
+      break;
+    case 3:
+      first.v4f[0] = vec_sld(first.v4f[1],  second.v4f[0], 8);
+      first.v4f[1] = vec_sld(second.v4f[0], second.v4f[1], 8);
+      break;
+    }
+  }
+};
+
+
+template<int Offset>
+struct palign_impl<Offset,Packet2d>
+{
+  static EIGEN_STRONG_INLINE void run(Packet2d& first, const Packet2d& second)
+  {
+    if (Offset == 1)
+      first = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(first), reinterpret_cast<Packet4i>(second), 8));
+  }
+};
+
+template<> EIGEN_STRONG_INLINE Packet4i pload<Packet4i>(const int*     from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_LOAD
+  Packet *vfrom;
+  vfrom = (Packet *) from;
+  return vfrom->v4i;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pload<Packet4f>(const float*   from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_LOAD
+  Packet4f vfrom;
+  vfrom.v4f[0] = vec_ld2f(&from[0]);
+  vfrom.v4f[1] = vec_ld2f(&from[2]);
+  return vfrom;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pload<Packet2d>(const double* from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_LOAD
+  Packet *vfrom;
+  vfrom = (Packet *) from;
+  return vfrom->v2d;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<int>(int*       to, const Packet4i& from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_STORE
+  Packet *vto;
+  vto = (Packet *) to;
+  vto->v4i = from;
+}
+
+template<> EIGEN_STRONG_INLINE void pstore<float>(float*   to, const Packet4f& from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_STORE
+  vec_st2f(from.v4f[0], &to[0]);
+  vec_st2f(from.v4f[1], &to[2]);
+}
+
+
+template<> EIGEN_STRONG_INLINE void pstore<double>(double*   to, const Packet2d& from)
+{
+  // FIXME: No intrinsic yet
+  EIGEN_DEBUG_ALIGNED_STORE
+  Packet *vto;
+  vto = (Packet *) to;
+  vto->v2d = from;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pset1<Packet4i>(const int&    from)
+{
+  return vec_splats(from);
+}
+template<> EIGEN_STRONG_INLINE Packet2d pset1<Packet2d>(const double& from) {
+  return vec_splats(from);
+}
+template<> EIGEN_STRONG_INLINE Packet4f pset1<Packet4f>(const float&    from)
+{
+  Packet4f to;
+  to.v4f[0] = pset1<Packet2d>(static_cast<const double&>(from));
+  to.v4f[1] = to.v4f[0];
+  return to;
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4i>(const int *a,
+                      Packet4i& a0, Packet4i& a1, Packet4i& a2, Packet4i& a3)
+{
+  a3 = pload<Packet4i>(a);
+  a0 = vec_splat(a3, 0);
+  a1 = vec_splat(a3, 1);
+  a2 = vec_splat(a3, 2);
+  a3 = vec_splat(a3, 3);
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet4f>(const float *a,
+                      Packet4f& a0, Packet4f& a1, Packet4f& a2, Packet4f& a3)
+{
+  a3 = pload<Packet4f>(a);
+  a0 = vec_splat_packet4f<0>(a3);
+  a1 = vec_splat_packet4f<1>(a3);
+  a2 = vec_splat_packet4f<2>(a3);
+  a3 = vec_splat_packet4f<3>(a3);
+}
+
+template<> EIGEN_STRONG_INLINE void
+pbroadcast4<Packet2d>(const double *a,
+                      Packet2d& a0, Packet2d& a1, Packet2d& a2, Packet2d& a3)
+{
+  a1 = pload<Packet2d>(a);
+  a0 = vec_splat(a1, 0);
+  a1 = vec_splat(a1, 1);
+  a3 = pload<Packet2d>(a+2);
+  a2 = vec_splat(a3, 0);
+  a3 = vec_splat(a3, 1);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet4i pgather<int, Packet4i>(const int* from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return pload<Packet4i>(ai);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet4f pgather<float, Packet4f>(const float* from, Index stride)
+{
+  float EIGEN_ALIGN16 ai[4];
+  ai[0] = from[0*stride];
+  ai[1] = from[1*stride];
+  ai[2] = from[2*stride];
+  ai[3] = from[3*stride];
+ return pload<Packet4f>(ai);
+}
+
+template<> EIGEN_DEVICE_FUNC inline Packet2d pgather<double, Packet2d>(const double* from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  af[0] = from[0*stride];
+  af[1] = from[1*stride];
+ return pload<Packet2d>(af);
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<int, Packet4i>(int* to, const Packet4i& from, Index stride)
+{
+  int EIGEN_ALIGN16 ai[4];
+  pstore<int>((int *)ai, from);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<float, Packet4f>(float* to, const Packet4f& from, Index stride)
+{
+  float EIGEN_ALIGN16 ai[4];
+  pstore<float>((float *)ai, from);
+  to[0*stride] = ai[0];
+  to[1*stride] = ai[1];
+  to[2*stride] = ai[2];
+  to[3*stride] = ai[3];
+}
+
+template<> EIGEN_DEVICE_FUNC inline void pscatter<double, Packet2d>(double* to, const Packet2d& from, Index stride)
+{
+  double EIGEN_ALIGN16 af[2];
+  pstore<double>(af, from);
+  to[0*stride] = af[0];
+  to[1*stride] = af[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i padd<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a + b); }
+template<> EIGEN_STRONG_INLINE Packet4f padd<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] + b.v4f[0];
+  c.v4f[1] = a.v4f[1] + b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d padd<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a + b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i psub<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a - b); }
+template<> EIGEN_STRONG_INLINE Packet4f psub<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] - b.v4f[0];
+  c.v4f[1] = a.v4f[1] - b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d psub<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a - b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pmul<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a * b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmul<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] * b.v4f[0];
+  c.v4f[1] = a.v4f[1] * b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pmul<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a * b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pdiv<Packet4i>(const Packet4i& a, const Packet4i& b) { return (a / b); }
+template<> EIGEN_STRONG_INLINE Packet4f pdiv<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f c;
+  c.v4f[0] = a.v4f[0] / b.v4f[0];
+  c.v4f[1] = a.v4f[1] / b.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pdiv<Packet2d>(const Packet2d& a, const Packet2d& b) { return (a / b); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pnegate(const Packet4i& a) { return (-a); }
+template<> EIGEN_STRONG_INLINE Packet4f pnegate(const Packet4f& a)
+{
+  Packet4f c;
+  c.v4f[0] = -a.v4f[0];
+  c.v4f[1] = -a.v4f[1];
+  return c;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pnegate(const Packet2d& a) { return (-a); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pconj(const Packet4i& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet4f pconj(const Packet4f& a) { return a; }
+template<> EIGEN_STRONG_INLINE Packet2d pconj(const Packet2d& a) { return a; }
+
+template<> EIGEN_STRONG_INLINE Packet4i pmadd(const Packet4i& a, const Packet4i& b, const Packet4i& c) { return padd<Packet4i>(pmul<Packet4i>(a, b), c); }
+template<> EIGEN_STRONG_INLINE Packet4f pmadd(const Packet4f& a, const Packet4f& b, const Packet4f& c)
+{
+  Packet4f res;
+  res.v4f[0] = vec_madd(a.v4f[0], b.v4f[0], c.v4f[0]);
+  res.v4f[1] = vec_madd(a.v4f[1], b.v4f[1], c.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pmadd(const Packet2d& a, const Packet2d& b, const Packet2d& c) { return vec_madd(a, b, c); }
+
+template<> EIGEN_STRONG_INLINE Packet4i plset<Packet4i>(const int& a)    { return padd<Packet4i>(pset1<Packet4i>(a), p4i_COUNTDOWN); }
+template<> EIGEN_STRONG_INLINE Packet4f plset<Packet4f>(const float& a)  { return padd<Packet4f>(pset1<Packet4f>(a), p4f_COUNTDOWN); }
+template<> EIGEN_STRONG_INLINE Packet2d plset<Packet2d>(const double& a) { return padd<Packet2d>(pset1<Packet2d>(a), p2d_COUNTDOWN); }
+
+template<> EIGEN_STRONG_INLINE Packet4i pmin<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_min(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pmin<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_min(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmin<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pmin(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pmin(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pmax<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_max(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pmax<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_max(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pmax<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pmax(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pmax(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pand<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_and(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pand<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pand<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i por<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_or(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d por<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_or(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f por<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pxor<Packet4i>(const Packet4i& a, const Packet4i& b) { return vec_xor(a, b); }
+template<> EIGEN_STRONG_INLINE Packet2d pxor<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_xor(a, b); }
+template<> EIGEN_STRONG_INLINE Packet4f pxor<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pand(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pand(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pandnot<Packet4i>(const Packet4i& a, const Packet4i& b) { return pand<Packet4i>(a, vec_nor(b, b)); }
+template<> EIGEN_STRONG_INLINE Packet2d pandnot<Packet2d>(const Packet2d& a, const Packet2d& b) { return vec_and(a, vec_nor(b, b)); }
+template<> EIGEN_STRONG_INLINE Packet4f pandnot<Packet4f>(const Packet4f& a, const Packet4f& b)
+{
+  Packet4f res;
+  res.v4f[0] = pandnot(a.v4f[0], b.v4f[0]);
+  res.v4f[1] = pandnot(a.v4f[1], b.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pround<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_round(a.v4f[0]);
+  res.v4f[1] = vec_round(a.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pround<Packet2d>(const Packet2d& a) { return vec_round(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pceil<Packet4f>(const  Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_ceil(a.v4f[0]);
+  res.v4f[1] = vec_ceil(a.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pceil<Packet2d>(const  Packet2d& a) { return vec_ceil(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pfloor<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = vec_floor(a.v4f[0]);
+  res.v4f[1] = vec_floor(a.v4f[1]);
+  return res;
+}
+template<> EIGEN_STRONG_INLINE Packet2d pfloor<Packet2d>(const Packet2d& a) { return vec_floor(a); }
+
+template<> EIGEN_STRONG_INLINE Packet4i ploadu<Packet4i>(const int*       from) { return pload<Packet4i>(from); }
+template<> EIGEN_STRONG_INLINE Packet4f ploadu<Packet4f>(const float*     from) { return pload<Packet4f>(from); }
+template<> EIGEN_STRONG_INLINE Packet2d ploadu<Packet2d>(const double*    from) { return pload<Packet2d>(from); }
+
+
+template<> EIGEN_STRONG_INLINE Packet4i ploaddup<Packet4i>(const int*     from)
+{
+  Packet4i p = pload<Packet4i>(from);
+  return vec_perm(p, p, p16uc_DUPLICATE32_HI);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f ploaddup<Packet4f>(const float*    from)
+{
+  Packet4f p = pload<Packet4f>(from);
+  p.v4f[1] = vec_splat(p.v4f[0], 1);
+  p.v4f[0] = vec_splat(p.v4f[0], 0);
+  return p;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d ploaddup<Packet2d>(const double*   from)
+{
+  Packet2d p = pload<Packet2d>(from);
+  return vec_perm(p, p, p16uc_PSET64_HI);
+}
+
+template<> EIGEN_STRONG_INLINE void pstoreu<int>(int*        to, const Packet4i& from) { pstore<int>(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<float>(float*    to, const Packet4f& from) { pstore<float>(to, from); }
+template<> EIGEN_STRONG_INLINE void pstoreu<double>(double*  to, const Packet2d& from) { pstore<double>(to, from); }
+
+template<> EIGEN_STRONG_INLINE void prefetch<int>(const int*       addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<float>(const float*   addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+template<> EIGEN_STRONG_INLINE void prefetch<double>(const double* addr) { EIGEN_ZVECTOR_PREFETCH(addr); }
+
+template<> EIGEN_STRONG_INLINE int    pfirst<Packet4i>(const Packet4i& a) { int    EIGEN_ALIGN16 x[4]; pstore(x, a); return x[0]; }
+template<> EIGEN_STRONG_INLINE float  pfirst<Packet4f>(const Packet4f& a) { float  EIGEN_ALIGN16 x[2]; vec_st2f(a.v4f[0], &x[0]); return x[0]; }
+template<> EIGEN_STRONG_INLINE double pfirst<Packet2d>(const Packet2d& a) { double EIGEN_ALIGN16 x[2]; pstore(x, a); return x[0]; }
+
+template<> EIGEN_STRONG_INLINE Packet4i preverse(const Packet4i& a)
+{
+  return reinterpret_cast<Packet4i>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE32));
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d preverse(const Packet2d& a)
+{
+  return reinterpret_cast<Packet2d>(vec_perm(reinterpret_cast<Packet16uc>(a), reinterpret_cast<Packet16uc>(a), p16uc_REVERSE64));
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f preverse(const Packet4f& a)
+{
+  Packet4f rev;
+  rev.v4f[0] = preverse<Packet2d>(a.v4f[1]);
+  rev.v4f[1] = preverse<Packet2d>(a.v4f[0]);
+  return rev;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pabs<Packet4i>(const Packet4i& a) { return vec_abs(a); }
+template<> EIGEN_STRONG_INLINE Packet2d pabs<Packet2d>(const Packet2d& a) { return vec_abs(a); }
+template<> EIGEN_STRONG_INLINE Packet4f pabs<Packet4f>(const Packet4f& a)
+{
+  Packet4f res;
+  res.v4f[0] = pabs(a.v4f[0]);
+  res.v4f[1] = pabs(a.v4f[1]);
+  return res;
+}
+
+template<> EIGEN_STRONG_INLINE int predux<Packet4i>(const Packet4i& a)
+{
+  Packet4i b, sum;
+  b   = vec_sld(a, a, 8);
+  sum = padd<Packet4i>(a, b);
+  b   = vec_sld(sum, sum, 4);
+  sum = padd<Packet4i>(sum, b);
+  return pfirst(sum);
+}
+
+template<> EIGEN_STRONG_INLINE double predux<Packet2d>(const Packet2d& a)
+{
+  Packet2d b, sum;
+  b   = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8));
+  sum = padd<Packet2d>(a, b);
+  return pfirst(sum);
+}
+template<> EIGEN_STRONG_INLINE float predux<Packet4f>(const Packet4f& a)
+{
+  Packet2d sum;
+  sum = padd<Packet2d>(a.v4f[0], a.v4f[1]);
+  double first = predux<Packet2d>(sum);
+  return static_cast<float>(first);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i preduxp<Packet4i>(const Packet4i* vecs)
+{
+  Packet4i v[4], sum[4];
+
+  // It's easier and faster to transpose then add as columns
+  // Check: http://www.freevec.org/function/matrix_4x4_transpose_floats for explanation
+  // Do the transpose, first set of moves
+  v[0] = vec_mergeh(vecs[0], vecs[2]);
+  v[1] = vec_mergel(vecs[0], vecs[2]);
+  v[2] = vec_mergeh(vecs[1], vecs[3]);
+  v[3] = vec_mergel(vecs[1], vecs[3]);
+  // Get the resulting vectors
+  sum[0] = vec_mergeh(v[0], v[2]);
+  sum[1] = vec_mergel(v[0], v[2]);
+  sum[2] = vec_mergeh(v[1], v[3]);
+  sum[3] = vec_mergel(v[1], v[3]);
+
+  // Now do the summation:
+  // Lines 0+1
+  sum[0] = padd<Packet4i>(sum[0], sum[1]);
+  // Lines 2+3
+  sum[1] = padd<Packet4i>(sum[2], sum[3]);
+  // Add the results
+  sum[0] = padd<Packet4i>(sum[0], sum[1]);
+
+  return sum[0];
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d preduxp<Packet2d>(const Packet2d* vecs)
+{
+  Packet2d v[2], sum;
+  v[0] = padd<Packet2d>(vecs[0], reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(vecs[0]), reinterpret_cast<Packet4ui>(vecs[0]), 8)));
+  v[1] = padd<Packet2d>(vecs[1], reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(vecs[1]), reinterpret_cast<Packet4ui>(vecs[1]), 8)));
+ 
+  sum = reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4ui>(v[0]), reinterpret_cast<Packet4ui>(v[1]), 8));
+
+  return sum;
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f preduxp<Packet4f>(const Packet4f* vecs)
+{
+  PacketBlock<Packet4f,4> transpose;
+  transpose.packet[0] = vecs[0];
+  transpose.packet[1] = vecs[1];
+  transpose.packet[2] = vecs[2];
+  transpose.packet[3] = vecs[3];
+  ptranspose(transpose);
+
+  Packet4f sum = padd(transpose.packet[0], transpose.packet[1]);
+  sum = padd(sum, transpose.packet[2]);
+  sum = padd(sum, transpose.packet[3]);
+  return sum;
+}
+
+// Other reduction functions:
+// mul
+template<> EIGEN_STRONG_INLINE int predux_mul<Packet4i>(const Packet4i& a)
+{
+  EIGEN_ALIGN16 int aux[4];
+  pstore(aux, a);
+  return aux[0] * aux[1] * aux[2] * aux[3];
+}
+
+template<> EIGEN_STRONG_INLINE double predux_mul<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmul(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_mul<Packet4f>(const Packet4f& a)
+{
+  // Return predux_mul<Packet2d> of the subvectors product
+  return static_cast<float>(pfirst(predux_mul(pmul(a.v4f[0], a.v4f[1]))));
+}
+
+// min
+template<> EIGEN_STRONG_INLINE int predux_min<Packet4i>(const Packet4i& a)
+{
+  Packet4i b, res;
+  b   = pmin<Packet4i>(a, vec_sld(a, a, 8));
+  res = pmin<Packet4i>(b, vec_sld(b, b, 4));
+  return pfirst(res);
+}
+
+template<> EIGEN_STRONG_INLINE double predux_min<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmin<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_min<Packet4f>(const Packet4f& a)
+{
+  Packet2d b, res;
+  b   = pmin<Packet2d>(a.v4f[0], a.v4f[1]);
+  res = pmin<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8)));
+  return static_cast<float>(pfirst(res));
+}
+
+// max
+template<> EIGEN_STRONG_INLINE int predux_max<Packet4i>(const Packet4i& a)
+{
+  Packet4i b, res;
+  b = pmax<Packet4i>(a, vec_sld(a, a, 8));
+  res = pmax<Packet4i>(b, vec_sld(b, b, 4));
+  return pfirst(res);
+}
+
+// max
+template<> EIGEN_STRONG_INLINE double predux_max<Packet2d>(const Packet2d& a)
+{
+  return pfirst(pmax<Packet2d>(a, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(a), reinterpret_cast<Packet4i>(a), 8))));
+}
+
+template<> EIGEN_STRONG_INLINE float predux_max<Packet4f>(const Packet4f& a)
+{
+  Packet2d b, res;
+  b   = pmax<Packet2d>(a.v4f[0], a.v4f[1]);
+  res = pmax<Packet2d>(b, reinterpret_cast<Packet2d>(vec_sld(reinterpret_cast<Packet4i>(b), reinterpret_cast<Packet4i>(b), 8)));
+  return static_cast<float>(pfirst(res));
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4i,4>& kernel) {
+  Packet4i t0 = vec_mergeh(kernel.packet[0], kernel.packet[2]);
+  Packet4i t1 = vec_mergel(kernel.packet[0], kernel.packet[2]);
+  Packet4i t2 = vec_mergeh(kernel.packet[1], kernel.packet[3]);
+  Packet4i t3 = vec_mergel(kernel.packet[1], kernel.packet[3]);
+  kernel.packet[0] = vec_mergeh(t0, t2);
+  kernel.packet[1] = vec_mergel(t0, t2);
+  kernel.packet[2] = vec_mergeh(t1, t3);
+  kernel.packet[3] = vec_mergel(t1, t3);
+}
+
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet2d,2>& kernel) {
+  Packet2d t0 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_HI);
+  Packet2d t1 = vec_perm(kernel.packet[0], kernel.packet[1], p16uc_TRANSPOSE64_LO);
+  kernel.packet[0] = t0;
+  kernel.packet[1] = t1;
+}
+
+/* Split the Packet4f PacketBlock into 4 Packet2d PacketBlocks and transpose each one
+ */
+EIGEN_DEVICE_FUNC inline void
+ptranspose(PacketBlock<Packet4f,4>& kernel) {
+  PacketBlock<Packet2d,2> t0,t1,t2,t3;
+  // copy top-left 2x2 Packet2d block
+  t0.packet[0] = kernel.packet[0].v4f[0];
+  t0.packet[1] = kernel.packet[1].v4f[0];
+
+  // copy top-right 2x2 Packet2d block
+  t1.packet[0] = kernel.packet[0].v4f[1];
+  t1.packet[1] = kernel.packet[1].v4f[1];
+
+  // copy bottom-left 2x2 Packet2d block
+  t2.packet[0] = kernel.packet[2].v4f[0];
+  t2.packet[1] = kernel.packet[3].v4f[0];
+
+  // copy bottom-right 2x2 Packet2d block
+  t3.packet[0] = kernel.packet[2].v4f[1];
+  t3.packet[1] = kernel.packet[3].v4f[1];
+
+  // Transpose all 2x2 blocks
+  ptranspose(t0);
+  ptranspose(t1);
+  ptranspose(t2);
+  ptranspose(t3);
+
+  // Copy back transposed blocks, but exchange t1 and t2 due to transposition
+  kernel.packet[0].v4f[0] = t0.packet[0];
+  kernel.packet[0].v4f[1] = t2.packet[0];
+  kernel.packet[1].v4f[0] = t0.packet[1];
+  kernel.packet[1].v4f[1] = t2.packet[1];
+  kernel.packet[2].v4f[0] = t1.packet[0];
+  kernel.packet[2].v4f[1] = t3.packet[0];
+  kernel.packet[3].v4f[0] = t1.packet[1];
+  kernel.packet[3].v4f[1] = t3.packet[1];
+}
+
+template<> EIGEN_STRONG_INLINE Packet4i pblend(const Selector<4>& ifPacket, const Packet4i& thenPacket, const Packet4i& elsePacket) {
+  Packet4ui select = { ifPacket.select[0], ifPacket.select[1], ifPacket.select[2], ifPacket.select[3] };
+  Packet4ui mask = vec_cmpeq(select, reinterpret_cast<Packet4ui>(p4i_ONE));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+template<> EIGEN_STRONG_INLINE Packet4f pblend(const Selector<4>& ifPacket, const Packet4f& thenPacket, const Packet4f& elsePacket) {
+  Packet2ul select_hi = { ifPacket.select[0], ifPacket.select[1] };
+  Packet2ul select_lo = { ifPacket.select[2], ifPacket.select[3] };
+  Packet2ul mask_hi = vec_cmpeq(select_hi, reinterpret_cast<Packet2ul>(p2l_ONE));
+  Packet2ul mask_lo = vec_cmpeq(select_lo, reinterpret_cast<Packet2ul>(p2l_ONE));
+  Packet4f result;
+  result.v4f[0] = vec_sel(elsePacket.v4f[0], thenPacket.v4f[0], mask_hi);
+  result.v4f[1] = vec_sel(elsePacket.v4f[1], thenPacket.v4f[1], mask_lo);
+  return result;
+}
+
+template<> EIGEN_STRONG_INLINE Packet2d pblend(const Selector<2>& ifPacket, const Packet2d& thenPacket, const Packet2d& elsePacket) {
+  Packet2ul select = { ifPacket.select[0], ifPacket.select[1] };
+  Packet2ul mask = vec_cmpeq(select, reinterpret_cast<Packet2ul>(p2l_ONE));
+  return vec_sel(elsePacket, thenPacket, mask);
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_PACKET_MATH_ZVECTOR_H
diff --git a/Eigen/src/Core/functors/AssignmentFunctors.h b/Eigen/src/Core/functors/AssignmentFunctors.h
new file mode 100644
index 0000000..4153b87
--- /dev/null
+++ b/Eigen/src/Core/functors/AssignmentFunctors.h
@@ -0,0 +1,168 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_ASSIGNMENT_FUNCTORS_H
+#define EIGEN_ASSIGNMENT_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+  
+/** \internal
+  * \brief Template functor for scalar/packet assignment
+  *
+  */
+template<typename DstScalar,typename SrcScalar> struct assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a = b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,b); }
+};
+
+// Empty overload for void type (used by PermutationMatrix)
+template<typename DstScalar> struct assign_op<DstScalar,void> {};
+
+template<typename DstScalar,typename SrcScalar>
+struct functor_traits<assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::Vectorizable && packet_traits<SrcScalar>::Vectorizable
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with addition
+  *
+  */
+template<typename DstScalar,typename SrcScalar> struct add_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(add_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a += b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::padd(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar,typename SrcScalar>
+struct functor_traits<add_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::AddCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasAdd
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with subtraction
+  *
+  */
+template<typename DstScalar,typename SrcScalar> struct sub_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(sub_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a -= b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::psub(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar,typename SrcScalar>
+struct functor_traits<sub_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::AddCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasSub
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with multiplication
+  *
+  */
+template<typename DstScalar, typename SrcScalar=DstScalar>
+struct mul_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(mul_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a *= b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::pmul(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar, typename SrcScalar>
+struct functor_traits<mul_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::MulCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasMul
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with diviving
+  *
+  */
+template<typename DstScalar, typename SrcScalar=DstScalar> struct div_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(div_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(DstScalar& a, const SrcScalar& b) const { a /= b; }
+  
+  template<int Alignment, typename Packet>
+  EIGEN_STRONG_INLINE void assignPacket(DstScalar* a, const Packet& b) const
+  { internal::pstoret<DstScalar,Packet,Alignment>(a,internal::pdiv(internal::ploadt<Packet,Alignment>(a),b)); }
+};
+template<typename DstScalar, typename SrcScalar>
+struct functor_traits<div_assign_op<DstScalar,SrcScalar> > {
+  enum {
+    Cost = NumTraits<DstScalar>::ReadCost + NumTraits<DstScalar>::MulCost,
+    PacketAccess = is_same<DstScalar,SrcScalar>::value && packet_traits<DstScalar>::HasDiv
+  };
+};
+
+/** \internal
+  * \brief Template functor for scalar/packet assignment with swapping
+  *
+  * It works as follow. For a non-vectorized evaluation loop, we have:
+  *   for(i) func(A.coeffRef(i), B.coeff(i));
+  * where B is a SwapWrapper expression. The trick is to make SwapWrapper::coeff behaves like a non-const coeffRef.
+  * Actually, SwapWrapper might not even be needed since even if B is a plain expression, since it has to be writable
+  * B.coeff already returns a const reference to the underlying scalar value.
+  * 
+  * The case of a vectorized loop is more tricky:
+  *   for(i,j) func.assignPacket<A_Align>(&A.coeffRef(i,j), B.packet<B_Align>(i,j));
+  * Here, B must be a SwapWrapper whose packet function actually returns a proxy object holding a Scalar*,
+  * the actual alignment and Packet type.
+  *
+  */
+template<typename Scalar> struct swap_assign_op {
+
+  EIGEN_EMPTY_STRUCT_CTOR(swap_assign_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Scalar& a, const Scalar& b) const
+  {
+#ifdef __CUDACC__
+    // FIXME is there some kind of cuda::swap?
+    Scalar t=b; const_cast<Scalar&>(b)=a; a=t;
+#else
+    using std::swap;
+    swap(a,const_cast<Scalar&>(b));
+#endif
+  }
+};
+template<typename Scalar>
+struct functor_traits<swap_assign_op<Scalar> > {
+  enum {
+    Cost = 3 * NumTraits<Scalar>::ReadCost,
+    PacketAccess = packet_traits<Scalar>::Vectorizable
+  };
+};
+
+} // namespace internal
+
+} // namespace Eigen
+
+#endif // EIGEN_ASSIGNMENT_FUNCTORS_H
diff --git a/Eigen/src/Core/functors/BinaryFunctors.h b/Eigen/src/Core/functors/BinaryFunctors.h
new file mode 100644
index 0000000..3eae6b8
--- /dev/null
+++ b/Eigen/src/Core/functors/BinaryFunctors.h
@@ -0,0 +1,475 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_BINARY_FUNCTORS_H
+#define EIGEN_BINARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+//---------- associative binary functors ----------
+
+template<typename Arg1, typename Arg2>
+struct binary_op_base
+{
+  typedef Arg1 first_argument_type;
+  typedef Arg2 second_argument_type;
+};
+
+/** \internal
+  * \brief Template functor to compute the sum of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::operator+, class VectorwiseOp, DenseBase::sum()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_sum_op : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_sum_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sum_op)
+#else
+  scalar_sum_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a + b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::padd(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_sum_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2, // rough estimate!
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasAdd && packet_traits<RhsScalar>::HasAdd
+    // TODO vectorize mixed sum
+  };
+};
+
+/** \internal
+  * \brief Template specialization to deprecate the summation of boolean expressions.
+  * This is required to solve Bug 426.
+  * \sa DenseBase::count(), DenseBase::any(), ArrayBase::cast(), MatrixBase::cast()
+  */
+template<> struct scalar_sum_op<bool,bool> : scalar_sum_op<int,int> {
+  EIGEN_DEPRECATED
+  scalar_sum_op() {}
+};
+
+
+/** \internal
+  * \brief Template functor to compute the product of two scalars
+  *
+  * \sa class CwiseBinaryOp, Cwise::operator*(), class VectorwiseOp, MatrixBase::redux()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_product_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_product_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_product_op)
+#else
+  scalar_product_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a * b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmul(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux_mul(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_product_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::MulCost + NumTraits<RhsScalar>::MulCost)/2, // rough estimate!
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasMul && packet_traits<RhsScalar>::HasMul
+    // TODO vectorize mixed product
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the conjugate product of two scalars
+  *
+  * This is a short cut for conj(x) * y which is needed for optimization purpose; in Eigen2 support mode, this becomes x * conj(y)
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_conj_product_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+
+  enum {
+    Conj = NumTraits<LhsScalar>::IsComplex
+  };
+  
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_conj_product_op>::ReturnType result_type;
+  
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_conj_product_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const
+  { return conj_helper<LhsScalar,RhsScalar,Conj,false>().pmul(a,b); }
+  
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return conj_helper<Packet,Packet,Conj,false>().pmul(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_conj_product_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = NumTraits<LhsScalar>::MulCost,
+    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMul
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the min of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::cwiseMin, class VectorwiseOp, MatrixBase::minCoeff()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_min_op : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_min_op>::ReturnType result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_min_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return numext::mini(a, b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmin(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux_min(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_min_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMin
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the max of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::cwiseMax, class VectorwiseOp, MatrixBase::maxCoeff()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_max_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_max_op>::ReturnType result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_max_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return numext::maxi(a, b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pmax(a,b); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const
+  { return internal::predux_max(a); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_max_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = internal::is_same<LhsScalar, RhsScalar>::value && packet_traits<LhsScalar>::HasMax
+  };
+};
+
+/** \internal
+  * \brief Template functors for comparison of two scalars
+  * \todo Implement packet-comparisons
+  */
+template<typename LhsScalar, typename RhsScalar, ComparisonName cmp> struct scalar_cmp_op;
+
+template<typename LhsScalar, typename RhsScalar, ComparisonName cmp>
+struct functor_traits<scalar_cmp_op<LhsScalar,RhsScalar, cmp> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = false
+  };
+};
+
+template<ComparisonName Cmp, typename LhsScalar, typename RhsScalar>
+struct result_of<scalar_cmp_op<LhsScalar, RhsScalar, Cmp>(LhsScalar,RhsScalar)> {
+  typedef bool type;
+};
+
+
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_EQ> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a==b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_LT> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a<b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_LE> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a<=b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_GT> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a>b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_GE> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a>=b;}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_UNORD> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return !(a<=b || b<=a);}
+};
+template<typename LhsScalar, typename RhsScalar>
+struct scalar_cmp_op<LhsScalar,RhsScalar, cmp_NEQ> : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef bool result_type;
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator()(const LhsScalar& a, const RhsScalar& b) const {return a!=b;}
+};
+
+
+/** \internal
+  * \brief Template functor to compute the hypot of two \b positive \b and \b real scalars
+  *
+  * \sa MatrixBase::stableNorm(), class Redux
+  */
+template<typename Scalar>
+struct scalar_hypot_op<Scalar,Scalar> : binary_op_base<Scalar,Scalar>
+{
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_hypot_op)
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar &x, const Scalar &y) const
+  {
+    // This functor is used by hypotNorm only for which it is faster to first apply abs
+    // on all coefficients prior to reduction through hypot.
+    // This way we avoid calling abs on positive and real entries, and this also permits
+    // to seamlessly handle complexes. Otherwise we would have to handle both real and complexes
+    // through the same functor...
+    return internal::positive_real_hypot(x,y);
+  }
+};
+template<typename Scalar>
+struct functor_traits<scalar_hypot_op<Scalar,Scalar> > {
+  enum
+  {
+    Cost = 3 * NumTraits<Scalar>::AddCost +
+           2 * NumTraits<Scalar>::MulCost +
+           2 * scalar_div_cost<Scalar,false>::value,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the pow of two scalars
+  */
+template<typename Scalar, typename Exponent>
+struct scalar_pow_op  : binary_op_base<Scalar,Exponent>
+{
+  typedef typename ScalarBinaryOpTraits<Scalar,Exponent,scalar_pow_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_pow_op)
+#else
+  scalar_pow_op() {
+    typedef Scalar LhsScalar;
+    typedef Exponent RhsScalar;
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC
+  inline result_type operator() (const Scalar& a, const Exponent& b) const { return numext::pow(a, b); }
+};
+template<typename Scalar, typename Exponent>
+struct functor_traits<scalar_pow_op<Scalar,Exponent> > {
+  enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = false };
+};
+
+
+
+//---------- non associative binary functors ----------
+
+/** \internal
+  * \brief Template functor to compute the difference of two scalars
+  *
+  * \sa class CwiseBinaryOp, MatrixBase::operator-
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_difference_op : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_difference_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_difference_op)
+#else
+  scalar_difference_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a - b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::psub(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_difference_op<LhsScalar,RhsScalar> > {
+  enum {
+    Cost = (NumTraits<LhsScalar>::AddCost+NumTraits<RhsScalar>::AddCost)/2,
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasSub && packet_traits<RhsScalar>::HasSub
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the quotient of two scalars
+  *
+  * \sa class CwiseBinaryOp, Cwise::operator/()
+  */
+template<typename LhsScalar,typename RhsScalar>
+struct scalar_quotient_op  : binary_op_base<LhsScalar,RhsScalar>
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar,scalar_quotient_op>::ReturnType result_type;
+#ifndef EIGEN_SCALAR_BINARY_OP_PLUGIN
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_op)
+#else
+  scalar_quotient_op() {
+    EIGEN_SCALAR_BINARY_OP_PLUGIN
+  }
+#endif
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a / b; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const
+  { return internal::pdiv(a,b); }
+};
+template<typename LhsScalar,typename RhsScalar>
+struct functor_traits<scalar_quotient_op<LhsScalar,RhsScalar> > {
+  typedef typename scalar_quotient_op<LhsScalar,RhsScalar>::result_type result_type;
+  enum {
+    PacketAccess = is_same<LhsScalar,RhsScalar>::value && packet_traits<LhsScalar>::HasDiv && packet_traits<RhsScalar>::HasDiv,
+    Cost = scalar_div_cost<result_type,PacketAccess>::value
+  };
+};
+
+
+
+/** \internal
+  * \brief Template functor to compute the and of two booleans
+  *
+  * \sa class CwiseBinaryOp, ArrayBase::operator&&
+  */
+struct scalar_boolean_and_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_and_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a && b; }
+};
+template<> struct functor_traits<scalar_boolean_and_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the or of two booleans
+  *
+  * \sa class CwiseBinaryOp, ArrayBase::operator||
+  */
+struct scalar_boolean_or_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_or_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a || b; }
+};
+template<> struct functor_traits<scalar_boolean_or_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+ * \brief Template functor to compute the xor of two booleans
+ *
+ * \sa class CwiseBinaryOp, ArrayBase::operator^
+ */
+struct scalar_boolean_xor_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_xor_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a ^ b; }
+};
+template<> struct functor_traits<scalar_boolean_xor_op> {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+
+
+//---------- binary functors bound to a constant, thus appearing as a unary functor ----------
+
+// The following two classes permits to turn any binary functor into a unary one with one argument bound to a constant value.
+// They are analogues to std::binder1st/binder2nd but with the following differences:
+//  - they are compatible with packetOp
+//  - they are portable across C++ versions (the std::binder* are deprecated in C++11)
+template<typename BinaryOp> struct bind1st_op : BinaryOp {
+
+  typedef typename BinaryOp::first_argument_type  first_argument_type;
+  typedef typename BinaryOp::second_argument_type second_argument_type;
+  typedef typename BinaryOp::result_type          result_type;
+
+  bind1st_op(const first_argument_type &val) : m_value(val) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const second_argument_type& b) const { return BinaryOp::operator()(m_value,b); }
+
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& b) const
+  { return BinaryOp::packetOp(internal::pset1<Packet>(m_value), b); }
+
+  first_argument_type m_value;
+};
+template<typename BinaryOp> struct functor_traits<bind1st_op<BinaryOp> > : functor_traits<BinaryOp> {};
+
+
+template<typename BinaryOp> struct bind2nd_op : BinaryOp {
+
+  typedef typename BinaryOp::first_argument_type  first_argument_type;
+  typedef typename BinaryOp::second_argument_type second_argument_type;
+  typedef typename BinaryOp::result_type          result_type;
+
+  bind2nd_op(const second_argument_type &val) : m_value(val) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const first_argument_type& a) const { return BinaryOp::operator()(a,m_value); }
+
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return BinaryOp::packetOp(a,internal::pset1<Packet>(m_value)); }
+
+  second_argument_type m_value;
+};
+template<typename BinaryOp> struct functor_traits<bind2nd_op<BinaryOp> > : functor_traits<BinaryOp> {};
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_BINARY_FUNCTORS_H
diff --git a/Eigen/src/Core/functors/NullaryFunctors.h b/Eigen/src/Core/functors/NullaryFunctors.h
new file mode 100644
index 0000000..b03be02
--- /dev/null
+++ b/Eigen/src/Core/functors/NullaryFunctors.h
@@ -0,0 +1,188 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_NULLARY_FUNCTORS_H
+#define EIGEN_NULLARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<typename Scalar>
+struct scalar_constant_op {
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE scalar_constant_op(const scalar_constant_op& other) : m_other(other.m_other) { }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE scalar_constant_op(const Scalar& other) : m_other(other) { }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() () const { return m_other; }
+  template<typename PacketType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const PacketType packetOp() const { return internal::pset1<PacketType>(m_other); }
+  const Scalar m_other;
+};
+template<typename Scalar>
+struct functor_traits<scalar_constant_op<Scalar> >
+{ enum { Cost = 0 /* as the constant value should be loaded in register only once for the whole expression */,
+         PacketAccess = packet_traits<Scalar>::Vectorizable, IsRepeatable = true }; };
+
+template<typename Scalar> struct scalar_identity_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_identity_op)
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (IndexType row, IndexType col) const { return row==col ? Scalar(1) : Scalar(0); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_identity_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::AddCost, PacketAccess = false, IsRepeatable = true }; };
+
+template <typename Scalar, typename Packet, bool IsInteger> struct linspaced_op_impl;
+
+template <typename Scalar, typename Packet>
+struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/false>
+{
+  linspaced_op_impl(const Scalar& low, const Scalar& high, Index num_steps) :
+    m_low(low), m_high(high), m_size1(num_steps==1 ? 1 : num_steps-1), m_step(num_steps==1 ? Scalar() : (high-low)/Scalar(num_steps-1)),
+    m_flip(numext::abs(high)<numext::abs(low))
+  {}
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (IndexType i) const {
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    if(m_flip)
+      return (i==0)? m_low : (m_high - RealScalar(m_size1-i)*m_step);
+    else
+      return (i==m_size1)? m_high : (m_low + RealScalar(i)*m_step);
+  }
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(IndexType i) const
+  {
+    // Principle:
+    // [low, ..., low] + ( [step, ..., step] * ( [i, ..., i] + [0, ..., size] ) )
+    if(m_flip)
+    {
+      Packet pi = plset<Packet>(Scalar(i-m_size1));
+      Packet res = padd(pset1<Packet>(m_high), pmul(pset1<Packet>(m_step), pi));
+      if(i==0)
+        res = pinsertfirst(res, m_low);
+      return res;
+    }
+    else
+    {
+      Packet pi = plset<Packet>(Scalar(i));
+      Packet res = padd(pset1<Packet>(m_low), pmul(pset1<Packet>(m_step), pi));
+      if(i==m_size1-unpacket_traits<Packet>::size+1)
+        res = pinsertlast(res, m_high);
+      return res;
+    }
+  }
+
+  const Scalar m_low;
+  const Scalar m_high;
+  const Index m_size1;
+  const Scalar m_step;
+  const bool m_flip;
+};
+
+template <typename Scalar, typename Packet>
+struct linspaced_op_impl<Scalar,Packet,/*IsInteger*/true>
+{
+  linspaced_op_impl(const Scalar& low, const Scalar& high, Index num_steps) :
+    m_low(low),
+    m_multiplier((high-low)/convert_index<Scalar>(num_steps<=1 ? 1 : num_steps-1)),
+    m_divisor(convert_index<Scalar>((high>=low?num_steps:-num_steps)+(high-low))/((numext::abs(high-low)+1)==0?1:(numext::abs(high-low)+1))),
+    m_use_divisor(num_steps>1 && (numext::abs(high-low)+1)<num_steps)
+  {}
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE
+  const Scalar operator() (IndexType i) const
+  {
+    if(m_use_divisor) return m_low + convert_index<Scalar>(i)/m_divisor;
+    else              return m_low + convert_index<Scalar>(i)*m_multiplier;
+  }
+
+  const Scalar m_low;
+  const Scalar m_multiplier;
+  const Scalar m_divisor;
+  const bool m_use_divisor;
+};
+
+// ----- Linspace functor ----------------------------------------------------------------
+
+// Forward declaration (we default to random access which does not really give
+// us a speed gain when using packet access but it allows to use the functor in
+// nested expressions).
+template <typename Scalar, typename PacketType> struct linspaced_op;
+template <typename Scalar, typename PacketType> struct functor_traits< linspaced_op<Scalar,PacketType> >
+{
+  enum
+  {
+    Cost = 1,
+    PacketAccess =   (!NumTraits<Scalar>::IsInteger) && packet_traits<Scalar>::HasSetLinear && packet_traits<Scalar>::HasBlend,
+                  /*&& ((!NumTraits<Scalar>::IsInteger) || packet_traits<Scalar>::HasDiv),*/ // <- vectorization for integer is currently disabled
+    IsRepeatable = true
+  };
+};
+template <typename Scalar, typename PacketType> struct linspaced_op
+{
+  linspaced_op(const Scalar& low, const Scalar& high, Index num_steps)
+    : impl((num_steps==1 ? high : low),high,num_steps)
+  {}
+
+  template<typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (IndexType i) const { return impl(i); }
+
+  template<typename Packet,typename IndexType>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(IndexType i) const { return impl.packetOp(i); }
+
+  // This proxy object handles the actual required temporaries and the different
+  // implementations (integer vs. floating point).
+  const linspaced_op_impl<Scalar,PacketType,NumTraits<Scalar>::IsInteger> impl;
+};
+
+// Linear access is automatically determined from the operator() prototypes available for the given functor.
+// If it exposes an operator()(i,j), then we assume the i and j coefficients are required independently
+// and linear access is not possible. In all other cases, linear access is enabled.
+// Users should not have to deal with this structure.
+template<typename Functor> struct functor_has_linear_access { enum { ret = !has_binary_operator<Functor>::value }; };
+
+// For unreliable compilers, let's specialize the has_*ary_operator
+// helpers so that at least built-in nullary functors work fine.
+#if !( (EIGEN_COMP_MSVC>1600) || (EIGEN_GNUC_AT_LEAST(4,8)) || (EIGEN_COMP_ICC>=1600))
+template<typename Scalar,typename IndexType>
+struct has_nullary_operator<scalar_constant_op<Scalar>,IndexType> { enum { value = 1}; };
+template<typename Scalar,typename IndexType>
+struct has_unary_operator<scalar_constant_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_binary_operator<scalar_constant_op<Scalar>,IndexType> { enum { value = 0}; };
+
+template<typename Scalar,typename IndexType>
+struct has_nullary_operator<scalar_identity_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_unary_operator<scalar_identity_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_binary_operator<scalar_identity_op<Scalar>,IndexType> { enum { value = 1}; };
+
+template<typename Scalar, typename PacketType,typename IndexType>
+struct has_nullary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 0}; };
+template<typename Scalar, typename PacketType,typename IndexType>
+struct has_unary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 1}; };
+template<typename Scalar, typename PacketType,typename IndexType>
+struct has_binary_operator<linspaced_op<Scalar,PacketType>,IndexType> { enum { value = 0}; };
+
+template<typename Scalar,typename IndexType>
+struct has_nullary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 1}; };
+template<typename Scalar,typename IndexType>
+struct has_unary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 0}; };
+template<typename Scalar,typename IndexType>
+struct has_binary_operator<scalar_random_op<Scalar>,IndexType> { enum { value = 0}; };
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_NULLARY_FUNCTORS_H
diff --git a/Eigen/src/Core/functors/StlFunctors.h b/Eigen/src/Core/functors/StlFunctors.h
new file mode 100644
index 0000000..9c1d758
--- /dev/null
+++ b/Eigen/src/Core/functors/StlFunctors.h
@@ -0,0 +1,136 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_STL_FUNCTORS_H
+#define EIGEN_STL_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+// default functor traits for STL functors:
+
+template<typename T>
+struct functor_traits<std::multiplies<T> >
+{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::divides<T> >
+{ enum { Cost = NumTraits<T>::MulCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::plus<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::minus<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::negate<T> >
+{ enum { Cost = NumTraits<T>::AddCost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_or<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_and<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::logical_not<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::greater<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::less<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::greater_equal<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::less_equal<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::equal_to<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::not_equal_to<T> >
+{ enum { Cost = 1, PacketAccess = false }; };
+
+#if (__cplusplus < 201103L) && (EIGEN_COMP_MSVC <= 1900)
+// std::binder* are deprecated since c++11 and will be removed in c++17
+template<typename T>
+struct functor_traits<std::binder2nd<T> >
+{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
+
+template<typename T>
+struct functor_traits<std::binder1st<T> >
+{ enum { Cost = functor_traits<T>::Cost, PacketAccess = false }; };
+#endif
+
+#if (__cplusplus < 201703L) && (EIGEN_COMP_MSVC < 1910)
+// std::unary_negate is deprecated since c++17 and will be removed in c++20
+template<typename T>
+struct functor_traits<std::unary_negate<T> >
+{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
+
+// std::binary_negate is deprecated since c++17 and will be removed in c++20
+template<typename T>
+struct functor_traits<std::binary_negate<T> >
+{ enum { Cost = 1 + functor_traits<T>::Cost, PacketAccess = false }; };
+#endif
+
+#ifdef EIGEN_STDEXT_SUPPORT
+
+template<typename T0,typename T1>
+struct functor_traits<std::project1st<T0,T1> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::project2nd<T0,T1> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::select2nd<std::pair<T0,T1> > >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::select1st<std::pair<T0,T1> > >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+template<typename T0,typename T1>
+struct functor_traits<std::unary_compose<T0,T1> >
+{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost, PacketAccess = false }; };
+
+template<typename T0,typename T1,typename T2>
+struct functor_traits<std::binary_compose<T0,T1,T2> >
+{ enum { Cost = functor_traits<T0>::Cost + functor_traits<T1>::Cost + functor_traits<T2>::Cost, PacketAccess = false }; };
+
+#endif // EIGEN_STDEXT_SUPPORT
+
+// allow to add new functors and specializations of functor_traits from outside Eigen.
+// this macro is really needed because functor_traits must be specialized after it is declared but before it is used...
+#ifdef EIGEN_FUNCTORS_PLUGIN
+#include EIGEN_FUNCTORS_PLUGIN
+#endif
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_STL_FUNCTORS_H
diff --git a/Eigen/src/Core/functors/TernaryFunctors.h b/Eigen/src/Core/functors/TernaryFunctors.h
new file mode 100644
index 0000000..b254e96
--- /dev/null
+++ b/Eigen/src/Core/functors/TernaryFunctors.h
@@ -0,0 +1,25 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Eugene Brevdo <ebrevdo@gmail.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_TERNARY_FUNCTORS_H
+#define EIGEN_TERNARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+//---------- associative ternary functors ----------
+
+
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_TERNARY_FUNCTORS_H
diff --git a/Eigen/src/Core/functors/UnaryFunctors.h b/Eigen/src/Core/functors/UnaryFunctors.h
new file mode 100644
index 0000000..2e6a00f
--- /dev/null
+++ b/Eigen/src/Core/functors/UnaryFunctors.h
@@ -0,0 +1,792 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_UNARY_FUNCTORS_H
+#define EIGEN_UNARY_FUNCTORS_H
+
+namespace Eigen {
+
+namespace internal {
+
+/** \internal
+  * \brief Template functor to compute the opposite of a scalar
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::operator-
+  */
+template<typename Scalar> struct scalar_opposite_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_opposite_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return -a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pnegate(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_opposite_op<Scalar> >
+{ enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasNegate };
+};
+
+/** \internal
+  * \brief Template functor to compute the absolute value of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::abs
+  */
+template<typename Scalar> struct scalar_abs_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return numext::abs(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pabs(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_abs_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasAbs
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the score of a scalar, to chose a pivot
+  *
+  * \sa class CwiseUnaryOp
+  */
+template<typename Scalar> struct scalar_score_coeff_op : scalar_abs_op<Scalar>
+{
+  typedef void Score_is_abs;
+};
+template<typename Scalar>
+struct functor_traits<scalar_score_coeff_op<Scalar> > : functor_traits<scalar_abs_op<Scalar> > {};
+
+/* Avoid recomputing abs when we know the score and they are the same. Not a true Eigen functor.  */
+template<typename Scalar, typename=void> struct abs_knowing_score
+{
+  EIGEN_EMPTY_STRUCT_CTOR(abs_knowing_score)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  template<typename Score>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a, const Score&) const { return numext::abs(a); }
+};
+template<typename Scalar> struct abs_knowing_score<Scalar, typename scalar_score_coeff_op<Scalar>::Score_is_abs>
+{
+  EIGEN_EMPTY_STRUCT_CTOR(abs_knowing_score)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  template<typename Scal>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scal&, const result_type& a) const { return a; }
+};
+
+/** \internal
+  * \brief Template functor to compute the squared absolute value of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::abs2
+  */
+template<typename Scalar> struct scalar_abs2_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_abs2_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return numext::abs2(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::pmul(a,a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_abs2_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasAbs2 }; };
+
+/** \internal
+  * \brief Template functor to compute the conjugate of a complex value
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::conjugate()
+  */
+template<typename Scalar> struct scalar_conjugate_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_conjugate_op)
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { using numext::conj; return conj(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const { return internal::pconj(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_conjugate_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::IsComplex ? NumTraits<Scalar>::AddCost : 0,
+    PacketAccess = packet_traits<Scalar>::HasConj
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the phase angle of a complex
+  *
+  * \sa class CwiseUnaryOp, Cwise::arg
+  */
+template<typename Scalar> struct scalar_arg_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_arg_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { using numext::arg; return arg(a); }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const
+  { return internal::parg(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_arg_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::IsComplex ? 5 * NumTraits<Scalar>::MulCost : NumTraits<Scalar>::AddCost,
+    PacketAccess = packet_traits<Scalar>::HasArg
+  };
+};
+/** \internal
+  * \brief Template functor to cast a scalar to another type
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::cast()
+  */
+template<typename Scalar, typename NewType>
+struct scalar_cast_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op)
+  typedef NewType result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const NewType operator() (const Scalar& a) const { return cast<Scalar, NewType>(a); }
+};
+template<typename Scalar, typename NewType>
+struct functor_traits<scalar_cast_op<Scalar,NewType> >
+{ enum { Cost = is_same<Scalar, NewType>::value ? 0 : NumTraits<NewType>::AddCost, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the real part of a complex
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::real()
+  */
+template<typename Scalar>
+struct scalar_real_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::real(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_real_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the imaginary part of a complex
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::imag()
+  */
+template<typename Scalar>
+struct scalar_imag_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::imag(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_imag_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the real part of a complex as a reference
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::real()
+  */
+template<typename Scalar>
+struct scalar_real_ref_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_real_ref_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::real_ref(*const_cast<Scalar*>(&a)); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_real_ref_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  * \brief Template functor to extract the imaginary part of a complex as a reference
+  *
+  * \sa class CwiseUnaryOp, MatrixBase::imag()
+  */
+template<typename Scalar>
+struct scalar_imag_ref_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_ref_op)
+  typedef typename NumTraits<Scalar>::Real result_type;
+  EIGEN_DEVICE_FUNC
+  EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::imag_ref(*const_cast<Scalar*>(&a)); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_imag_ref_op<Scalar> >
+{ enum { Cost = 0, PacketAccess = false }; };
+
+/** \internal
+  *
+  * \brief Template functor to compute the exponential of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::exp()
+  */
+template<typename Scalar> struct scalar_exp_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_exp_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::exp(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pexp(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_exp_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasExp,
+    // The following numbers are based on the AVX implementation.
+#ifdef EIGEN_VECTORIZE_FMA
+    // Haswell can issue 2 add/mul/madd per cycle.
+    Cost =
+    (sizeof(Scalar) == 4
+     // float: 8 pmadd, 4 pmul, 2 padd/psub, 6 other
+     ? (8 * NumTraits<Scalar>::AddCost + 6 * NumTraits<Scalar>::MulCost)
+     // double: 7 pmadd, 5 pmul, 3 padd/psub, 1 div,  13 other
+     : (14 * NumTraits<Scalar>::AddCost +
+        6 * NumTraits<Scalar>::MulCost +
+        scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value))
+#else
+    Cost =
+    (sizeof(Scalar) == 4
+     // float: 7 pmadd, 6 pmul, 4 padd/psub, 10 other
+     ? (21 * NumTraits<Scalar>::AddCost + 13 * NumTraits<Scalar>::MulCost)
+     // double: 7 pmadd, 5 pmul, 3 padd/psub, 1 div,  13 other
+     : (23 * NumTraits<Scalar>::AddCost +
+        12 * NumTraits<Scalar>::MulCost +
+        scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value))
+#endif
+  };
+};
+
+/** \internal
+  *
+  * \brief Template functor to compute the logarithm of a scalar
+  *
+  * \sa class CwiseUnaryOp, ArrayBase::log()
+  */
+template<typename Scalar> struct scalar_log_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_log_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::log(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::plog(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_log_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasLog,
+    Cost =
+    (PacketAccess
+     // The following numbers are based on the AVX implementation.
+#ifdef EIGEN_VECTORIZE_FMA
+     // 8 pmadd, 6 pmul, 8 padd/psub, 16 other, can issue 2 add/mul/madd per cycle.
+     ? (20 * NumTraits<Scalar>::AddCost + 7 * NumTraits<Scalar>::MulCost)
+#else
+     // 8 pmadd, 6 pmul, 8 padd/psub, 20 other
+     ? (36 * NumTraits<Scalar>::AddCost + 14 * NumTraits<Scalar>::MulCost)
+#endif
+     // Measured cost of std::log.
+     : sizeof(Scalar)==4 ? 40 : 85)
+  };
+};
+
+/** \internal
+  *
+  * \brief Template functor to compute the logarithm of 1 plus a scalar value
+  *
+  * \sa class CwiseUnaryOp, ArrayBase::log1p()
+  */
+template<typename Scalar> struct scalar_log1p_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_log1p_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::log1p(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::plog1p(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_log1p_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasLog1p,
+    Cost = functor_traits<scalar_log_op<Scalar> >::Cost // TODO measure cost of log1p
+  };
+};
+
+/** \internal
+  *
+  * \brief Template functor to compute the base-10 logarithm of a scalar
+  *
+  * \sa class CwiseUnaryOp, Cwise::log10()
+  */
+template<typename Scalar> struct scalar_log10_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_log10_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { EIGEN_USING_STD_MATH(log10) return log10(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::plog10(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_log10_op<Scalar> >
+{ enum { Cost = 5 * NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasLog10 }; };
+
+/** \internal
+  * \brief Template functor to compute the square root of a scalar
+  * \sa class CwiseUnaryOp, Cwise::sqrt()
+  */
+template<typename Scalar> struct scalar_sqrt_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sqrt_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::sqrt(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psqrt(a); }
+};
+template <typename Scalar>
+struct functor_traits<scalar_sqrt_op<Scalar> > {
+  enum {
+#if EIGEN_FAST_MATH
+    // The following numbers are based on the AVX implementation.
+    Cost = (sizeof(Scalar) == 8 ? 28
+                                // 4 pmul, 1 pmadd, 3 other
+                                : (3 * NumTraits<Scalar>::AddCost +
+                                   5 * NumTraits<Scalar>::MulCost)),
+#else
+    // The following numbers are based on min VSQRT throughput on Haswell.
+    Cost = (sizeof(Scalar) == 8 ? 28 : 14),
+#endif
+    PacketAccess = packet_traits<Scalar>::HasSqrt
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the reciprocal square root of a scalar
+  * \sa class CwiseUnaryOp, Cwise::rsqrt()
+  */
+template<typename Scalar> struct scalar_rsqrt_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_rsqrt_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return Scalar(1)/numext::sqrt(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::prsqrt(a); }
+};
+
+template<typename Scalar>
+struct functor_traits<scalar_rsqrt_op<Scalar> >
+{ enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasRsqrt
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the cosine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::cos()
+  */
+template<typename Scalar> struct scalar_cos_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cos_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return numext::cos(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pcos(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cos_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasCos
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the sine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::sin()
+  */
+template<typename Scalar> struct scalar_sin_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sin_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::sin(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psin(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sin_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasSin
+  };
+};
+
+
+/** \internal
+  * \brief Template functor to compute the tan of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::tan()
+  */
+template<typename Scalar> struct scalar_tan_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_tan_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::tan(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::ptan(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_tan_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasTan
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the arc cosine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::acos()
+  */
+template<typename Scalar> struct scalar_acos_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_acos_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::acos(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pacos(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_acos_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasACos
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the arc sine of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::asin()
+  */
+template<typename Scalar> struct scalar_asin_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_asin_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::asin(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pasin(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_asin_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasASin
+  };
+};
+
+
+/** \internal
+  * \brief Template functor to compute the atan of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::atan()
+  */
+template<typename Scalar> struct scalar_atan_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_atan_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::atan(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::patan(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_atan_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasATan
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the tanh of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::tanh()
+  */
+template <typename Scalar>
+struct scalar_tanh_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_tanh_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator()(const Scalar& a) const { return numext::tanh(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& x) const { return ptanh(x); }
+};
+
+template <typename Scalar>
+struct functor_traits<scalar_tanh_op<Scalar> > {
+  enum {
+    PacketAccess = packet_traits<Scalar>::HasTanh,
+    Cost = ( (EIGEN_FAST_MATH && is_same<Scalar,float>::value)
+// The following numbers are based on the AVX implementation,
+#ifdef EIGEN_VECTORIZE_FMA
+                // Haswell can issue 2 add/mul/madd per cycle.
+                // 9 pmadd, 2 pmul, 1 div, 2 other
+                ? (2 * NumTraits<Scalar>::AddCost +
+                   6 * NumTraits<Scalar>::MulCost +
+                   scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value)
+#else
+                ? (11 * NumTraits<Scalar>::AddCost +
+                   11 * NumTraits<Scalar>::MulCost +
+                   scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value)
+#endif
+                // This number assumes a naive implementation of tanh
+                : (6 * NumTraits<Scalar>::AddCost +
+                   3 * NumTraits<Scalar>::MulCost +
+                   2 * scalar_div_cost<Scalar,packet_traits<Scalar>::HasDiv>::value +
+                   functor_traits<scalar_exp_op<Scalar> >::Cost))
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the sinh of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::sinh()
+  */
+template<typename Scalar> struct scalar_sinh_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sinh_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::sinh(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psinh(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sinh_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasSinh
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the cosh of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::cosh()
+  */
+template<typename Scalar> struct scalar_cosh_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cosh_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const { return numext::cosh(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pcosh(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cosh_op<Scalar> >
+{
+  enum {
+    Cost = 5 * NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasCosh
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the inverse of a scalar
+  * \sa class CwiseUnaryOp, Cwise::inverse()
+  */
+template<typename Scalar>
+struct scalar_inverse_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_inverse_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return Scalar(1)/a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC inline const Packet packetOp(const Packet& a) const
+  { return internal::pdiv(pset1<Packet>(Scalar(1)),a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_inverse_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasDiv }; };
+
+/** \internal
+  * \brief Template functor to compute the square of a scalar
+  * \sa class CwiseUnaryOp, Cwise::square()
+  */
+template<typename Scalar>
+struct scalar_square_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_square_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return a*a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC inline const Packet packetOp(const Packet& a) const
+  { return internal::pmul(a,a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_square_op<Scalar> >
+{ enum { Cost = NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
+
+/** \internal
+  * \brief Template functor to compute the cube of a scalar
+  * \sa class CwiseUnaryOp, Cwise::cube()
+  */
+template<typename Scalar>
+struct scalar_cube_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_cube_op)
+  EIGEN_DEVICE_FUNC inline Scalar operator() (const Scalar& a) const { return a*a*a; }
+  template<typename Packet>
+  EIGEN_DEVICE_FUNC inline const Packet packetOp(const Packet& a) const
+  { return internal::pmul(a,pmul(a,a)); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_cube_op<Scalar> >
+{ enum { Cost = 2*NumTraits<Scalar>::MulCost, PacketAccess = packet_traits<Scalar>::HasMul }; };
+
+/** \internal
+  * \brief Template functor to compute the rounded value of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::round()
+  */
+template<typename Scalar> struct scalar_round_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_round_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return numext::round(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pround(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_round_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasRound
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the floor of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::floor()
+  */
+template<typename Scalar> struct scalar_floor_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_floor_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return numext::floor(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pfloor(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_floor_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasFloor
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the ceil of a scalar
+  * \sa class CwiseUnaryOp, ArrayBase::ceil()
+  */
+template<typename Scalar> struct scalar_ceil_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_ceil_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return numext::ceil(a); }
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::pceil(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_ceil_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = packet_traits<Scalar>::HasCeil
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute whether a scalar is NaN
+  * \sa class CwiseUnaryOp, ArrayBase::isnan()
+  */
+template<typename Scalar> struct scalar_isnan_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_isnan_op)
+  typedef bool result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return (numext::isnan)(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_isnan_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to check whether a scalar is +/-inf
+  * \sa class CwiseUnaryOp, ArrayBase::isinf()
+  */
+template<typename Scalar> struct scalar_isinf_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_isinf_op)
+  typedef bool result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return (numext::isinf)(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_isinf_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to check whether a scalar has a finite value
+  * \sa class CwiseUnaryOp, ArrayBase::isfinite()
+  */
+template<typename Scalar> struct scalar_isfinite_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_isfinite_op)
+  typedef bool result_type;
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return (numext::isfinite)(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_isfinite_op<Scalar> >
+{
+  enum {
+    Cost = NumTraits<Scalar>::MulCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the logical not of a boolean
+  *
+  * \sa class CwiseUnaryOp, ArrayBase::operator!
+  */
+template<typename Scalar> struct scalar_boolean_not_op {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_not_op)
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool operator() (const bool& a) const { return !a; }
+};
+template<typename Scalar>
+struct functor_traits<scalar_boolean_not_op<Scalar> > {
+  enum {
+    Cost = NumTraits<bool>::AddCost,
+    PacketAccess = false
+  };
+};
+
+/** \internal
+  * \brief Template functor to compute the signum of a scalar
+  * \sa class CwiseUnaryOp, Cwise::sign()
+  */
+template<typename Scalar,bool iscpx=(NumTraits<Scalar>::IsComplex!=0) > struct scalar_sign_op;
+template<typename Scalar>
+struct scalar_sign_op<Scalar,false> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sign_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const
+  {
+      return Scalar( (a>Scalar(0)) - (a<Scalar(0)) );
+  }
+  //TODO
+  //template <typename Packet>
+  //EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psign(a); }
+};
+template<typename Scalar>
+struct scalar_sign_op<Scalar,true> {
+  EIGEN_EMPTY_STRUCT_CTOR(scalar_sign_op)
+  EIGEN_DEVICE_FUNC inline const Scalar operator() (const Scalar& a) const
+  {
+    typedef typename NumTraits<Scalar>::Real real_type;
+    real_type aa = numext::abs(a);
+    if (aa==real_type(0))
+      return Scalar(0);
+    aa = real_type(1)/aa;
+    return Scalar(real(a)*aa, imag(a)*aa );
+  }
+  //TODO
+  //template <typename Packet>
+  //EIGEN_DEVICE_FUNC inline Packet packetOp(const Packet& a) const { return internal::psign(a); }
+};
+template<typename Scalar>
+struct functor_traits<scalar_sign_op<Scalar> >
+{ enum {
+    Cost = 
+        NumTraits<Scalar>::IsComplex
+        ? ( 8*NumTraits<Scalar>::MulCost  ) // roughly
+        : ( 3*NumTraits<Scalar>::AddCost),
+    PacketAccess = packet_traits<Scalar>::HasSign
+  };
+};
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_FUNCTORS_H
diff --git a/Eigen/src/Core/products/CMakeLists.txt b/Eigen/src/Core/products/CMakeLists.txt
deleted file mode 100644
index 21fc94a..0000000
--- a/Eigen/src/Core/products/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Core_Product_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Core_Product_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core/products COMPONENT Devel
-  )
diff --git a/Eigen/src/Core/products/CoeffBasedProduct.h b/Eigen/src/Core/products/CoeffBasedProduct.h
deleted file mode 100644
index 2a9d65b..0000000
--- a/Eigen/src/Core/products/CoeffBasedProduct.h
+++ /dev/null
@@ -1,476 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_COEFFBASED_PRODUCT_H
-#define EIGEN_COEFFBASED_PRODUCT_H
-
-namespace Eigen { 
-
-namespace internal {
-
-/*********************************************************************************
-*  Coefficient based product implementation.
-*  It is designed for the following use cases:
-*  - small fixed sizes
-*  - lazy products
-*********************************************************************************/
-
-/* Since the all the dimensions of the product are small, here we can rely
- * on the generic Assign mechanism to evaluate the product per coeff (or packet).
- *
- * Note that here the inner-loops should always be unrolled.
- */
-
-template<int Traversal, int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl;
-
-template<int StorageOrder, int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl;
-
-template<typename LhsNested, typename RhsNested, int NestingFlags>
-struct traits<CoeffBasedProduct<LhsNested,RhsNested,NestingFlags> >
-{
-  typedef MatrixXpr XprKind;
-  typedef typename remove_all<LhsNested>::type _LhsNested;
-  typedef typename remove_all<RhsNested>::type _RhsNested;
-  typedef typename scalar_product_traits<typename _LhsNested::Scalar, typename _RhsNested::Scalar>::ReturnType Scalar;
-  typedef typename promote_storage_type<typename traits<_LhsNested>::StorageKind,
-                                           typename traits<_RhsNested>::StorageKind>::ret StorageKind;
-  typedef typename promote_index_type<typename traits<_LhsNested>::Index,
-                                         typename traits<_RhsNested>::Index>::type Index;
-
-  enum {
-      LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-      RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-      LhsFlags = _LhsNested::Flags,
-      RhsFlags = _RhsNested::Flags,
-
-      RowsAtCompileTime = _LhsNested::RowsAtCompileTime,
-      ColsAtCompileTime = _RhsNested::ColsAtCompileTime,
-      InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(_LhsNested::ColsAtCompileTime, _RhsNested::RowsAtCompileTime),
-
-      MaxRowsAtCompileTime = _LhsNested::MaxRowsAtCompileTime,
-      MaxColsAtCompileTime = _RhsNested::MaxColsAtCompileTime,
-
-      LhsRowMajor = LhsFlags & RowMajorBit,
-      RhsRowMajor = RhsFlags & RowMajorBit,
-
-      SameType = is_same<typename _LhsNested::Scalar,typename _RhsNested::Scalar>::value,
-
-      CanVectorizeRhs = RhsRowMajor && (RhsFlags & PacketAccessBit)
-                      && (ColsAtCompileTime == Dynamic
-                          || ( (ColsAtCompileTime % packet_traits<Scalar>::size) == 0
-                              && (RhsFlags&AlignedBit)
-                             )
-                         ),
-
-      CanVectorizeLhs = (!LhsRowMajor) && (LhsFlags & PacketAccessBit)
-                      && (RowsAtCompileTime == Dynamic
-                          || ( (RowsAtCompileTime % packet_traits<Scalar>::size) == 0
-                              && (LhsFlags&AlignedBit)
-                             )
-                         ),
-
-      EvalToRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1
-                     : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0
-                     : (RhsRowMajor && !CanVectorizeLhs),
-
-      Flags = ((unsigned int)(LhsFlags | RhsFlags) & HereditaryBits & ~RowMajorBit)
-            | (EvalToRowMajor ? RowMajorBit : 0)
-            | NestingFlags
-            | (LhsFlags & RhsFlags & AlignedBit)
-            // TODO enable vectorization for mixed types
-            | (SameType && (CanVectorizeLhs || CanVectorizeRhs) ? PacketAccessBit : 0),
-
-      CoeffReadCost = InnerSize == Dynamic ? Dynamic
-                    : InnerSize == 0 ? 0
-                    : InnerSize * (NumTraits<Scalar>::MulCost + LhsCoeffReadCost + RhsCoeffReadCost)
-                      + (InnerSize - 1) * NumTraits<Scalar>::AddCost,
-
-      /* CanVectorizeInner deserves special explanation. It does not affect the product flags. It is not used outside
-      * of Product. If the Product itself is not a packet-access expression, there is still a chance that the inner
-      * loop of the product might be vectorized. This is the meaning of CanVectorizeInner. Since it doesn't affect
-      * the Flags, it is safe to make this value depend on ActualPacketAccessBit, that doesn't affect the ABI.
-      */
-      CanVectorizeInner =    SameType
-                          && LhsRowMajor
-                          && (!RhsRowMajor)
-                          && (LhsFlags & RhsFlags & ActualPacketAccessBit)
-                          && (LhsFlags & RhsFlags & AlignedBit)
-                          && (InnerSize % packet_traits<Scalar>::size == 0)
-    };
-};
-
-} // end namespace internal
-
-template<typename LhsNested, typename RhsNested, int NestingFlags>
-class CoeffBasedProduct
-  : internal::no_assignment_operator,
-    public MatrixBase<CoeffBasedProduct<LhsNested, RhsNested, NestingFlags> >
-{
-  public:
-
-    typedef MatrixBase<CoeffBasedProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(CoeffBasedProduct)
-    typedef typename Base::PlainObject PlainObject;
-
-  private:
-
-    typedef typename internal::traits<CoeffBasedProduct>::_LhsNested _LhsNested;
-    typedef typename internal::traits<CoeffBasedProduct>::_RhsNested _RhsNested;
-
-    enum {
-      PacketSize = internal::packet_traits<Scalar>::size,
-      InnerSize  = internal::traits<CoeffBasedProduct>::InnerSize,
-      Unroll = CoeffReadCost != Dynamic && CoeffReadCost <= EIGEN_UNROLLING_LIMIT,
-      CanVectorizeInner = internal::traits<CoeffBasedProduct>::CanVectorizeInner
-    };
-
-    typedef internal::product_coeff_impl<CanVectorizeInner ? InnerVectorizedTraversal : DefaultTraversal,
-                                   Unroll ? InnerSize : Dynamic,
-                                   _LhsNested, _RhsNested, Scalar> ScalarCoeffImpl;
-
-    typedef CoeffBasedProduct<LhsNested,RhsNested,NestByRefBit> LazyCoeffBasedProductType;
-
-  public:
-
-    inline CoeffBasedProduct(const CoeffBasedProduct& other)
-      : Base(), m_lhs(other.m_lhs), m_rhs(other.m_rhs)
-    {}
-
-    template<typename Lhs, typename Rhs>
-    inline CoeffBasedProduct(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      // we don't allow taking products of matrices of different real types, as that wouldn't be vectorizable.
-      // We still allow to mix T and complex<T>.
-      EIGEN_STATIC_ASSERT((internal::scalar_product_traits<typename Lhs::RealScalar, typename Rhs::RealScalar>::Defined),
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-      eigen_assert(lhs.cols() == rhs.rows()
-        && "invalid matrix product"
-        && "if you wanted a coeff-wise or a dot product use the respective explicit functions");
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
-
-    EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const
-    {
-      Scalar res;
-      ScalarCoeffImpl::run(row, col, m_lhs, m_rhs, res);
-      return res;
-    }
-
-    /* Allow index-based non-packet access. It is impossible though to allow index-based packed access,
-     * which is why we don't set the LinearAccessBit.
-     */
-    EIGEN_STRONG_INLINE const Scalar coeff(Index index) const
-    {
-      Scalar res;
-      const Index row = RowsAtCompileTime == 1 ? 0 : index;
-      const Index col = RowsAtCompileTime == 1 ? index : 0;
-      ScalarCoeffImpl::run(row, col, m_lhs, m_rhs, res);
-      return res;
-    }
-
-    template<int LoadMode>
-    EIGEN_STRONG_INLINE const PacketScalar packet(Index row, Index col) const
-    {
-      PacketScalar res;
-      internal::product_packet_impl<Flags&RowMajorBit ? RowMajor : ColMajor,
-                              Unroll ? InnerSize : Dynamic,
-                              _LhsNested, _RhsNested, PacketScalar, LoadMode>
-        ::run(row, col, m_lhs, m_rhs, res);
-      return res;
-    }
-
-    // Implicit conversion to the nested type (trigger the evaluation of the product)
-    EIGEN_STRONG_INLINE operator const PlainObject& () const
-    {
-      m_result.lazyAssign(*this);
-      return m_result;
-    }
-
-    const _LhsNested& lhs() const { return m_lhs; }
-    const _RhsNested& rhs() const { return m_rhs; }
-
-    const Diagonal<const LazyCoeffBasedProductType,0> diagonal() const
-    { return reinterpret_cast<const LazyCoeffBasedProductType&>(*this); }
-
-    template<int DiagonalIndex>
-    const Diagonal<const LazyCoeffBasedProductType,DiagonalIndex> diagonal() const
-    { return reinterpret_cast<const LazyCoeffBasedProductType&>(*this); }
-
-    const Diagonal<const LazyCoeffBasedProductType,Dynamic> diagonal(Index index) const
-    { return reinterpret_cast<const LazyCoeffBasedProductType&>(*this).diagonal(index); }
-
-  protected:
-    typename internal::add_const_on_value_type<LhsNested>::type m_lhs;
-    typename internal::add_const_on_value_type<RhsNested>::type m_rhs;
-
-    mutable PlainObject m_result;
-};
-
-namespace internal {
-
-// here we need to overload the nested rule for products
-// such that the nested type is a const reference to a plain matrix
-template<typename Lhs, typename Rhs, int N, typename PlainObject>
-struct nested<CoeffBasedProduct<Lhs,Rhs,EvalBeforeNestingBit|EvalBeforeAssigningBit>, N, PlainObject>
-{
-  typedef PlainObject const& type;
-};
-
-/***************************************************************************
-* Normal product .coeff() implementation (with meta-unrolling)
-***************************************************************************/
-
-/**************************************
-*** Scalar path  - no vectorization ***
-**************************************/
-
-template<int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl<DefaultTraversal, UnrollingIndex, Lhs, Rhs, RetScalar>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar &res)
-  {
-    product_coeff_impl<DefaultTraversal, UnrollingIndex-1, Lhs, Rhs, RetScalar>::run(row, col, lhs, rhs, res);
-    res += lhs.coeff(row, UnrollingIndex-1) * rhs.coeff(UnrollingIndex-1, col);
-  }
-};
-
-template<typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl<DefaultTraversal, 1, Lhs, Rhs, RetScalar>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar &res)
-  {
-    res = lhs.coeff(row, 0) * rhs.coeff(0, col);
-  }
-};
-
-template<typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl<DefaultTraversal, 0, Lhs, Rhs, RetScalar>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, RetScalar &res)
-  {
-    res = RetScalar(0);
-  }
-};
-
-template<typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl<DefaultTraversal, Dynamic, Lhs, Rhs, RetScalar>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar& res)
-  {
-    res = (lhs.row(row).transpose().cwiseProduct( rhs.col(col) )).sum();
-  }
-};
-
-/*******************************************
-*** Scalar path with inner vectorization ***
-*******************************************/
-
-template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet>
-struct product_coeff_vectorized_unroller
-{
-  typedef typename Lhs::Index Index;
-  enum { PacketSize = packet_traits<typename Lhs::Scalar>::size };
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::PacketScalar &pres)
-  {
-    product_coeff_vectorized_unroller<UnrollingIndex-PacketSize, Lhs, Rhs, Packet>::run(row, col, lhs, rhs, pres);
-    pres = padd(pres, pmul( lhs.template packet<Aligned>(row, UnrollingIndex) , rhs.template packet<Aligned>(UnrollingIndex, col) ));
-  }
-};
-
-template<typename Lhs, typename Rhs, typename Packet>
-struct product_coeff_vectorized_unroller<0, Lhs, Rhs, Packet>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::PacketScalar &pres)
-  {
-    pres = pmul(lhs.template packet<Aligned>(row, 0) , rhs.template packet<Aligned>(0, col));
-  }
-};
-
-template<typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl<InnerVectorizedTraversal, 0, Lhs, Rhs, RetScalar>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, RetScalar &res)
-  {
-    res = 0;
-  }
-};
-
-template<int UnrollingIndex, typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl<InnerVectorizedTraversal, UnrollingIndex, Lhs, Rhs, RetScalar>
-{
-  typedef typename Lhs::PacketScalar Packet;
-  typedef typename Lhs::Index Index;
-  enum { PacketSize = packet_traits<typename Lhs::Scalar>::size };
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, RetScalar &res)
-  {
-    Packet pres;
-    product_coeff_vectorized_unroller<UnrollingIndex-PacketSize, Lhs, Rhs, Packet>::run(row, col, lhs, rhs, pres);
-    res = predux(pres);
-  }
-};
-
-template<typename Lhs, typename Rhs, int LhsRows = Lhs::RowsAtCompileTime, int RhsCols = Rhs::ColsAtCompileTime>
-struct product_coeff_vectorized_dyn_selector
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
-  {
-    res = lhs.row(row).transpose().cwiseProduct(rhs.col(col)).sum();
-  }
-};
-
-// NOTE the 3 following specializations are because taking .col(0) on a vector is a bit slower
-// NOTE maybe they are now useless since we have a specialization for Block<Matrix>
-template<typename Lhs, typename Rhs, int RhsCols>
-struct product_coeff_vectorized_dyn_selector<Lhs,Rhs,1,RhsCols>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
-  {
-    res = lhs.transpose().cwiseProduct(rhs.col(col)).sum();
-  }
-};
-
-template<typename Lhs, typename Rhs, int LhsRows>
-struct product_coeff_vectorized_dyn_selector<Lhs,Rhs,LhsRows,1>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index /*col*/, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
-  {
-    res = lhs.row(row).transpose().cwiseProduct(rhs).sum();
-  }
-};
-
-template<typename Lhs, typename Rhs>
-struct product_coeff_vectorized_dyn_selector<Lhs,Rhs,1,1>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
-  {
-    res = lhs.transpose().cwiseProduct(rhs).sum();
-  }
-};
-
-template<typename Lhs, typename Rhs, typename RetScalar>
-struct product_coeff_impl<InnerVectorizedTraversal, Dynamic, Lhs, Rhs, RetScalar>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, typename Lhs::Scalar &res)
-  {
-    product_coeff_vectorized_dyn_selector<Lhs,Rhs>::run(row, col, lhs, rhs, res);
-  }
-};
-
-/*******************
-*** Packet path  ***
-*******************/
-
-template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<RowMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
-  {
-    product_packet_impl<RowMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, res);
-    res =  pmadd(pset1<Packet>(lhs.coeff(row, UnrollingIndex-1)), rhs.template packet<LoadMode>(UnrollingIndex-1, col), res);
-  }
-};
-
-template<int UnrollingIndex, typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<ColMajor, UnrollingIndex, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
-  {
-    product_packet_impl<ColMajor, UnrollingIndex-1, Lhs, Rhs, Packet, LoadMode>::run(row, col, lhs, rhs, res);
-    res =  pmadd(lhs.template packet<LoadMode>(row, UnrollingIndex-1), pset1<Packet>(rhs.coeff(UnrollingIndex-1, col)), res);
-  }
-};
-
-template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<RowMajor, 1, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
-  {
-    res = pmul(pset1<Packet>(lhs.coeff(row, 0)),rhs.template packet<LoadMode>(0, col));
-  }
-};
-
-template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<ColMajor, 1, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet &res)
-  {
-    res = pmul(lhs.template packet<LoadMode>(row, 0), pset1<Packet>(rhs.coeff(0, col)));
-  }
-};
-
-template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<RowMajor, 0, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Packet &res)
-  {
-    res = pset1<Packet>(0);
-  }
-};
-
-template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<ColMajor, 0, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Packet &res)
-  {
-    res = pset1<Packet>(0);
-  }
-};
-
-template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<RowMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet& res)
-  {
-    res = pset1<Packet>(0);
-    for(Index i = 0; i < lhs.cols(); ++i)
-      res =  pmadd(pset1<Packet>(lhs.coeff(row, i)), rhs.template packet<LoadMode>(i, col), res);
-  }
-};
-
-template<typename Lhs, typename Rhs, typename Packet, int LoadMode>
-struct product_packet_impl<ColMajor, Dynamic, Lhs, Rhs, Packet, LoadMode>
-{
-  typedef typename Lhs::Index Index;
-  static EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Packet& res)
-  {
-    res = pset1<Packet>(0);
-    for(Index i = 0; i < lhs.cols(); ++i)
-      res =  pmadd(lhs.template packet<LoadMode>(row, i), pset1<Packet>(rhs.coeff(i, col)), res);
-  }
-};
-
-} // end namespace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_COEFFBASED_PRODUCT_H
diff --git a/Eigen/src/Core/products/GeneralBlockPanelKernel.h b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
index bcdca5b..e3980f6 100644
--- a/Eigen/src/Core/products/GeneralBlockPanelKernel.h
+++ b/Eigen/src/Core/products/GeneralBlockPanelKernel.h
@@ -10,8 +10,9 @@
 #ifndef EIGEN_GENERAL_BLOCK_PANEL_H
 #define EIGEN_GENERAL_BLOCK_PANEL_H
 
-namespace Eigen { 
-  
+
+namespace Eigen {
+
 namespace internal {
 
 template<typename _LhsScalar, typename _RhsScalar, bool _ConjLhs=false, bool _ConjRhs=false>
@@ -24,29 +25,51 @@
   return a<=0 ? b : a;
 }
 
+#if EIGEN_ARCH_i386_OR_x86_64
+const std::ptrdiff_t defaultL1CacheSize = 32*1024;
+const std::ptrdiff_t defaultL2CacheSize = 256*1024;
+const std::ptrdiff_t defaultL3CacheSize = 2*1024*1024;
+#else
+const std::ptrdiff_t defaultL1CacheSize = 16*1024;
+const std::ptrdiff_t defaultL2CacheSize = 512*1024;
+const std::ptrdiff_t defaultL3CacheSize = 512*1024;
+#endif
+
 /** \internal */
-inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1=0, std::ptrdiff_t* l2=0)
-{
-  static std::ptrdiff_t m_l1CacheSize = 0;
-  static std::ptrdiff_t m_l2CacheSize = 0;
-  if(m_l2CacheSize==0)
-  {
-    m_l1CacheSize = manage_caching_sizes_helper(queryL1CacheSize(),8 * 1024);
-    m_l2CacheSize = manage_caching_sizes_helper(queryTopLevelCacheSize(),1*1024*1024);
+struct CacheSizes {
+  CacheSizes(): m_l1(-1),m_l2(-1),m_l3(-1) {
+    int l1CacheSize, l2CacheSize, l3CacheSize;
+    queryCacheSizes(l1CacheSize, l2CacheSize, l3CacheSize);
+    m_l1 = manage_caching_sizes_helper(l1CacheSize, defaultL1CacheSize);
+    m_l2 = manage_caching_sizes_helper(l2CacheSize, defaultL2CacheSize);
+    m_l3 = manage_caching_sizes_helper(l3CacheSize, defaultL3CacheSize);
   }
-  
+
+  std::ptrdiff_t m_l1;
+  std::ptrdiff_t m_l2;
+  std::ptrdiff_t m_l3;
+};
+
+
+/** \internal */
+inline void manage_caching_sizes(Action action, std::ptrdiff_t* l1, std::ptrdiff_t* l2, std::ptrdiff_t* l3)
+{
+  static CacheSizes m_cacheSizes;
+
   if(action==SetAction)
   {
     // set the cpu cache size and cache all block sizes from a global cache size in byte
     eigen_internal_assert(l1!=0 && l2!=0);
-    m_l1CacheSize = *l1;
-    m_l2CacheSize = *l2;
+    m_cacheSizes.m_l1 = *l1;
+    m_cacheSizes.m_l2 = *l2;
+    m_cacheSizes.m_l3 = *l3;
   }
   else if(action==GetAction)
   {
     eigen_internal_assert(l1!=0 && l2!=0);
-    *l1 = m_l1CacheSize;
-    *l2 = m_l2CacheSize;
+    *l1 = m_cacheSizes.m_l1;
+    *l2 = m_cacheSizes.m_l2;
+    *l3 = m_cacheSizes.m_l3;
   }
   else
   {
@@ -54,6 +77,206 @@
   }
 }
 
+/* Helper for computeProductBlockingSizes.
+ *
+ * Given a m x k times k x n matrix product of scalar types \c LhsScalar and \c RhsScalar,
+ * this function computes the blocking size parameters along the respective dimensions
+ * for matrix products and related algorithms. The blocking sizes depends on various
+ * parameters:
+ * - the L1 and L2 cache sizes,
+ * - the register level blocking sizes defined by gebp_traits,
+ * - the number of scalars that fit into a packet (when vectorization is enabled).
+ *
+ * \sa setCpuCacheSizes */
+
+template<typename LhsScalar, typename RhsScalar, int KcFactor, typename Index>
+void evaluateProductBlockingSizesHeuristic(Index& k, Index& m, Index& n, Index num_threads = 1)
+{
+  typedef gebp_traits<LhsScalar,RhsScalar> Traits;
+
+  // Explanations:
+  // Let's recall that the product algorithms form mc x kc vertical panels A' on the lhs and
+  // kc x nc blocks B' on the rhs. B' has to fit into L2/L3 cache. Moreover, A' is processed
+  // per mr x kc horizontal small panels where mr is the blocking size along the m dimension
+  // at the register level. This small horizontal panel has to stay within L1 cache.
+  std::ptrdiff_t l1, l2, l3;
+  manage_caching_sizes(GetAction, &l1, &l2, &l3);
+
+  if (num_threads > 1) {
+    typedef typename Traits::ResScalar ResScalar;
+    enum {
+      kdiv = KcFactor * (Traits::mr * sizeof(LhsScalar) + Traits::nr * sizeof(RhsScalar)),
+      ksub = Traits::mr * Traits::nr * sizeof(ResScalar),
+      kr = 8,
+      mr = Traits::mr,
+      nr = Traits::nr
+    };
+    // Increasing k gives us more time to prefetch the content of the "C"
+    // registers. However once the latency is hidden there is no point in
+    // increasing the value of k, so we'll cap it at 320 (value determined
+    // experimentally).
+    const Index k_cache = (numext::mini<Index>)((l1-ksub)/kdiv, 320);
+    if (k_cache < k) {
+      k = k_cache - (k_cache % kr);
+      eigen_internal_assert(k > 0);
+    }
+
+    const Index n_cache = (l2-l1) / (nr * sizeof(RhsScalar) * k);
+    const Index n_per_thread = numext::div_ceil(n, num_threads);
+    if (n_cache <= n_per_thread) {
+      // Don't exceed the capacity of the l2 cache.
+      eigen_internal_assert(n_cache >= static_cast<Index>(nr));
+      n = n_cache - (n_cache % nr);
+      eigen_internal_assert(n > 0);
+    } else {
+      n = (numext::mini<Index>)(n, (n_per_thread + nr - 1) - ((n_per_thread + nr - 1) % nr));
+    }
+
+    if (l3 > l2) {
+      // l3 is shared between all cores, so we'll give each thread its own chunk of l3.
+      const Index m_cache = (l3-l2) / (sizeof(LhsScalar) * k * num_threads);
+      const Index m_per_thread = numext::div_ceil(m, num_threads);
+      if(m_cache < m_per_thread && m_cache >= static_cast<Index>(mr)) {
+        m = m_cache - (m_cache % mr);
+        eigen_internal_assert(m > 0);
+      } else {
+        m = (numext::mini<Index>)(m, (m_per_thread + mr - 1) - ((m_per_thread + mr - 1) % mr));
+      }
+    }
+  }
+  else {
+    // In unit tests we do not want to use extra large matrices,
+    // so we reduce the cache size to check the blocking strategy is not flawed
+#ifdef EIGEN_DEBUG_SMALL_PRODUCT_BLOCKS
+    l1 = 9*1024;
+    l2 = 32*1024;
+    l3 = 512*1024;
+#endif
+
+    // Early return for small problems because the computation below are time consuming for small problems.
+    // Perhaps it would make more sense to consider k*n*m??
+    // Note that for very tiny problem, this function should be bypassed anyway
+    // because we use the coefficient-based implementation for them.
+    if((numext::maxi)(k,(numext::maxi)(m,n))<48)
+      return;
+
+    typedef typename Traits::ResScalar ResScalar;
+    enum {
+      k_peeling = 8,
+      k_div = KcFactor * (Traits::mr * sizeof(LhsScalar) + Traits::nr * sizeof(RhsScalar)),
+      k_sub = Traits::mr * Traits::nr * sizeof(ResScalar)
+    };
+
+    // ---- 1st level of blocking on L1, yields kc ----
+
+    // Blocking on the third dimension (i.e., k) is chosen so that an horizontal panel
+    // of size mr x kc of the lhs plus a vertical panel of kc x nr of the rhs both fits within L1 cache.
+    // We also include a register-level block of the result (mx x nr).
+    // (In an ideal world only the lhs panel would stay in L1)
+    // Moreover, kc has to be a multiple of 8 to be compatible with loop peeling, leading to a maximum blocking size of:
+    const Index max_kc = numext::maxi<Index>(((l1-k_sub)/k_div) & (~(k_peeling-1)),1);
+    const Index old_k = k;
+    if(k>max_kc)
+    {
+      // We are really blocking on the third dimension:
+      // -> reduce blocking size to make sure the last block is as large as possible
+      //    while keeping the same number of sweeps over the result.
+      k = (k%max_kc)==0 ? max_kc
+                        : max_kc - k_peeling * ((max_kc-1-(k%max_kc))/(k_peeling*(k/max_kc+1)));
+
+      eigen_internal_assert(((old_k/k) == (old_k/max_kc)) && "the number of sweeps has to remain the same");
+    }
+
+    // ---- 2nd level of blocking on max(L2,L3), yields nc ----
+
+    // TODO find a reliable way to get the actual amount of cache per core to use for 2nd level blocking, that is:
+    //      actual_l2 = max(l2, l3/nb_core_sharing_l3)
+    // The number below is quite conservative: it is better to underestimate the cache size rather than overestimating it)
+    // For instance, it corresponds to 6MB of L3 shared among 4 cores.
+    #ifdef EIGEN_DEBUG_SMALL_PRODUCT_BLOCKS
+    const Index actual_l2 = l3;
+    #else
+    const Index actual_l2 = 1572864; // == 1.5 MB
+    #endif
+
+    // Here, nc is chosen such that a block of kc x nc of the rhs fit within half of L2.
+    // The second half is implicitly reserved to access the result and lhs coefficients.
+    // When k<max_kc, then nc can arbitrarily growth. In practice, it seems to be fruitful
+    // to limit this growth: we bound nc to growth by a factor x1.5.
+    // However, if the entire lhs block fit within L1, then we are not going to block on the rows at all,
+    // and it becomes fruitful to keep the packed rhs blocks in L1 if there is enough remaining space.
+    Index max_nc;
+    const Index lhs_bytes = m * k * sizeof(LhsScalar);
+    const Index remaining_l1 = l1- k_sub - lhs_bytes;
+    if(remaining_l1 >= Index(Traits::nr*sizeof(RhsScalar))*k)
+    {
+      // L1 blocking
+      max_nc = remaining_l1 / (k*sizeof(RhsScalar));
+    }
+    else
+    {
+      // L2 blocking
+      max_nc = (3*actual_l2)/(2*2*max_kc*sizeof(RhsScalar));
+    }
+    // WARNING Below, we assume that Traits::nr is a power of two.
+    Index nc = numext::mini<Index>(actual_l2/(2*k*sizeof(RhsScalar)), max_nc) & (~(Traits::nr-1));
+    if(n>nc)
+    {
+      // We are really blocking over the columns:
+      // -> reduce blocking size to make sure the last block is as large as possible
+      //    while keeping the same number of sweeps over the packed lhs.
+      //    Here we allow one more sweep if this gives us a perfect match, thus the commented "-1"
+      n = (n%nc)==0 ? nc
+                    : (nc - Traits::nr * ((nc/*-1*/-(n%nc))/(Traits::nr*(n/nc+1))));
+    }
+    else if(old_k==k)
+    {
+      // So far, no blocking at all, i.e., kc==k, and nc==n.
+      // In this case, let's perform a blocking over the rows such that the packed lhs data is kept in cache L1/L2
+      // TODO: part of this blocking strategy is now implemented within the kernel itself, so the L1-based heuristic here should be obsolete.
+      Index problem_size = k*n*sizeof(LhsScalar);
+      Index actual_lm = actual_l2;
+      Index max_mc = m;
+      if(problem_size<=1024)
+      {
+        // problem is small enough to keep in L1
+        // Let's choose m such that lhs's block fit in 1/3 of L1
+        actual_lm = l1;
+      }
+      else if(l3!=0 && problem_size<=32768)
+      {
+        // we have both L2 and L3, and problem is small enough to be kept in L2
+        // Let's choose m such that lhs's block fit in 1/3 of L2
+        actual_lm = l2;
+        max_mc = (numext::mini<Index>)(576,max_mc);
+      }
+      Index mc = (numext::mini<Index>)(actual_lm/(3*k*sizeof(LhsScalar)), max_mc);
+      if (mc > Traits::mr) mc -= mc % Traits::mr;
+      else if (mc==0) return;
+      m = (m%mc)==0 ? mc
+                    : (mc - Traits::mr * ((mc/*-1*/-(m%mc))/(Traits::mr*(m/mc+1))));
+    }
+  }
+}
+
+template <typename Index>
+inline bool useSpecificBlockingSizes(Index& k, Index& m, Index& n)
+{
+#ifdef EIGEN_TEST_SPECIFIC_BLOCKING_SIZES
+  if (EIGEN_TEST_SPECIFIC_BLOCKING_SIZES) {
+    k = numext::mini<Index>(k, EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_K);
+    m = numext::mini<Index>(m, EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_M);
+    n = numext::mini<Index>(n, EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_N);
+    return true;
+  }
+#else
+  EIGEN_UNUSED_VARIABLE(k)
+  EIGEN_UNUSED_VARIABLE(m)
+  EIGEN_UNUSED_VARIABLE(n)
+#endif
+  return false;
+}
+
 /** \brief Computes the blocking parameters for a m x k times k x n matrix product
   *
   * \param[in,out] k Input: the third dimension of the product. Output: the blocking size along the same dimension.
@@ -62,48 +285,30 @@
   *
   * Given a m x k times k x n matrix product of scalar types \c LhsScalar and \c RhsScalar,
   * this function computes the blocking size parameters along the respective dimensions
-  * for matrix products and related algorithms. The blocking sizes depends on various
-  * parameters:
-  * - the L1 and L2 cache sizes,
-  * - the register level blocking sizes defined by gebp_traits,
-  * - the number of scalars that fit into a packet (when vectorization is enabled).
+  * for matrix products and related algorithms.
+  *
+  * The blocking size parameters may be evaluated:
+  *   - either by a heuristic based on cache sizes;
+  *   - or using fixed prescribed values (for testing purposes).
   *
   * \sa setCpuCacheSizes */
-template<typename LhsScalar, typename RhsScalar, int KcFactor, typename SizeType>
-void computeProductBlockingSizes(SizeType& k, SizeType& m, SizeType& n)
+
+template<typename LhsScalar, typename RhsScalar, int KcFactor, typename Index>
+void computeProductBlockingSizes(Index& k, Index& m, Index& n, Index num_threads = 1)
 {
-  EIGEN_UNUSED_VARIABLE(n);
-  // Explanations:
-  // Let's recall the product algorithms form kc x nc horizontal panels B' on the rhs and
-  // mc x kc blocks A' on the lhs. A' has to fit into L2 cache. Moreover, B' is processed
-  // per kc x nr vertical small panels where nr is the blocking size along the n dimension
-  // at the register level. For vectorization purpose, these small vertical panels are unpacked,
-  // e.g., each coefficient is replicated to fit a packet. This small vertical panel has to
-  // stay in L1 cache.
-  std::ptrdiff_t l1, l2;
-
-  typedef gebp_traits<LhsScalar,RhsScalar> Traits;
-  enum {
-    kdiv = KcFactor * 2 * Traits::nr
-         * Traits::RhsProgress * sizeof(RhsScalar),
-    mr = gebp_traits<LhsScalar,RhsScalar>::mr,
-    mr_mask = (0xffffffff/mr)*mr
-  };
-
-  manage_caching_sizes(GetAction, &l1, &l2);
-  k = std::min<SizeType>(k, l1/kdiv);
-  SizeType _m = k>0 ? l2/(4 * sizeof(LhsScalar) * k) : 0;
-  if(_m<m) m = _m & mr_mask;
+  if (!useSpecificBlockingSizes(k, m, n)) {
+    evaluateProductBlockingSizesHeuristic<LhsScalar, RhsScalar, KcFactor, Index>(k, m, n, num_threads);
+  }
 }
 
-template<typename LhsScalar, typename RhsScalar, typename SizeType>
-inline void computeProductBlockingSizes(SizeType& k, SizeType& m, SizeType& n)
+template<typename LhsScalar, typename RhsScalar, typename Index>
+inline void computeProductBlockingSizes(Index& k, Index& m, Index& n, Index num_threads = 1)
 {
-  computeProductBlockingSizes<LhsScalar,RhsScalar,1>(k, m, n);
+  computeProductBlockingSizes<LhsScalar,RhsScalar,1,Index>(k, m, n, num_threads);
 }
 
-#ifdef EIGEN_HAS_FUSE_CJMADD
-  #define MADD(CJ,A,B,C,T)  C = CJ.pmadd(A,B,C);
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_CJMADD
+  #define CJMADD(CJ,A,B,C,T)  C = CJ.pmadd(A,B,C);
 #else
 
   // FIXME (a bit overkill maybe ?)
@@ -128,8 +333,8 @@
     gebp_madd_selector<CJ,A,B,C,T>::run(cj,a,b,c,t);
   }
 
-  #define MADD(CJ,A,B,C,T)  gebp_madd(CJ,A,B,C,T);
-//   #define MADD(CJ,A,B,C,T)  T = B; T = CJ.pmul(A,T); C = padd(C,T);
+  #define CJMADD(CJ,A,B,C,T)  gebp_madd(CJ,A,B,C,T);
+//   #define CJMADD(CJ,A,B,C,T)  T = B; T = CJ.pmul(A,T); C = padd(C,T);
 #endif
 
 /* Vectorization logic
@@ -148,7 +353,7 @@
 public:
   typedef _LhsScalar LhsScalar;
   typedef _RhsScalar RhsScalar;
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
   enum {
     ConjLhs = _ConjLhs,
@@ -160,16 +365,22 @@
     
     NumberOfRegisters = EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS,
 
-    // register block size along the N direction (must be either 2 or 4)
-    nr = NumberOfRegisters/4,
+    // register block size along the N direction must be 1 or 4
+    nr = 4,
 
     // register block size along the M direction (currently, this one cannot be modified)
-    mr = 2 * LhsPacketSize,
+    default_mr = (EIGEN_PLAIN_ENUM_MIN(16,NumberOfRegisters)/2/nr)*LhsPacketSize,
+#if defined(EIGEN_HAS_SINGLE_INSTRUCTION_MADD) && !defined(EIGEN_VECTORIZE_ALTIVEC) && !defined(EIGEN_VECTORIZE_VSX)
+    // we assume 16 registers
+    // See bug 992, if the scalar type is not vectorizable but that EIGEN_HAS_SINGLE_INSTRUCTION_MADD is defined,
+    // then using 3*LhsPacketSize triggers non-implemented paths in syrk.
+    mr = Vectorizable ? 3*LhsPacketSize : default_mr,
+#else
+    mr = default_mr,
+#endif
     
-    WorkSpaceFactor = nr * RhsPacketSize,
-
     LhsProgress = LhsPacketSize,
-    RhsProgress = RhsPacketSize
+    RhsProgress = 1
   };
 
   typedef typename packet_traits<LhsScalar>::type  _LhsPacket;
@@ -186,36 +397,67 @@
   {
     p = pset1<ResPacket>(ResScalar(0));
   }
-
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const RhsScalar* rhs, RhsScalar* b)
+  
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
   {
-    for(DenseIndex k=0; k<n; k++)
-      pstore1<RhsPacket>(&b[k*RhsPacketSize], rhs[k]);
+    pbroadcast4(b, b0, b1, b2, b3);
+  }
+  
+//   EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1)
+//   {
+//     pbroadcast2(b, b0, b1);
+//   }
+  
+  template<typename RhsPacketType>
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacketType& dest) const
+  {
+    dest = pset1<RhsPacketType>(*b);
+  }
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, RhsPacket& dest) const
+  {
+    dest = ploadquad<RhsPacket>(b);
   }
 
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
+  template<typename LhsPacketType>
+  EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacketType& dest) const
   {
-    dest = pload<RhsPacket>(b);
+    dest = pload<LhsPacketType>(a);
   }
 
-  EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacket& dest) const
+  template<typename LhsPacketType>
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacketType& dest) const
   {
-    dest = pload<LhsPacket>(a);
+    dest = ploadu<LhsPacketType>(a);
   }
 
-  EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, AccPacket& c, AccPacket& tmp) const
+  template<typename LhsPacketType, typename RhsPacketType, typename AccPacketType>
+  EIGEN_STRONG_INLINE void madd(const LhsPacketType& a, const RhsPacketType& b, AccPacketType& c, AccPacketType& tmp) const
   {
-    tmp = b; tmp = pmul(a,tmp); c = padd(c,tmp);
+    conj_helper<LhsPacketType,RhsPacketType,ConjLhs,ConjRhs> cj;
+    // It would be a lot cleaner to call pmadd all the time. Unfortunately if we
+    // let gcc allocate the register in which to store the result of the pmul
+    // (in the case where there is no FMA) gcc fails to figure out how to avoid
+    // spilling register.
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+    EIGEN_UNUSED_VARIABLE(tmp);
+    c = cj.pmadd(a,b,c);
+#else
+    tmp = b; tmp = cj.pmul(a,tmp); c = padd(c,tmp);
+#endif
   }
 
   EIGEN_STRONG_INLINE void acc(const AccPacket& c, const ResPacket& alpha, ResPacket& r) const
   {
     r = pmadd(c,alpha,r);
   }
+  
+  template<typename ResPacketHalf>
+  EIGEN_STRONG_INLINE void acc(const ResPacketHalf& c, const ResPacketHalf& alpha, ResPacketHalf& r) const
+  {
+    r = pmadd(c,alpha,r);
+  }
 
-protected:
-//   conj_helper<LhsScalar,RhsScalar,ConjLhs,ConjRhs> cj;
-//   conj_helper<LhsPacket,RhsPacket,ConjLhs,ConjRhs> pcj;
 };
 
 template<typename RealScalar, bool _ConjLhs>
@@ -224,7 +466,7 @@
 public:
   typedef std::complex<RealScalar> LhsScalar;
   typedef RealScalar RhsScalar;
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
   enum {
     ConjLhs = _ConjLhs,
@@ -235,12 +477,16 @@
     ResPacketSize = Vectorizable ? packet_traits<ResScalar>::size : 1,
     
     NumberOfRegisters = EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS,
-    nr = NumberOfRegisters/4,
-    mr = 2 * LhsPacketSize,
-    WorkSpaceFactor = nr*RhsPacketSize,
+    nr = 4,
+#if defined(EIGEN_HAS_SINGLE_INSTRUCTION_MADD) && !defined(EIGEN_VECTORIZE_ALTIVEC) && !defined(EIGEN_VECTORIZE_VSX)
+    // we assume 16 registers
+    mr = 3*LhsPacketSize,
+#else
+    mr = (EIGEN_PLAIN_ENUM_MIN(16,NumberOfRegisters)/2/nr)*LhsPacketSize,
+#endif
 
     LhsProgress = LhsPacketSize,
-    RhsProgress = RhsPacketSize
+    RhsProgress = 1
   };
 
   typedef typename packet_traits<LhsScalar>::type  _LhsPacket;
@@ -258,15 +504,14 @@
     p = pset1<ResPacket>(ResScalar(0));
   }
 
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const RhsScalar* rhs, RhsScalar* b)
-  {
-    for(DenseIndex k=0; k<n; k++)
-      pstore1<RhsPacket>(&b[k*RhsPacketSize], rhs[k]);
-  }
-
   EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
   {
-    dest = pload<RhsPacket>(b);
+    dest = pset1<RhsPacket>(*b);
+  }
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, RhsPacket& dest) const
+  {
+    dest = pset1<RhsPacket>(*b);
   }
 
   EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacket& dest) const
@@ -274,6 +519,21 @@
     dest = pload<LhsPacket>(a);
   }
 
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacket& dest) const
+  {
+    dest = ploadu<LhsPacket>(a);
+  }
+
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
+  {
+    pbroadcast4(b, b0, b1, b2, b3);
+  }
+  
+//   EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1)
+//   {
+//     pbroadcast2(b, b0, b1);
+//   }
+
   EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp) const
   {
     madd_impl(a, b, c, tmp, typename conditional<Vectorizable,true_type,false_type>::type());
@@ -281,7 +541,12 @@
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp, const true_type&) const
   {
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+    EIGEN_UNUSED_VARIABLE(tmp);
+    c.v = pmadd(a.v,b,c.v);
+#else
     tmp = b; tmp = pmul(a.v,tmp); c.v = padd(c.v,tmp);
+#endif
   }
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsScalar& a, const RhsScalar& b, ResScalar& c, RhsScalar& /*tmp*/, const false_type&) const
@@ -298,6 +563,38 @@
   conj_helper<ResPacket,ResPacket,ConjLhs,false> cj;
 };
 
+template<typename Packet>
+struct DoublePacket
+{
+  Packet first;
+  Packet second;
+};
+
+template<typename Packet>
+DoublePacket<Packet> padd(const DoublePacket<Packet> &a, const DoublePacket<Packet> &b)
+{
+  DoublePacket<Packet> res;
+  res.first  = padd(a.first, b.first);
+  res.second = padd(a.second,b.second);
+  return res;
+}
+
+template<typename Packet>
+const DoublePacket<Packet>& predux_downto4(const DoublePacket<Packet> &a)
+{
+  return a;
+}
+
+template<typename Packet> struct unpacket_traits<DoublePacket<Packet> > { typedef DoublePacket<Packet> half; };
+// template<typename Packet>
+// DoublePacket<Packet> pmadd(const DoublePacket<Packet> &a, const DoublePacket<Packet> &b)
+// {
+//   DoublePacket<Packet> res;
+//   res.first  = padd(a.first, b.first);
+//   res.second = padd(a.second,b.second);
+//   return res;
+// }
+
 template<typename RealScalar, bool _ConjLhs, bool _ConjRhs>
 class gebp_traits<std::complex<RealScalar>, std::complex<RealScalar>, _ConjLhs, _ConjRhs >
 {
@@ -314,60 +611,80 @@
                 && packet_traits<Scalar>::Vectorizable,
     RealPacketSize  = Vectorizable ? packet_traits<RealScalar>::size : 1,
     ResPacketSize   = Vectorizable ? packet_traits<ResScalar>::size : 1,
-    
-    nr = 2,
-    mr = 2 * ResPacketSize,
-    WorkSpaceFactor = Vectorizable ? 2*nr*RealPacketSize : nr,
+    LhsPacketSize = Vectorizable ? packet_traits<LhsScalar>::size : 1,
+    RhsPacketSize = Vectorizable ? packet_traits<RhsScalar>::size : 1,
+
+    // FIXME: should depend on NumberOfRegisters
+    nr = 4,
+    mr = ResPacketSize,
 
     LhsProgress = ResPacketSize,
-    RhsProgress = Vectorizable ? 2*ResPacketSize : 1
+    RhsProgress = 1
   };
   
   typedef typename packet_traits<RealScalar>::type RealPacket;
   typedef typename packet_traits<Scalar>::type     ScalarPacket;
-  struct DoublePacket
-  {
-    RealPacket first;
-    RealPacket second;
-  };
+  typedef DoublePacket<RealPacket> DoublePacketType;
 
   typedef typename conditional<Vectorizable,RealPacket,  Scalar>::type LhsPacket;
-  typedef typename conditional<Vectorizable,DoublePacket,Scalar>::type RhsPacket;
+  typedef typename conditional<Vectorizable,DoublePacketType,Scalar>::type RhsPacket;
   typedef typename conditional<Vectorizable,ScalarPacket,Scalar>::type ResPacket;
-  typedef typename conditional<Vectorizable,DoublePacket,Scalar>::type AccPacket;
+  typedef typename conditional<Vectorizable,DoublePacketType,Scalar>::type AccPacket;
   
   EIGEN_STRONG_INLINE void initAcc(Scalar& p) { p = Scalar(0); }
 
-  EIGEN_STRONG_INLINE void initAcc(DoublePacket& p)
+  EIGEN_STRONG_INLINE void initAcc(DoublePacketType& p)
   {
     p.first   = pset1<RealPacket>(RealScalar(0));
     p.second  = pset1<RealPacket>(RealScalar(0));
   }
 
-  /* Unpack the rhs coeff such that each complex coefficient is spread into
-   * two packects containing respectively the real and imaginary coefficient
-   * duplicated as many time as needed: (x+iy) => [x, ..., x] [y, ..., y]
-   */
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const Scalar* rhs, Scalar* b)
+  // Scalar path
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, ResPacket& dest) const
   {
-    for(DenseIndex k=0; k<n; k++)
-    {
-      if(Vectorizable)
-      {
-        pstore1<RealPacket>((RealScalar*)&b[k*ResPacketSize*2+0],             real(rhs[k]));
-        pstore1<RealPacket>((RealScalar*)&b[k*ResPacketSize*2+ResPacketSize], imag(rhs[k]));
-      }
-      else
-        b[k] = rhs[k];
-    }
+    dest = pset1<ResPacket>(*b);
   }
 
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, ResPacket& dest) const { dest = *b; }
-
-  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, DoublePacket& dest) const
+  // Vectorized path
+  EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, DoublePacketType& dest) const
   {
-    dest.first  = pload<RealPacket>((const RealScalar*)b);
-    dest.second = pload<RealPacket>((const RealScalar*)(b+ResPacketSize));
+    dest.first  = pset1<RealPacket>(real(*b));
+    dest.second = pset1<RealPacket>(imag(*b));
+  }
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, ResPacket& dest) const
+  {
+    loadRhs(b,dest);
+  }
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, DoublePacketType& dest) const
+  {
+    eigen_internal_assert(unpacket_traits<ScalarPacket>::size<=4);
+    loadRhs(b,dest);
+  }
+  
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
+  {
+    // FIXME not sure that's the best way to implement it!
+    loadRhs(b+0, b0);
+    loadRhs(b+1, b1);
+    loadRhs(b+2, b2);
+    loadRhs(b+3, b3);
+  }
+  
+  // Vectorized path
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, DoublePacketType& b0, DoublePacketType& b1)
+  {
+    // FIXME not sure that's the best way to implement it!
+    loadRhs(b+0, b0);
+    loadRhs(b+1, b1);
+  }
+  
+  // Scalar path
+  EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsScalar& b0, RhsScalar& b1)
+  {
+    // FIXME not sure that's the best way to implement it!
+    loadRhs(b+0, b0);
+    loadRhs(b+1, b1);
   }
 
   // nothing special here
@@ -376,7 +693,12 @@
     dest = pload<LhsPacket>((const typename unpacket_traits<LhsPacket>::type*)(a));
   }
 
-  EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, DoublePacket& c, RhsPacket& /*tmp*/) const
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacket& dest) const
+  {
+    dest = ploadu<LhsPacket>((const typename unpacket_traits<LhsPacket>::type*)(a));
+  }
+
+  EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, DoublePacketType& c, RhsPacket& /*tmp*/) const
   {
     c.first   = padd(pmul(a,b.first), c.first);
     c.second  = padd(pmul(a,b.second),c.second);
@@ -389,7 +711,7 @@
   
   EIGEN_STRONG_INLINE void acc(const Scalar& c, const Scalar& alpha, Scalar& r) const { r += alpha * c; }
   
-  EIGEN_STRONG_INLINE void acc(const DoublePacket& c, const ResPacket& alpha, ResPacket& r) const
+  EIGEN_STRONG_INLINE void acc(const DoublePacketType& c, const ResPacket& alpha, ResPacket& r) const
   {
     // assemble c
     ResPacket tmp;
@@ -440,12 +762,12 @@
     ResPacketSize = Vectorizable ? packet_traits<ResScalar>::size : 1,
     
     NumberOfRegisters = EIGEN_ARCH_DEFAULT_NUMBER_OF_REGISTERS,
+    // FIXME: should depend on NumberOfRegisters
     nr = 4,
-    mr = 2*ResPacketSize,
-    WorkSpaceFactor = nr*RhsPacketSize,
+    mr = (EIGEN_PLAIN_ENUM_MIN(16,NumberOfRegisters)/2/nr)*ResPacketSize,
 
     LhsProgress = ResPacketSize,
-    RhsProgress = ResPacketSize
+    RhsProgress = 1
   };
 
   typedef typename packet_traits<LhsScalar>::type  _LhsPacket;
@@ -463,21 +785,38 @@
     p = pset1<ResPacket>(ResScalar(0));
   }
 
-  EIGEN_STRONG_INLINE void unpackRhs(DenseIndex n, const RhsScalar* rhs, RhsScalar* b)
-  {
-    for(DenseIndex k=0; k<n; k++)
-      pstore1<RhsPacket>(&b[k*RhsPacketSize], rhs[k]);
-  }
-
   EIGEN_STRONG_INLINE void loadRhs(const RhsScalar* b, RhsPacket& dest) const
   {
-    dest = pload<RhsPacket>(b);
+    dest = pset1<RhsPacket>(*b);
   }
+  
+  void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1, RhsPacket& b2, RhsPacket& b3)
+  {
+    pbroadcast4(b, b0, b1, b2, b3);
+  }
+  
+//   EIGEN_STRONG_INLINE void broadcastRhs(const RhsScalar* b, RhsPacket& b0, RhsPacket& b1)
+//   {
+//     // FIXME not sure that's the best way to implement it!
+//     b0 = pload1<RhsPacket>(b+0);
+//     b1 = pload1<RhsPacket>(b+1);
+//   }
 
   EIGEN_STRONG_INLINE void loadLhs(const LhsScalar* a, LhsPacket& dest) const
   {
     dest = ploaddup<LhsPacket>(a);
   }
+  
+  EIGEN_STRONG_INLINE void loadRhsQuad(const RhsScalar* b, RhsPacket& dest) const
+  {
+    eigen_internal_assert(unpacket_traits<RhsPacket>::size<=4);
+    loadRhs(b,dest);
+  }
+
+  EIGEN_STRONG_INLINE void loadLhsUnaligned(const LhsScalar* a, LhsPacket& dest) const
+  {
+    dest = ploaddup<LhsPacket>(a);
+  }
 
   EIGEN_STRONG_INLINE void madd(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp) const
   {
@@ -486,7 +825,13 @@
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsPacket& a, const RhsPacket& b, AccPacket& c, RhsPacket& tmp, const true_type&) const
   {
+#ifdef EIGEN_HAS_SINGLE_INSTRUCTION_MADD
+    EIGEN_UNUSED_VARIABLE(tmp);
+    c.v = pmadd(a,b.v,c.v);
+#else
     tmp = b; tmp.v = pmul(a,tmp.v); c = padd(c,tmp);
+#endif
+    
   }
 
   EIGEN_STRONG_INLINE void madd_impl(const LhsScalar& a, const RhsScalar& b, ResScalar& c, RhsScalar& /*tmp*/, const false_type&) const
@@ -510,7 +855,7 @@
  *  |real |cplx | no vectorization yet, would require to pack A with duplication
  *  |cplx |real | easy vectorization
  */
-template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
+template<typename LhsScalar, typename RhsScalar, typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
 struct gebp_kernel
 {
   typedef gebp_traits<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> Traits;
@@ -520,6 +865,15 @@
   typedef typename Traits::ResPacket ResPacket;
   typedef typename Traits::AccPacket AccPacket;
 
+  typedef gebp_traits<RhsScalar,LhsScalar,ConjugateRhs,ConjugateLhs> SwappedTraits;
+  typedef typename SwappedTraits::ResScalar SResScalar;
+  typedef typename SwappedTraits::LhsPacket SLhsPacket;
+  typedef typename SwappedTraits::RhsPacket SRhsPacket;
+  typedef typename SwappedTraits::ResPacket SResPacket;
+  typedef typename SwappedTraits::AccPacket SAccPacket;
+
+  typedef typename DataMapper::LinearMapper LinearMapper;
+
   enum {
     Vectorizable  = Traits::Vectorizable,
     LhsProgress   = Traits::LhsProgress,
@@ -528,571 +882,795 @@
   };
 
   EIGEN_DONT_INLINE
-  void operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index rows, Index depth, Index cols, ResScalar alpha,
-                  Index strideA=-1, Index strideB=-1, Index offsetA=0, Index offsetB=0, RhsScalar* unpackedB=0);
+  void operator()(const DataMapper& res, const LhsScalar* blockA, const RhsScalar* blockB,
+                  Index rows, Index depth, Index cols, ResScalar alpha,
+                  Index strideA=-1, Index strideB=-1, Index offsetA=0, Index offsetB=0);
 };
 
-template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
+template<typename LhsScalar, typename RhsScalar, typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs, bool ConjugateRhs>
 EIGEN_DONT_INLINE
-void gebp_kernel<LhsScalar,RhsScalar,Index,mr,nr,ConjugateLhs,ConjugateRhs>
-  ::operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index rows, Index depth, Index cols, ResScalar alpha,
-               Index strideA, Index strideB, Index offsetA, Index offsetB, RhsScalar* unpackedB)
+void gebp_kernel<LhsScalar,RhsScalar,Index,DataMapper,mr,nr,ConjugateLhs,ConjugateRhs>
+  ::operator()(const DataMapper& res, const LhsScalar* blockA, const RhsScalar* blockB,
+               Index rows, Index depth, Index cols, ResScalar alpha,
+               Index strideA, Index strideB, Index offsetA, Index offsetB)
   {
     Traits traits;
+    SwappedTraits straits;
     
     if(strideA==-1) strideA = depth;
     if(strideB==-1) strideB = depth;
     conj_helper<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> cj;
-//     conj_helper<LhsPacket,RhsPacket,ConjugateLhs,ConjugateRhs> pcj;
-    Index packet_cols = (cols/nr) * nr;
-    const Index peeled_mc = (rows/mr)*mr;
-    // FIXME:
-    const Index peeled_mc2 = peeled_mc + (rows-peeled_mc >= LhsProgress ? LhsProgress : 0);
-    const Index peeled_kc = (depth/4)*4;
+    Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
+    const Index peeled_mc3 = mr>=3*Traits::LhsProgress ? (rows/(3*LhsProgress))*(3*LhsProgress) : 0;
+    const Index peeled_mc2 = mr>=2*Traits::LhsProgress ? peeled_mc3+((rows-peeled_mc3)/(2*LhsProgress))*(2*LhsProgress) : 0;
+    const Index peeled_mc1 = mr>=1*Traits::LhsProgress ? (rows/(1*LhsProgress))*(1*LhsProgress) : 0;
+    enum { pk = 8 }; // NOTE Such a large peeling factor is important for large matrices (~ +5% when >1000 on Haswell)
+    const Index peeled_kc  = depth & ~(pk-1);
+    const Index prefetch_res_offset = 32/sizeof(ResScalar);    
+//     const Index depth2     = depth & ~1;
 
-    if(unpackedB==0)
-      unpackedB = const_cast<RhsScalar*>(blockB - strideB * nr * RhsProgress);
-
-    // loops on each micro vertical panel of rhs (depth x nr)
-    for(Index j2=0; j2<packet_cols; j2+=nr)
+    //---------- Process 3 * LhsProgress rows at once ----------
+    // This corresponds to 3*LhsProgress x nr register blocks.
+    // Usually, make sense only with FMA
+    if(mr>=3*Traits::LhsProgress)
     {
-      traits.unpackRhs(depth*nr,&blockB[j2*strideB+offsetB*nr],unpackedB); 
-
-      // loops on each largest micro horizontal panel of lhs (mr x depth)
-      // => we select a mr x nr micro block of res which is entirely
-      //    stored into mr/packet_size x nr registers.
-      for(Index i=0; i<peeled_mc; i+=mr)
+      // Here, the general idea is to loop on each largest micro horizontal panel of the lhs (3*Traits::LhsProgress x depth)
+      // and on each largest micro vertical panel of the rhs (depth * nr).
+      // Blocking sizes, i.e., 'depth' has been computed so that the micro horizontal panel of the lhs fit in L1.
+      // However, if depth is too small, we can extend the number of rows of these horizontal panels.
+      // This actual number of rows is computed as follow:
+      const Index l1 = defaultL1CacheSize; // in Bytes, TODO, l1 should be passed to this function.
+      // The max(1, ...) here is needed because we may be using blocking params larger than what our known l1 cache size
+      // suggests we should be using: either because our known l1 cache size is inaccurate (e.g. on Android, we can only guess),
+      // or because we are testing specific blocking sizes.
+      const Index actual_panel_rows = (3*LhsProgress) * std::max<Index>(1,( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 3*LhsProgress) ));
+      for(Index i1=0; i1<peeled_mc3; i1+=actual_panel_rows)
       {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*mr];
-        prefetch(&blA[0]);
-
-        // gets res block as register
-        AccPacket C0, C1, C2, C3, C4, C5, C6, C7;
-                  traits.initAcc(C0);
-                  traits.initAcc(C1);
-        if(nr==4) traits.initAcc(C2);
-        if(nr==4) traits.initAcc(C3);
-                  traits.initAcc(C4);
-                  traits.initAcc(C5);
-        if(nr==4) traits.initAcc(C6);
-        if(nr==4) traits.initAcc(C7);
-
-        ResScalar* r0 = &res[(j2+0)*resStride + i];
-        ResScalar* r1 = r0 + resStride;
-        ResScalar* r2 = r1 + resStride;
-        ResScalar* r3 = r2 + resStride;
-
-        prefetch(r0+16);
-        prefetch(r1+16);
-        prefetch(r2+16);
-        prefetch(r3+16);
-
-        // performs "inner" product
-        // TODO let's check wether the folowing peeled loop could not be
-        //      optimized via optimal prefetching from one loop to the other
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<peeled_kc; k+=4)
+        const Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc3);
+        for(Index j2=0; j2<packet_cols4; j2+=nr)
         {
-          if(nr==2)
+          for(Index i=i1; i<actual_panel_end; i+=3*LhsProgress)
           {
-            LhsPacket A0, A1;
-            RhsPacket B_0;
-            RhsPacket T0;
-            
-EIGEN_ASM_COMMENT("mybegin2");
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.loadLhs(&blA[3*LhsProgress], A1);
-            traits.loadRhs(&blB[2*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[3*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-
-            traits.loadLhs(&blA[4*LhsProgress], A0);
-            traits.loadLhs(&blA[5*LhsProgress], A1);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[5*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-
-            traits.loadLhs(&blA[6*LhsProgress], A0);
-            traits.loadLhs(&blA[7*LhsProgress], A1);
-            traits.loadRhs(&blB[6*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[7*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-EIGEN_ASM_COMMENT("myend");
-          }
-          else
-          {
-EIGEN_ASM_COMMENT("mybegin4");
-            LhsPacket A0, A1;
-            RhsPacket B_0, B1, B2, B3;
-            RhsPacket T0;
-            
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-
-            traits.madd(A0,B_0,C0,T0);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.loadRhs(&blB[5*RhsProgress], B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.loadRhs(&blB[6*RhsProgress], B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.madd(A1,B3,C7,B3);
-            traits.loadLhs(&blA[3*LhsProgress], A1);
-            traits.loadRhs(&blB[7*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[8*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.loadRhs(&blB[9*RhsProgress], B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.loadRhs(&blB[10*RhsProgress], B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.loadLhs(&blA[4*LhsProgress], A0);
-            traits.madd(A1,B3,C7,B3);
-            traits.loadLhs(&blA[5*LhsProgress], A1);
-            traits.loadRhs(&blB[11*RhsProgress], B3);
-
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[12*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.loadRhs(&blB[13*RhsProgress], B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.loadRhs(&blB[14*RhsProgress], B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.loadLhs(&blA[6*LhsProgress], A0);
-            traits.madd(A1,B3,C7,B3);
-            traits.loadLhs(&blA[7*LhsProgress], A1);
-            traits.loadRhs(&blB[15*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.madd(A1,B3,C7,B3);
-          }
-
-          blB += 4*nr*RhsProgress;
-          blA += 4*mr;
-        }
-        // process remaining peeled loop
-        for(Index k=peeled_kc; k<depth; k++)
-        {
-          if(nr==2)
-          {
-            LhsPacket A0, A1;
-            RhsPacket B_0;
-            RhsPacket T0;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.madd(A0,B_0,C0,T0);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B_0);
-            traits.madd(A0,B_0,C1,T0);
-            traits.madd(A1,B_0,C5,B_0);
-          }
-          else
-          {
-            LhsPacket A0, A1;
-            RhsPacket B_0, B1, B2, B3;
-            RhsPacket T0;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadLhs(&blA[1*LhsProgress], A1);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-
-            traits.madd(A0,B_0,C0,T0);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.madd(A1,B_0,C4,B_0);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
-            traits.madd(A0,B1,C1,T0);
-            traits.madd(A1,B1,C5,B1);
-            traits.madd(A0,B2,C2,T0);
-            traits.madd(A1,B2,C6,B2);
-            traits.madd(A0,B3,C3,T0);
-            traits.madd(A1,B3,C7,B3);
-          }
-
-          blB += nr*RhsProgress;
-          blA += mr;
-        }
-
-        if(nr==4)
-        {
-          ResPacket R0, R1, R2, R3, R4, R5, R6;
-          ResPacket alphav = pset1<ResPacket>(alpha);
-
-          R0 = ploadu<ResPacket>(r0);
-          R1 = ploadu<ResPacket>(r1);
-          R2 = ploadu<ResPacket>(r2);
-          R3 = ploadu<ResPacket>(r3);
-          R4 = ploadu<ResPacket>(r0 + ResPacketSize);
-          R5 = ploadu<ResPacket>(r1 + ResPacketSize);
-          R6 = ploadu<ResPacket>(r2 + ResPacketSize);
-          traits.acc(C0, alphav, R0);
-          pstoreu(r0, R0);
-          R0 = ploadu<ResPacket>(r3 + ResPacketSize);
-
-          traits.acc(C1, alphav, R1);
-          traits.acc(C2, alphav, R2);
-          traits.acc(C3, alphav, R3);
-          traits.acc(C4, alphav, R4);
-          traits.acc(C5, alphav, R5);
-          traits.acc(C6, alphav, R6);
-          traits.acc(C7, alphav, R0);
           
-          pstoreu(r1, R1);
-          pstoreu(r2, R2);
-          pstoreu(r3, R3);
-          pstoreu(r0 + ResPacketSize, R4);
-          pstoreu(r1 + ResPacketSize, R5);
-          pstoreu(r2 + ResPacketSize, R6);
-          pstoreu(r3 + ResPacketSize, R0);
-        }
-        else
-        {
-          ResPacket R0, R1, R4;
+          // We selected a 3*Traits::LhsProgress x nr micro block of res which is entirely
+          // stored into 3 x nr registers.
+          
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(3*LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C1, C2,  C3,
+                    C4, C5, C6,  C7,
+                    C8, C9, C10, C11;
+          traits.initAcc(C0);  traits.initAcc(C1);  traits.initAcc(C2);  traits.initAcc(C3);
+          traits.initAcc(C4);  traits.initAcc(C5);  traits.initAcc(C6);  traits.initAcc(C7);
+          traits.initAcc(C8);  traits.initAcc(C9);  traits.initAcc(C10); traits.initAcc(C11);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2 + 0);
+          LinearMapper r1 = res.getLinearMapper(i, j2 + 1);
+          LinearMapper r2 = res.getLinearMapper(i, j2 + 2);
+          LinearMapper r3 = res.getLinearMapper(i, j2 + 3);
+
+          r0.prefetch(0);
+          r1.prefetch(0);
+          r2.prefetch(0);
+          r3.prefetch(0);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+          prefetch(&blB[0]);
+          LhsPacket A0, A1;
+
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 3pX4");
+            RhsPacket B_0, T0;
+            LhsPacket A2;
+
+#define EIGEN_GEBP_ONESTEP(K) \
+            do { \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 3pX4"); \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              internal::prefetch(blA+(3*K+16)*LhsProgress); \
+              if (EIGEN_ARCH_ARM) { internal::prefetch(blB+(4*K+16)*RhsProgress); } /* Bug 953 */ \
+              traits.loadLhs(&blA[(0+3*K)*LhsProgress], A0);  \
+              traits.loadLhs(&blA[(1+3*K)*LhsProgress], A1);  \
+              traits.loadLhs(&blA[(2+3*K)*LhsProgress], A2);  \
+              traits.loadRhs(blB + (0+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C0, T0); \
+              traits.madd(A1, B_0, C4, T0); \
+              traits.madd(A2, B_0, C8, B_0); \
+              traits.loadRhs(blB + (1+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C1, T0); \
+              traits.madd(A1, B_0, C5, T0); \
+              traits.madd(A2, B_0, C9, B_0); \
+              traits.loadRhs(blB + (2+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C2,  T0); \
+              traits.madd(A1, B_0, C6,  T0); \
+              traits.madd(A2, B_0, C10, B_0); \
+              traits.loadRhs(blB + (3+4*K)*Traits::RhsProgress, B_0); \
+              traits.madd(A0, B_0, C3 , T0); \
+              traits.madd(A1, B_0, C7,  T0); \
+              traits.madd(A2, B_0, C11, B_0); \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 3pX4"); \
+            } while(false)
+
+            internal::prefetch(blB);
+            EIGEN_GEBP_ONESTEP(0);
+            EIGEN_GEBP_ONESTEP(1);
+            EIGEN_GEBP_ONESTEP(2);
+            EIGEN_GEBP_ONESTEP(3);
+            EIGEN_GEBP_ONESTEP(4);
+            EIGEN_GEBP_ONESTEP(5);
+            EIGEN_GEBP_ONESTEP(6);
+            EIGEN_GEBP_ONESTEP(7);
+
+            blB += pk*4*RhsProgress;
+            blA += pk*3*Traits::LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 3pX4");
+          }
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0, T0;
+            LhsPacket A2;
+            EIGEN_GEBP_ONESTEP(0);
+            blB += 4*RhsProgress;
+            blA += 3*Traits::LhsProgress;
+          }
+
+#undef EIGEN_GEBP_ONESTEP
+
+          ResPacket R0, R1, R2;
           ResPacket alphav = pset1<ResPacket>(alpha);
 
-          R0 = ploadu<ResPacket>(r0);
-          R1 = ploadu<ResPacket>(r1);
-          R4 = ploadu<ResPacket>(r0 + ResPacketSize);
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r0.loadPacket(2 * Traits::ResPacketSize);
           traits.acc(C0, alphav, R0);
-          pstoreu(r0, R0);
-          R0 = ploadu<ResPacket>(r1 + ResPacketSize);
-          traits.acc(C1, alphav, R1);
-          traits.acc(C4, alphav, R4);
-          traits.acc(C5, alphav, R0);
-          pstoreu(r1, R1);
-          pstoreu(r0 + ResPacketSize, R4);
-          pstoreu(r1 + ResPacketSize, R0);
+          traits.acc(C4, alphav, R1);
+          traits.acc(C8, alphav, R2);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          r0.storePacket(2 * Traits::ResPacketSize, R2);
+
+          R0 = r1.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r1.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r1.loadPacket(2 * Traits::ResPacketSize);
+          traits.acc(C1, alphav, R0);
+          traits.acc(C5, alphav, R1);
+          traits.acc(C9, alphav, R2);
+          r1.storePacket(0 * Traits::ResPacketSize, R0);
+          r1.storePacket(1 * Traits::ResPacketSize, R1);
+          r1.storePacket(2 * Traits::ResPacketSize, R2);
+
+          R0 = r2.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r2.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r2.loadPacket(2 * Traits::ResPacketSize);
+          traits.acc(C2, alphav, R0);
+          traits.acc(C6, alphav, R1);
+          traits.acc(C10, alphav, R2);
+          r2.storePacket(0 * Traits::ResPacketSize, R0);
+          r2.storePacket(1 * Traits::ResPacketSize, R1);
+          r2.storePacket(2 * Traits::ResPacketSize, R2);
+
+          R0 = r3.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r3.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r3.loadPacket(2 * Traits::ResPacketSize);
+          traits.acc(C3, alphav, R0);
+          traits.acc(C7, alphav, R1);
+          traits.acc(C11, alphav, R2);
+          r3.storePacket(0 * Traits::ResPacketSize, R0);
+          r3.storePacket(1 * Traits::ResPacketSize, R1);
+          r3.storePacket(2 * Traits::ResPacketSize, R2);          
+          }
         }
+
+        // Deal with remaining columns of the rhs
+        for(Index j2=packet_cols4; j2<cols; j2++)
+        {
+          for(Index i=i1; i<actual_panel_end; i+=3*LhsProgress)
+          {
+          // One column at a time
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(3*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C4, C8;
+          traits.initAcc(C0);
+          traits.initAcc(C4);
+          traits.initAcc(C8);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2);
+          r0.prefetch(0);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          LhsPacket A0, A1, A2;
+          
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 3pX1");
+            RhsPacket B_0;
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do { \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 3pX1"); \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+3*K)*LhsProgress], A0);  \
+              traits.loadLhs(&blA[(1+3*K)*LhsProgress], A1);  \
+              traits.loadLhs(&blA[(2+3*K)*LhsProgress], A2);  \
+              traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);   \
+              traits.madd(A0, B_0, C0, B_0); \
+              traits.madd(A1, B_0, C4, B_0); \
+              traits.madd(A2, B_0, C8, B_0); \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 3pX1"); \
+            } while(false)
         
-      }
-      
-      if(rows-peeled_mc>=LhsProgress)
-      {
-        Index i = peeled_mc;
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*LhsProgress];
-        prefetch(&blA[0]);
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
 
-        // gets res block as register
-        AccPacket C0, C1, C2, C3;
-                  traits.initAcc(C0);
-                  traits.initAcc(C1);
-        if(nr==4) traits.initAcc(C2);
-        if(nr==4) traits.initAcc(C3);
+            blB += pk*RhsProgress;
+            blA += pk*3*Traits::LhsProgress;
 
-        // performs "inner" product
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<peeled_kc; k+=4)
-        {
-          if(nr==2)
-          {
-            LhsPacket A0;
-            RhsPacket B_0, B1;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[2*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadLhs(&blA[1*LhsProgress], A0);
-            traits.loadRhs(&blB[3*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.loadRhs(&blB[5*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[6*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadLhs(&blA[3*LhsProgress], A0);
-            traits.loadRhs(&blB[7*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
-          }
-          else
-          {
-            LhsPacket A0;
-            RhsPacket B_0, B1, B2, B3;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
-            traits.loadRhs(&blB[4*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadRhs(&blB[5*RhsProgress], B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.loadRhs(&blB[6*RhsProgress], B2);
-            traits.madd(A0,B3,C3,B3);
-            traits.loadLhs(&blA[1*LhsProgress], A0);
-            traits.loadRhs(&blB[7*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[8*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadRhs(&blB[9*RhsProgress], B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.loadRhs(&blB[10*RhsProgress], B2);
-            traits.madd(A0,B3,C3,B3);
-            traits.loadLhs(&blA[2*LhsProgress], A0);
-            traits.loadRhs(&blB[11*RhsProgress], B3);
-
-            traits.madd(A0,B_0,C0,B_0);
-            traits.loadRhs(&blB[12*RhsProgress], B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.loadRhs(&blB[13*RhsProgress], B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.loadRhs(&blB[14*RhsProgress], B2);
-            traits.madd(A0,B3,C3,B3);
-
-            traits.loadLhs(&blA[3*LhsProgress], A0);
-            traits.loadRhs(&blB[15*RhsProgress], B3);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.madd(A0,B3,C3,B3);
+            EIGEN_ASM_COMMENT("end gebp micro kernel 3pX1");
           }
 
-          blB += nr*4*RhsProgress;
-          blA += 4*LhsProgress;
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += RhsProgress;
+            blA += 3*Traits::LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
+          ResPacket R0, R1, R2;
+          ResPacket alphav = pset1<ResPacket>(alpha);
+
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r0.loadPacket(2 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          traits.acc(C4, alphav, R1);
+          traits.acc(C8, alphav, R2);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          r0.storePacket(2 * Traits::ResPacketSize, R2);          
+          }
         }
-        // process remaining peeled loop
-        for(Index k=peeled_kc; k<depth; k++)
-        {
-          if(nr==2)
-          {
-            LhsPacket A0;
-            RhsPacket B_0, B1;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
-          }
-          else
-          {
-            LhsPacket A0;
-            RhsPacket B_0, B1, B2, B3;
-
-            traits.loadLhs(&blA[0*LhsProgress], A0);
-            traits.loadRhs(&blB[0*RhsProgress], B_0);
-            traits.loadRhs(&blB[1*RhsProgress], B1);
-            traits.loadRhs(&blB[2*RhsProgress], B2);
-            traits.loadRhs(&blB[3*RhsProgress], B3);
-
-            traits.madd(A0,B_0,C0,B_0);
-            traits.madd(A0,B1,C1,B1);
-            traits.madd(A0,B2,C2,B2);
-            traits.madd(A0,B3,C3,B3);
-          }
-
-          blB += nr*RhsProgress;
-          blA += LhsProgress;
-        }
-
-        ResPacket R0, R1, R2, R3;
-        ResPacket alphav = pset1<ResPacket>(alpha);
-
-        ResScalar* r0 = &res[(j2+0)*resStride + i];
-        ResScalar* r1 = r0 + resStride;
-        ResScalar* r2 = r1 + resStride;
-        ResScalar* r3 = r2 + resStride;
-
-                  R0 = ploadu<ResPacket>(r0);
-                  R1 = ploadu<ResPacket>(r1);
-        if(nr==4) R2 = ploadu<ResPacket>(r2);
-        if(nr==4) R3 = ploadu<ResPacket>(r3);
-
-                  traits.acc(C0, alphav, R0);
-                  traits.acc(C1, alphav, R1);
-        if(nr==4) traits.acc(C2, alphav, R2);
-        if(nr==4) traits.acc(C3, alphav, R3);
-
-                  pstoreu(r0, R0);
-                  pstoreu(r1, R1);
-        if(nr==4) pstoreu(r2, R2);
-        if(nr==4) pstoreu(r3, R3);
-      }
-      for(Index i=peeled_mc2; i<rows; i++)
-      {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA];
-        prefetch(&blA[0]);
-
-        // gets a 1 x nr res block as registers
-        ResScalar C0(0), C1(0), C2(0), C3(0);
-        // TODO directly use blockB ???
-        const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
-        for(Index k=0; k<depth; k++)
-        {
-          if(nr==2)
-          {
-            LhsScalar A0;
-            RhsScalar B_0, B1;
-
-            A0 = blA[k];
-            B_0 = blB[0];
-            B1 = blB[1];
-            MADD(cj,A0,B_0,C0,B_0);
-            MADD(cj,A0,B1,C1,B1);
-          }
-          else
-          {
-            LhsScalar A0;
-            RhsScalar B_0, B1, B2, B3;
-
-            A0 = blA[k];
-            B_0 = blB[0];
-            B1 = blB[1];
-            B2 = blB[2];
-            B3 = blB[3];
-
-            MADD(cj,A0,B_0,C0,B_0);
-            MADD(cj,A0,B1,C1,B1);
-            MADD(cj,A0,B2,C2,B2);
-            MADD(cj,A0,B3,C3,B3);
-          }
-
-          blB += nr;
-        }
-                  res[(j2+0)*resStride + i] += alpha*C0;
-                  res[(j2+1)*resStride + i] += alpha*C1;
-        if(nr==4) res[(j2+2)*resStride + i] += alpha*C2;
-        if(nr==4) res[(j2+3)*resStride + i] += alpha*C3;
       }
     }
-    // process remaining rhs/res columns one at a time
-    // => do the same but with nr==1
-    for(Index j2=packet_cols; j2<cols; j2++)
+
+    //---------- Process 2 * LhsProgress rows at once ----------
+    if(mr>=2*Traits::LhsProgress)
     {
-      // unpack B
-      traits.unpackRhs(depth, &blockB[j2*strideB+offsetB], unpackedB);
+      const Index l1 = defaultL1CacheSize; // in Bytes, TODO, l1 should be passed to this function.
+      // The max(1, ...) here is needed because we may be using blocking params larger than what our known l1 cache size
+      // suggests we should be using: either because our known l1 cache size is inaccurate (e.g. on Android, we can only guess),
+      // or because we are testing specific blocking sizes.
+      Index actual_panel_rows = (2*LhsProgress) * std::max<Index>(1,( (l1 - sizeof(ResScalar)*mr*nr - depth*nr*sizeof(RhsScalar)) / (depth * sizeof(LhsScalar) * 2*LhsProgress) ));
 
-      for(Index i=0; i<peeled_mc; i+=mr)
+      for(Index i1=peeled_mc3; i1<peeled_mc2; i1+=actual_panel_rows)
       {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*mr];
-        prefetch(&blA[0]);
-
-        // TODO move the res loads to the stores
-
-        // get res block as registers
-        AccPacket C0, C4;
-        traits.initAcc(C0);
-        traits.initAcc(C4);
-
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<depth; k++)
+        Index actual_panel_end = (std::min)(i1+actual_panel_rows, peeled_mc2);
+        for(Index j2=0; j2<packet_cols4; j2+=nr)
         {
+          for(Index i=i1; i<actual_panel_end; i+=2*LhsProgress)
+          {
+          
+          // We selected a 2*Traits::LhsProgress x nr micro block of res which is entirely
+          // stored into 2 x nr registers.
+          
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(2*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C1, C2, C3,
+                    C4, C5, C6, C7;
+          traits.initAcc(C0); traits.initAcc(C1); traits.initAcc(C2); traits.initAcc(C3);
+          traits.initAcc(C4); traits.initAcc(C5); traits.initAcc(C6); traits.initAcc(C7);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2 + 0);
+          LinearMapper r1 = res.getLinearMapper(i, j2 + 1);
+          LinearMapper r2 = res.getLinearMapper(i, j2 + 2);
+          LinearMapper r3 = res.getLinearMapper(i, j2 + 3);
+
+          r0.prefetch(prefetch_res_offset);
+          r1.prefetch(prefetch_res_offset);
+          r2.prefetch(prefetch_res_offset);
+          r3.prefetch(prefetch_res_offset);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+          prefetch(&blB[0]);
           LhsPacket A0, A1;
-          RhsPacket B_0;
-          RhsPacket T0;
 
-          traits.loadLhs(&blA[0*LhsProgress], A0);
-          traits.loadLhs(&blA[1*LhsProgress], A1);
-          traits.loadRhs(&blB[0*RhsProgress], B_0);
-          traits.madd(A0,B_0,C0,T0);
-          traits.madd(A1,B_0,C4,B_0);
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 2pX4");
+            RhsPacket B_0, B1, B2, B3, T0;
 
-          blB += RhsProgress;
-          blA += 2*LhsProgress;
+          // NOTE: the begin/end asm comments below work around bug 935!
+          // but they are not enough for gcc>=6 without FMA (bug 1637)
+          #if EIGEN_GNUC_AT_LEAST(6,0) && defined(EIGEN_VECTORIZE_SSE)
+            #define EIGEN_GEBP_2PX4_SPILLING_WORKAROUND __asm__  ("" : [a0] "+x,m" (A0),[a1] "+x,m" (A1));
+          #else
+            #define EIGEN_GEBP_2PX4_SPILLING_WORKAROUND
+          #endif
+          #define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 2pX4");        \
+              traits.loadLhs(&blA[(0+2*K)*LhsProgress], A0);                    \
+              traits.loadLhs(&blA[(1+2*K)*LhsProgress], A1);                    \
+              traits.broadcastRhs(&blB[(0+4*K)*RhsProgress], B_0, B1, B2, B3);  \
+              traits.madd(A0, B_0, C0, T0);                                     \
+              traits.madd(A1, B_0, C4, B_0);                                    \
+              traits.madd(A0, B1,  C1, T0);                                     \
+              traits.madd(A1, B1,  C5, B1);                                     \
+              traits.madd(A0, B2,  C2, T0);                                     \
+              traits.madd(A1, B2,  C6, B2);                                     \
+              traits.madd(A0, B3,  C3, T0);                                     \
+              traits.madd(A1, B3,  C7, B3);                                     \
+              EIGEN_GEBP_2PX4_SPILLING_WORKAROUND                               \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 2pX4");          \
+            } while(false)
+            
+            internal::prefetch(blB+(48+0));
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            internal::prefetch(blB+(48+16));
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*4*RhsProgress;
+            blA += pk*(2*Traits::LhsProgress);
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 2pX4");
+          }
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0, B1, B2, B3, T0;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += 4*RhsProgress;
+            blA += 2*Traits::LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
+
+          ResPacket R0, R1, R2, R3;
+          ResPacket alphav = pset1<ResPacket>(alpha);
+
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r1.loadPacket(0 * Traits::ResPacketSize);
+          R3 = r1.loadPacket(1 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          traits.acc(C4, alphav, R1);
+          traits.acc(C1, alphav, R2);
+          traits.acc(C5, alphav, R3);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          r1.storePacket(0 * Traits::ResPacketSize, R2);
+          r1.storePacket(1 * Traits::ResPacketSize, R3);
+
+          R0 = r2.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r2.loadPacket(1 * Traits::ResPacketSize);
+          R2 = r3.loadPacket(0 * Traits::ResPacketSize);
+          R3 = r3.loadPacket(1 * Traits::ResPacketSize);
+          traits.acc(C2,  alphav, R0);
+          traits.acc(C6,  alphav, R1);
+          traits.acc(C3,  alphav, R2);
+          traits.acc(C7,  alphav, R3);
+          r2.storePacket(0 * Traits::ResPacketSize, R0);
+          r2.storePacket(1 * Traits::ResPacketSize, R1);
+          r3.storePacket(0 * Traits::ResPacketSize, R2);
+          r3.storePacket(1 * Traits::ResPacketSize, R3);
+          }
         }
-        ResPacket R0, R4;
-        ResPacket alphav = pset1<ResPacket>(alpha);
-
-        ResScalar* r0 = &res[(j2+0)*resStride + i];
-
-        R0 = ploadu<ResPacket>(r0);
-        R4 = ploadu<ResPacket>(r0+ResPacketSize);
-
-        traits.acc(C0, alphav, R0);
-        traits.acc(C4, alphav, R4);
-
-        pstoreu(r0,               R0);
-        pstoreu(r0+ResPacketSize, R4);
-      }
-      if(rows-peeled_mc>=LhsProgress)
-      {
-        Index i = peeled_mc;
-        const LhsScalar* blA = &blockA[i*strideA+offsetA*LhsProgress];
-        prefetch(&blA[0]);
-
-        AccPacket C0;
-        traits.initAcc(C0);
-
-        const RhsScalar* blB = unpackedB;
-        for(Index k=0; k<depth; k++)
+      
+        // Deal with remaining columns of the rhs
+        for(Index j2=packet_cols4; j2<cols; j2++)
         {
+          for(Index i=i1; i<actual_panel_end; i+=2*LhsProgress)
+          {
+          // One column at a time
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(2*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C4;
+          traits.initAcc(C0);
+          traits.initAcc(C4);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2);
+          r0.prefetch(prefetch_res_offset);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          LhsPacket A0, A1;
+
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 2pX1");
+            RhsPacket B_0, B1;
+        
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                  \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 2pX1");          \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+2*K)*LhsProgress], A0);                      \
+              traits.loadLhs(&blA[(1+2*K)*LhsProgress], A1);                      \
+              traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);                       \
+              traits.madd(A0, B_0, C0, B1);                                       \
+              traits.madd(A1, B_0, C4, B_0);                                      \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 2pX1");            \
+            } while(false)
+        
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*RhsProgress;
+            blA += pk*2*Traits::LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 2pX1");
+          }
+
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0, B1;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += RhsProgress;
+            blA += 2*Traits::LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
+          ResPacket R0, R1;
+          ResPacket alphav = pset1<ResPacket>(alpha);
+
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r0.loadPacket(1 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          traits.acc(C4, alphav, R1);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r0.storePacket(1 * Traits::ResPacketSize, R1);
+          }
+        }
+      }
+    }
+    //---------- Process 1 * LhsProgress rows at once ----------
+    if(mr>=1*Traits::LhsProgress)
+    {
+      // loops on each largest micro horizontal panel of lhs (1*LhsProgress x depth)
+      for(Index i=peeled_mc2; i<peeled_mc1; i+=1*LhsProgress)
+      {
+        // loops on each largest micro vertical panel of rhs (depth * nr)
+        for(Index j2=0; j2<packet_cols4; j2+=nr)
+        {
+          // We select a 1*Traits::LhsProgress x nr micro block of res which is entirely
+          // stored into 1 x nr registers.
+          
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(1*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0, C1, C2, C3;
+          traits.initAcc(C0);
+          traits.initAcc(C1);
+          traits.initAcc(C2);
+          traits.initAcc(C3);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2 + 0);
+          LinearMapper r1 = res.getLinearMapper(i, j2 + 1);
+          LinearMapper r2 = res.getLinearMapper(i, j2 + 2);
+          LinearMapper r3 = res.getLinearMapper(i, j2 + 3);
+
+          r0.prefetch(prefetch_res_offset);
+          r1.prefetch(prefetch_res_offset);
+          r2.prefetch(prefetch_res_offset);
+          r3.prefetch(prefetch_res_offset);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+          prefetch(&blB[0]);
           LhsPacket A0;
-          RhsPacket B_0;
-          traits.loadLhs(blA, A0);
-          traits.loadRhs(blB, B_0);
-          traits.madd(A0, B_0, C0, B_0);
-          blB += RhsProgress;
-          blA += LhsProgress;
+
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 1pX4");
+            RhsPacket B_0, B1, B2, B3;
+               
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1pX4");        \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+1*K)*LhsProgress], A0);                    \
+              traits.broadcastRhs(&blB[(0+4*K)*RhsProgress], B_0, B1, B2, B3);  \
+              traits.madd(A0, B_0, C0, B_0);                                    \
+              traits.madd(A0, B1,  C1, B1);                                     \
+              traits.madd(A0, B2,  C2, B2);                                     \
+              traits.madd(A0, B3,  C3, B3);                                     \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 1pX4");          \
+            } while(false)
+            
+            internal::prefetch(blB+(48+0));
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            internal::prefetch(blB+(48+16));
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*4*RhsProgress;
+            blA += pk*1*LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 1pX4");
+          }
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0, B1, B2, B3;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += 4*RhsProgress;
+            blA += 1*LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
+
+          ResPacket R0, R1;
+          ResPacket alphav = pset1<ResPacket>(alpha);
+
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r1.loadPacket(0 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          traits.acc(C1,  alphav, R1);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
+          r1.storePacket(0 * Traits::ResPacketSize, R1);
+
+          R0 = r2.loadPacket(0 * Traits::ResPacketSize);
+          R1 = r3.loadPacket(0 * Traits::ResPacketSize);
+          traits.acc(C2,  alphav, R0);
+          traits.acc(C3,  alphav, R1);
+          r2.storePacket(0 * Traits::ResPacketSize, R0);
+          r3.storePacket(0 * Traits::ResPacketSize, R1);
         }
 
-        ResPacket alphav = pset1<ResPacket>(alpha);
-        ResPacket R0 = ploadu<ResPacket>(&res[(j2+0)*resStride + i]);
-        traits.acc(C0, alphav, R0);
-        pstoreu(&res[(j2+0)*resStride + i], R0);
-      }
-      for(Index i=peeled_mc2; i<rows; i++)
-      {
-        const LhsScalar* blA = &blockA[i*strideA+offsetA];
-        prefetch(&blA[0]);
-
-        // gets a 1 x 1 res block as registers
-        ResScalar C0(0);
-        // FIXME directly use blockB ??
-        const RhsScalar* blB = &blockB[j2*strideB+offsetB];
-        for(Index k=0; k<depth; k++)
+        // Deal with remaining columns of the rhs
+        for(Index j2=packet_cols4; j2<cols; j2++)
         {
-          LhsScalar A0 = blA[k];
-          RhsScalar B_0 = blB[k];
-          MADD(cj, A0, B_0, C0, B_0);
+          // One column at a time
+          const LhsScalar* blA = &blockA[i*strideA+offsetA*(1*Traits::LhsProgress)];
+          prefetch(&blA[0]);
+
+          // gets res block as register
+          AccPacket C0;
+          traits.initAcc(C0);
+
+          LinearMapper r0 = res.getLinearMapper(i, j2);
+
+          // performs "inner" products
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          LhsPacket A0;
+
+          for(Index k=0; k<peeled_kc; k+=pk)
+          {
+            EIGEN_ASM_COMMENT("begin gebp micro kernel 1pX1");
+            RhsPacket B_0;
+        
+#define EIGEN_GEBGP_ONESTEP(K) \
+            do {                                                                \
+              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1pX1");        \
+              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
+              traits.loadLhs(&blA[(0+1*K)*LhsProgress], A0);                    \
+              traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);                     \
+              traits.madd(A0, B_0, C0, B_0);                                    \
+              EIGEN_ASM_COMMENT("end step of gebp micro kernel 1pX1");          \
+            } while(false);
+
+            EIGEN_GEBGP_ONESTEP(0);
+            EIGEN_GEBGP_ONESTEP(1);
+            EIGEN_GEBGP_ONESTEP(2);
+            EIGEN_GEBGP_ONESTEP(3);
+            EIGEN_GEBGP_ONESTEP(4);
+            EIGEN_GEBGP_ONESTEP(5);
+            EIGEN_GEBGP_ONESTEP(6);
+            EIGEN_GEBGP_ONESTEP(7);
+
+            blB += pk*RhsProgress;
+            blA += pk*1*Traits::LhsProgress;
+
+            EIGEN_ASM_COMMENT("end gebp micro kernel 1pX1");
+          }
+
+          // process remaining peeled loop
+          for(Index k=peeled_kc; k<depth; k++)
+          {
+            RhsPacket B_0;
+            EIGEN_GEBGP_ONESTEP(0);
+            blB += RhsProgress;
+            blA += 1*Traits::LhsProgress;
+          }
+#undef EIGEN_GEBGP_ONESTEP
+          ResPacket R0;
+          ResPacket alphav = pset1<ResPacket>(alpha);
+          R0 = r0.loadPacket(0 * Traits::ResPacketSize);
+          traits.acc(C0, alphav, R0);
+          r0.storePacket(0 * Traits::ResPacketSize, R0);
         }
-        res[(j2+0)*resStride + i] += alpha*C0;
+      }
+    }
+    //---------- Process remaining rows, 1 at once ----------
+    if(peeled_mc1<rows)
+    {
+      // loop on each panel of the rhs
+      for(Index j2=0; j2<packet_cols4; j2+=nr)
+      {
+        // loop on each row of the lhs (1*LhsProgress x depth)
+        for(Index i=peeled_mc1; i<rows; i+=1)
+        {
+          const LhsScalar* blA = &blockA[i*strideA+offsetA];
+          prefetch(&blA[0]);
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB*nr];
+
+          // The following piece of code wont work for 512 bit registers
+          // Moreover, if LhsProgress==8 it assumes that there is a half packet of the same size
+          // as nr (which is currently 4) for the return type.
+          const int SResPacketHalfSize = unpacket_traits<typename unpacket_traits<SResPacket>::half>::size;
+          if ((SwappedTraits::LhsProgress % 4) == 0 &&
+              (SwappedTraits::LhsProgress <= 8) &&
+              (SwappedTraits::LhsProgress!=8 || SResPacketHalfSize==nr))
+          {
+            SAccPacket C0, C1, C2, C3;
+            straits.initAcc(C0);
+            straits.initAcc(C1);
+            straits.initAcc(C2);
+            straits.initAcc(C3);
+
+            const Index spk   = (std::max)(1,SwappedTraits::LhsProgress/4);
+            const Index endk  = (depth/spk)*spk;
+            const Index endk4 = (depth/(spk*4))*(spk*4);
+
+            Index k=0;
+            for(; k<endk4; k+=4*spk)
+            {
+              SLhsPacket A0,A1;
+              SRhsPacket B_0,B_1;
+
+              straits.loadLhsUnaligned(blB+0*SwappedTraits::LhsProgress, A0);
+              straits.loadLhsUnaligned(blB+1*SwappedTraits::LhsProgress, A1);
+
+              straits.loadRhsQuad(blA+0*spk, B_0);
+              straits.loadRhsQuad(blA+1*spk, B_1);
+              straits.madd(A0,B_0,C0,B_0);
+              straits.madd(A1,B_1,C1,B_1);
+
+              straits.loadLhsUnaligned(blB+2*SwappedTraits::LhsProgress, A0);
+              straits.loadLhsUnaligned(blB+3*SwappedTraits::LhsProgress, A1);
+              straits.loadRhsQuad(blA+2*spk, B_0);
+              straits.loadRhsQuad(blA+3*spk, B_1);
+              straits.madd(A0,B_0,C2,B_0);
+              straits.madd(A1,B_1,C3,B_1);
+
+              blB += 4*SwappedTraits::LhsProgress;
+              blA += 4*spk;
+            }
+            C0 = padd(padd(C0,C1),padd(C2,C3));
+            for(; k<endk; k+=spk)
+            {
+              SLhsPacket A0;
+              SRhsPacket B_0;
+
+              straits.loadLhsUnaligned(blB, A0);
+              straits.loadRhsQuad(blA, B_0);
+              straits.madd(A0,B_0,C0,B_0);
+
+              blB += SwappedTraits::LhsProgress;
+              blA += spk;
+            }
+            if(SwappedTraits::LhsProgress==8)
+            {
+              // Special case where we have to first reduce the accumulation register C0
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SResPacket>::half,SResPacket>::type SResPacketHalf;
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SLhsPacket>::half,SLhsPacket>::type SLhsPacketHalf;
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SLhsPacket>::half,SRhsPacket>::type SRhsPacketHalf;
+              typedef typename conditional<SwappedTraits::LhsProgress>=8,typename unpacket_traits<SAccPacket>::half,SAccPacket>::type SAccPacketHalf;
+
+              SResPacketHalf R = res.template gatherPacket<SResPacketHalf>(i, j2);
+              SResPacketHalf alphav = pset1<SResPacketHalf>(alpha);
+
+              if(depth-endk>0)
+              {
+                // We have to handle the last row of the rhs which corresponds to a half-packet
+                SLhsPacketHalf a0;
+                SRhsPacketHalf b0;
+                straits.loadLhsUnaligned(blB, a0);
+                straits.loadRhs(blA, b0);
+                SAccPacketHalf c0 = predux_downto4(C0);
+                straits.madd(a0,b0,c0,b0);
+                straits.acc(c0, alphav, R);
+              }
+              else
+              {
+                straits.acc(predux_downto4(C0), alphav, R);
+              }
+              res.scatterPacket(i, j2, R);
+            }
+            else
+            {
+              SResPacket R = res.template gatherPacket<SResPacket>(i, j2);
+              SResPacket alphav = pset1<SResPacket>(alpha);
+              straits.acc(C0, alphav, R);
+              res.scatterPacket(i, j2, R);
+            }
+          }
+          else // scalar path
+          {
+            // get a 1 x 4 res block as registers
+            ResScalar C0(0), C1(0), C2(0), C3(0);
+
+            for(Index k=0; k<depth; k++)
+            {
+              LhsScalar A0;
+              RhsScalar B_0, B_1;
+
+              A0 = blA[k];
+
+              B_0 = blB[0];
+              B_1 = blB[1];
+              CJMADD(cj,A0,B_0,C0,  B_0);
+              CJMADD(cj,A0,B_1,C1,  B_1);
+              
+              B_0 = blB[2];
+              B_1 = blB[3];
+              CJMADD(cj,A0,B_0,C2,  B_0);
+              CJMADD(cj,A0,B_1,C3,  B_1);
+              
+              blB += 4;
+            }
+            res(i, j2 + 0) += alpha * C0;
+            res(i, j2 + 1) += alpha * C1;
+            res(i, j2 + 2) += alpha * C2;
+            res(i, j2 + 3) += alpha * C3;
+          }
+        }
+      }
+      // remaining columns
+      for(Index j2=packet_cols4; j2<cols; j2++)
+      {
+        // loop on each row of the lhs (1*LhsProgress x depth)
+        for(Index i=peeled_mc1; i<rows; i+=1)
+        {
+          const LhsScalar* blA = &blockA[i*strideA+offsetA];
+          prefetch(&blA[0]);
+          // gets a 1 x 1 res block as registers
+          ResScalar C0(0);
+          const RhsScalar* blB = &blockB[j2*strideB+offsetB];
+          for(Index k=0; k<depth; k++)
+          {
+            LhsScalar A0 = blA[k];
+            RhsScalar B_0 = blB[k];
+            CJMADD(cj, A0, B_0, C0, B_0);
+          }
+          res(i, j2) += alpha * C0;
+        }
       }
     }
   }
@@ -1114,81 +1692,193 @@
 //
 //  32 33 34 35 ...
 //  36 36 38 39 ...
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder, bool Conjugate, bool PanelMode>
-struct gemm_pack_lhs
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+struct gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, ColMajor, Conjugate, PanelMode>
 {
-  EIGEN_DONT_INLINE void operator()(Scalar* blockA, const Scalar* EIGEN_RESTRICT _lhs, Index lhsStride, Index depth, Index rows, Index stride=0, Index offset=0);
+  typedef typename DataMapper::LinearMapper LinearMapper;
+  EIGEN_DONT_INLINE void operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride=0, Index offset=0);
 };
 
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder, bool Conjugate, bool PanelMode>
-EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, Pack1, Pack2, StorageOrder, Conjugate, PanelMode>
-  ::operator()(Scalar* blockA, const Scalar* EIGEN_RESTRICT _lhs, Index lhsStride, Index depth, Index rows, Index stride, Index offset)
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, ColMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride, Index offset)
 {
   typedef typename packet_traits<Scalar>::type Packet;
   enum { PacketSize = packet_traits<Scalar>::size };
 
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK LHS");
-  EIGEN_UNUSED_VARIABLE(stride)
-  EIGEN_UNUSED_VARIABLE(offset)
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
-  eigen_assert( (StorageOrder==RowMajor) || ((Pack1%PacketSize)==0 && Pack1<=4*PacketSize) );
+  eigen_assert( ((Pack1%PacketSize)==0 && Pack1<=4*PacketSize) || (Pack1<=4) );
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
-  const_blas_data_mapper<Scalar, Index, StorageOrder> lhs(_lhs,lhsStride);
   Index count = 0;
-  Index peeled_mc = (rows/Pack1)*Pack1;
-  for(Index i=0; i<peeled_mc; i+=Pack1)
-  {
-    if(PanelMode) count += Pack1 * offset;
 
-    if(StorageOrder==ColMajor)
+  const Index peeled_mc3 = Pack1>=3*PacketSize ? (rows/(3*PacketSize))*(3*PacketSize) : 0;
+  const Index peeled_mc2 = Pack1>=2*PacketSize ? peeled_mc3+((rows-peeled_mc3)/(2*PacketSize))*(2*PacketSize) : 0;
+  const Index peeled_mc1 = Pack1>=1*PacketSize ? (rows/(1*PacketSize))*(1*PacketSize) : 0;
+  const Index peeled_mc0 = Pack2>=1*PacketSize ? peeled_mc1
+                         : Pack2>1             ? (rows/Pack2)*Pack2 : 0;
+
+  Index i=0;
+
+  // Pack 3 packets
+  if(Pack1>=3*PacketSize)
+  {
+    for(; i<peeled_mc3; i+=3*PacketSize)
     {
+      if(PanelMode) count += (3*PacketSize) * offset;
+
       for(Index k=0; k<depth; k++)
       {
-        Packet A, B, C, D;
-        if(Pack1>=1*PacketSize) A = ploadu<Packet>(&lhs(i+0*PacketSize, k));
-        if(Pack1>=2*PacketSize) B = ploadu<Packet>(&lhs(i+1*PacketSize, k));
-        if(Pack1>=3*PacketSize) C = ploadu<Packet>(&lhs(i+2*PacketSize, k));
-        if(Pack1>=4*PacketSize) D = ploadu<Packet>(&lhs(i+3*PacketSize, k));
-        if(Pack1>=1*PacketSize) { pstore(blockA+count, cj.pconj(A)); count+=PacketSize; }
-        if(Pack1>=2*PacketSize) { pstore(blockA+count, cj.pconj(B)); count+=PacketSize; }
-        if(Pack1>=3*PacketSize) { pstore(blockA+count, cj.pconj(C)); count+=PacketSize; }
-        if(Pack1>=4*PacketSize) { pstore(blockA+count, cj.pconj(D)); count+=PacketSize; }
+        Packet A, B, C;
+        A = lhs.loadPacket(i+0*PacketSize, k);
+        B = lhs.loadPacket(i+1*PacketSize, k);
+        C = lhs.loadPacket(i+2*PacketSize, k);
+        pstore(blockA+count, cj.pconj(A)); count+=PacketSize;
+        pstore(blockA+count, cj.pconj(B)); count+=PacketSize;
+        pstore(blockA+count, cj.pconj(C)); count+=PacketSize;
       }
+      if(PanelMode) count += (3*PacketSize) * (stride-offset-depth);
     }
-    else
+  }
+  // Pack 2 packets
+  if(Pack1>=2*PacketSize)
+  {
+    for(; i<peeled_mc2; i+=2*PacketSize)
     {
+      if(PanelMode) count += (2*PacketSize) * offset;
+
       for(Index k=0; k<depth; k++)
       {
-        // TODO add a vectorized transpose here
+        Packet A, B;
+        A = lhs.loadPacket(i+0*PacketSize, k);
+        B = lhs.loadPacket(i+1*PacketSize, k);
+        pstore(blockA+count, cj.pconj(A)); count+=PacketSize;
+        pstore(blockA+count, cj.pconj(B)); count+=PacketSize;
+      }
+      if(PanelMode) count += (2*PacketSize) * (stride-offset-depth);
+    }
+  }
+  // Pack 1 packets
+  if(Pack1>=1*PacketSize)
+  {
+    for(; i<peeled_mc1; i+=1*PacketSize)
+    {
+      if(PanelMode) count += (1*PacketSize) * offset;
+
+      for(Index k=0; k<depth; k++)
+      {
+        Packet A;
+        A = lhs.loadPacket(i+0*PacketSize, k);
+        pstore(blockA+count, cj.pconj(A));
+        count+=PacketSize;
+      }
+      if(PanelMode) count += (1*PacketSize) * (stride-offset-depth);
+    }
+  }
+  // Pack scalars
+  if(Pack2<PacketSize && Pack2>1)
+  {
+    for(; i<peeled_mc0; i+=Pack2)
+    {
+      if(PanelMode) count += Pack2 * offset;
+
+      for(Index k=0; k<depth; k++)
+        for(Index w=0; w<Pack2; w++)
+          blockA[count++] = cj(lhs(i+w, k));
+
+      if(PanelMode) count += Pack2 * (stride-offset-depth);
+    }
+  }
+  for(; i<rows; i++)
+  {
+    if(PanelMode) count += offset;
+    for(Index k=0; k<depth; k++)
+      blockA[count++] = cj(lhs(i, k));
+    if(PanelMode) count += (stride-offset-depth);
+  }
+}
+
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+struct gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, RowMajor, Conjugate, PanelMode>
+{
+  typedef typename DataMapper::LinearMapper LinearMapper;
+  EIGEN_DONT_INLINE void operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride=0, Index offset=0);
+};
+
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_lhs<Scalar, Index, DataMapper, Pack1, Pack2, RowMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockA, const DataMapper& lhs, Index depth, Index rows, Index stride, Index offset)
+{
+  typedef typename packet_traits<Scalar>::type Packet;
+  enum { PacketSize = packet_traits<Scalar>::size };
+
+  EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK LHS");
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
+  eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
+  conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
+  Index count = 0;
+
+//   const Index peeled_mc3 = Pack1>=3*PacketSize ? (rows/(3*PacketSize))*(3*PacketSize) : 0;
+//   const Index peeled_mc2 = Pack1>=2*PacketSize ? peeled_mc3+((rows-peeled_mc3)/(2*PacketSize))*(2*PacketSize) : 0;
+//   const Index peeled_mc1 = Pack1>=1*PacketSize ? (rows/(1*PacketSize))*(1*PacketSize) : 0;
+
+  int pack = Pack1;
+  Index i = 0;
+  while(pack>0)
+  {
+    Index remaining_rows = rows-i;
+    Index peeled_mc = i+(remaining_rows/pack)*pack;
+    for(; i<peeled_mc; i+=pack)
+    {
+      if(PanelMode) count += pack * offset;
+
+      const Index peeled_k = (depth/PacketSize)*PacketSize;
+      Index k=0;
+      if(pack>=PacketSize)
+      {
+        for(; k<peeled_k; k+=PacketSize)
+        {
+          for (Index m = 0; m < pack; m += PacketSize)
+          {
+            PacketBlock<Packet> kernel;
+            for (int p = 0; p < PacketSize; ++p) kernel.packet[p] = lhs.loadPacket(i+p+m, k);
+            ptranspose(kernel);
+            for (int p = 0; p < PacketSize; ++p) pstore(blockA+count+m+(pack)*p, cj.pconj(kernel.packet[p]));
+          }
+          count += PacketSize*pack;
+        }
+      }
+      for(; k<depth; k++)
+      {
         Index w=0;
-        for(; w<Pack1-3; w+=4)
+        for(; w<pack-3; w+=4)
         {
           Scalar a(cj(lhs(i+w+0, k))),
-                  b(cj(lhs(i+w+1, k))),
-                  c(cj(lhs(i+w+2, k))),
-                  d(cj(lhs(i+w+3, k)));
+                 b(cj(lhs(i+w+1, k))),
+                 c(cj(lhs(i+w+2, k))),
+                 d(cj(lhs(i+w+3, k)));
           blockA[count++] = a;
           blockA[count++] = b;
           blockA[count++] = c;
           blockA[count++] = d;
         }
-        if(Pack1%4)
-          for(;w<Pack1;++w)
+        if(pack%4)
+          for(;w<pack;++w)
             blockA[count++] = cj(lhs(i+w, k));
       }
+
+      if(PanelMode) count += pack * (stride-offset-depth);
     }
-    if(PanelMode) count += Pack1 * (stride-offset-depth);
+
+    pack -= PacketSize;
+    if(pack<Pack2 && (pack+PacketSize)!=Pack2)
+      pack = Pack2;
   }
-  if(rows-peeled_mc>=Pack2)
-  {
-    if(PanelMode) count += Pack2*offset;
-    for(Index k=0; k<depth; k++)
-      for(Index w=0; w<Pack2; w++)
-        blockA[count++] = cj(lhs(peeled_mc+w, k));
-    if(PanelMode) count += Pack2 * (stride-offset-depth);
-    peeled_mc += Pack2;
-  }
-  for(Index i=peeled_mc; i<rows; i++)
+
+  for(; i<rows; i++)
   {
     if(PanelMode) count += offset;
     for(Index k=0; k<depth; k++)
@@ -1204,53 +1894,123 @@
 //  4  5  6  7   16 17 18 19   25 28
 //  8  9 10 11   20 21 22 23   26 29
 //  .  .  .  .    .  .  .  .    .  .
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-struct gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, PanelMode>
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+struct gemm_pack_rhs<Scalar, Index, DataMapper, nr, ColMajor, Conjugate, PanelMode>
 {
   typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename DataMapper::LinearMapper LinearMapper;
   enum { PacketSize = packet_traits<Scalar>::size };
-  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride=0, Index offset=0);
+  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride=0, Index offset=0);
 };
 
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, ColMajor, Conjugate, PanelMode>
-  ::operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride, Index offset)
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, DataMapper, nr, ColMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride, Index offset)
 {
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK RHS COLMAJOR");
-  EIGEN_UNUSED_VARIABLE(stride)
-  EIGEN_UNUSED_VARIABLE(offset)
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
-  Index packet_cols = (cols/nr) * nr;
+  Index packet_cols8 = nr>=8 ? (cols/8) * 8 : 0;
+  Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
   Index count = 0;
-  for(Index j2=0; j2<packet_cols; j2+=nr)
+  const Index peeled_k = (depth/PacketSize)*PacketSize;
+//   if(nr>=8)
+//   {
+//     for(Index j2=0; j2<packet_cols8; j2+=8)
+//     {
+//       // skip what we have before
+//       if(PanelMode) count += 8 * offset;
+//       const Scalar* b0 = &rhs[(j2+0)*rhsStride];
+//       const Scalar* b1 = &rhs[(j2+1)*rhsStride];
+//       const Scalar* b2 = &rhs[(j2+2)*rhsStride];
+//       const Scalar* b3 = &rhs[(j2+3)*rhsStride];
+//       const Scalar* b4 = &rhs[(j2+4)*rhsStride];
+//       const Scalar* b5 = &rhs[(j2+5)*rhsStride];
+//       const Scalar* b6 = &rhs[(j2+6)*rhsStride];
+//       const Scalar* b7 = &rhs[(j2+7)*rhsStride];
+//       Index k=0;
+//       if(PacketSize==8) // TODO enbale vectorized transposition for PacketSize==4
+//       {
+//         for(; k<peeled_k; k+=PacketSize) {
+//           PacketBlock<Packet> kernel;
+//           for (int p = 0; p < PacketSize; ++p) {
+//             kernel.packet[p] = ploadu<Packet>(&rhs[(j2+p)*rhsStride+k]);
+//           }
+//           ptranspose(kernel);
+//           for (int p = 0; p < PacketSize; ++p) {
+//             pstoreu(blockB+count, cj.pconj(kernel.packet[p]));
+//             count+=PacketSize;
+//           }
+//         }
+//       }
+//       for(; k<depth; k++)
+//       {
+//         blockB[count+0] = cj(b0[k]);
+//         blockB[count+1] = cj(b1[k]);
+//         blockB[count+2] = cj(b2[k]);
+//         blockB[count+3] = cj(b3[k]);
+//         blockB[count+4] = cj(b4[k]);
+//         blockB[count+5] = cj(b5[k]);
+//         blockB[count+6] = cj(b6[k]);
+//         blockB[count+7] = cj(b7[k]);
+//         count += 8;
+//       }
+//       // skip what we have after
+//       if(PanelMode) count += 8 * (stride-offset-depth);
+//     }
+//   }
+
+  if(nr>=4)
   {
-    // skip what we have before
-    if(PanelMode) count += nr * offset;
-    const Scalar* b0 = &rhs[(j2+0)*rhsStride];
-    const Scalar* b1 = &rhs[(j2+1)*rhsStride];
-    const Scalar* b2 = &rhs[(j2+2)*rhsStride];
-    const Scalar* b3 = &rhs[(j2+3)*rhsStride];
-    for(Index k=0; k<depth; k++)
+    for(Index j2=packet_cols8; j2<packet_cols4; j2+=4)
     {
-                blockB[count+0] = cj(b0[k]);
-                blockB[count+1] = cj(b1[k]);
-      if(nr==4) blockB[count+2] = cj(b2[k]);
-      if(nr==4) blockB[count+3] = cj(b3[k]);
-      count += nr;
+      // skip what we have before
+      if(PanelMode) count += 4 * offset;
+      const LinearMapper dm0 = rhs.getLinearMapper(0, j2 + 0);
+      const LinearMapper dm1 = rhs.getLinearMapper(0, j2 + 1);
+      const LinearMapper dm2 = rhs.getLinearMapper(0, j2 + 2);
+      const LinearMapper dm3 = rhs.getLinearMapper(0, j2 + 3);
+
+      Index k=0;
+      if((PacketSize%4)==0) // TODO enable vectorized transposition for PacketSize==2 ??
+      {
+        for(; k<peeled_k; k+=PacketSize) {
+          PacketBlock<Packet,(PacketSize%4)==0?4:PacketSize> kernel;
+          kernel.packet[0] = dm0.loadPacket(k);
+          kernel.packet[1%PacketSize] = dm1.loadPacket(k);
+          kernel.packet[2%PacketSize] = dm2.loadPacket(k);
+          kernel.packet[3%PacketSize] = dm3.loadPacket(k);
+          ptranspose(kernel);
+          pstoreu(blockB+count+0*PacketSize, cj.pconj(kernel.packet[0]));
+          pstoreu(blockB+count+1*PacketSize, cj.pconj(kernel.packet[1%PacketSize]));
+          pstoreu(blockB+count+2*PacketSize, cj.pconj(kernel.packet[2%PacketSize]));
+          pstoreu(blockB+count+3*PacketSize, cj.pconj(kernel.packet[3%PacketSize]));
+          count+=4*PacketSize;
+        }
+      }
+      for(; k<depth; k++)
+      {
+        blockB[count+0] = cj(dm0(k));
+        blockB[count+1] = cj(dm1(k));
+        blockB[count+2] = cj(dm2(k));
+        blockB[count+3] = cj(dm3(k));
+        count += 4;
+      }
+      // skip what we have after
+      if(PanelMode) count += 4 * (stride-offset-depth);
     }
-    // skip what we have after
-    if(PanelMode) count += nr * (stride-offset-depth);
   }
 
   // copy the remaining columns one at a time (nr==1)
-  for(Index j2=packet_cols; j2<cols; ++j2)
+  for(Index j2=packet_cols4; j2<cols; ++j2)
   {
     if(PanelMode) count += offset;
-    const Scalar* b0 = &rhs[(j2+0)*rhsStride];
+    const LinearMapper dm0 = rhs.getLinearMapper(0, j2);
     for(Index k=0; k<depth; k++)
     {
-      blockB[count] = cj(b0[k]);
+      blockB[count] = cj(dm0(k));
       count += 1;
     }
     if(PanelMode) count += (stride-offset-depth);
@@ -1258,48 +2018,93 @@
 }
 
 // this version is optimized for row major matrices
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-struct gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, PanelMode>
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+struct gemm_pack_rhs<Scalar, Index, DataMapper, nr, RowMajor, Conjugate, PanelMode>
 {
+  typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename DataMapper::LinearMapper LinearMapper;
   enum { PacketSize = packet_traits<Scalar>::size };
-  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride=0, Index offset=0);
+  EIGEN_DONT_INLINE void operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride=0, Index offset=0);
 };
 
-template<typename Scalar, typename Index, int nr, bool Conjugate, bool PanelMode>
-EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, nr, RowMajor, Conjugate, PanelMode>
-  ::operator()(Scalar* blockB, const Scalar* rhs, Index rhsStride, Index depth, Index cols, Index stride, Index offset)
+template<typename Scalar, typename Index, typename DataMapper, int nr, bool Conjugate, bool PanelMode>
+EIGEN_DONT_INLINE void gemm_pack_rhs<Scalar, Index, DataMapper, nr, RowMajor, Conjugate, PanelMode>
+  ::operator()(Scalar* blockB, const DataMapper& rhs, Index depth, Index cols, Index stride, Index offset)
 {
   EIGEN_ASM_COMMENT("EIGEN PRODUCT PACK RHS ROWMAJOR");
-  EIGEN_UNUSED_VARIABLE(stride)
-  EIGEN_UNUSED_VARIABLE(offset)
+  EIGEN_UNUSED_VARIABLE(stride);
+  EIGEN_UNUSED_VARIABLE(offset);
   eigen_assert(((!PanelMode) && stride==0 && offset==0) || (PanelMode && stride>=depth && offset<=stride));
   conj_if<NumTraits<Scalar>::IsComplex && Conjugate> cj;
-  Index packet_cols = (cols/nr) * nr;
+  Index packet_cols8 = nr>=8 ? (cols/8) * 8 : 0;
+  Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
   Index count = 0;
-  for(Index j2=0; j2<packet_cols; j2+=nr)
+
+//   if(nr>=8)
+//   {
+//     for(Index j2=0; j2<packet_cols8; j2+=8)
+//     {
+//       // skip what we have before
+//       if(PanelMode) count += 8 * offset;
+//       for(Index k=0; k<depth; k++)
+//       {
+//         if (PacketSize==8) {
+//           Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
+//           pstoreu(blockB+count, cj.pconj(A));
+//         } else if (PacketSize==4) {
+//           Packet A = ploadu<Packet>(&rhs[k*rhsStride + j2]);
+//           Packet B = ploadu<Packet>(&rhs[k*rhsStride + j2 + PacketSize]);
+//           pstoreu(blockB+count, cj.pconj(A));
+//           pstoreu(blockB+count+PacketSize, cj.pconj(B));
+//         } else {
+//           const Scalar* b0 = &rhs[k*rhsStride + j2];
+//           blockB[count+0] = cj(b0[0]);
+//           blockB[count+1] = cj(b0[1]);
+//           blockB[count+2] = cj(b0[2]);
+//           blockB[count+3] = cj(b0[3]);
+//           blockB[count+4] = cj(b0[4]);
+//           blockB[count+5] = cj(b0[5]);
+//           blockB[count+6] = cj(b0[6]);
+//           blockB[count+7] = cj(b0[7]);
+//         }
+//         count += 8;
+//       }
+//       // skip what we have after
+//       if(PanelMode) count += 8 * (stride-offset-depth);
+//     }
+//   }
+  if(nr>=4)
   {
-    // skip what we have before
-    if(PanelMode) count += nr * offset;
-    for(Index k=0; k<depth; k++)
+    for(Index j2=packet_cols8; j2<packet_cols4; j2+=4)
     {
-      const Scalar* b0 = &rhs[k*rhsStride + j2];
-                blockB[count+0] = cj(b0[0]);
-                blockB[count+1] = cj(b0[1]);
-      if(nr==4) blockB[count+2] = cj(b0[2]);
-      if(nr==4) blockB[count+3] = cj(b0[3]);
-      count += nr;
+      // skip what we have before
+      if(PanelMode) count += 4 * offset;
+      for(Index k=0; k<depth; k++)
+      {
+        if (PacketSize==4) {
+          Packet A = rhs.loadPacket(k, j2);
+          pstoreu(blockB+count, cj.pconj(A));
+          count += PacketSize;
+        } else {
+          const LinearMapper dm0 = rhs.getLinearMapper(k, j2);
+          blockB[count+0] = cj(dm0(0));
+          blockB[count+1] = cj(dm0(1));
+          blockB[count+2] = cj(dm0(2));
+          blockB[count+3] = cj(dm0(3));
+          count += 4;
+        }
+      }
+      // skip what we have after
+      if(PanelMode) count += 4 * (stride-offset-depth);
     }
-    // skip what we have after
-    if(PanelMode) count += nr * (stride-offset-depth);
   }
   // copy the remaining columns one at a time (nr==1)
-  for(Index j2=packet_cols; j2<cols; ++j2)
+  for(Index j2=packet_cols4; j2<cols; ++j2)
   {
     if(PanelMode) count += offset;
-    const Scalar* b0 = &rhs[j2];
     for(Index k=0; k<depth; k++)
     {
-      blockB[count] = cj(b0[k*rhsStride]);
+      blockB[count] = cj(rhs(k, j2));
       count += 1;
     }
     if(PanelMode) count += stride-offset-depth;
@@ -1312,8 +2117,8 @@
   * \sa setCpuCacheSize */
 inline std::ptrdiff_t l1CacheSize()
 {
-  std::ptrdiff_t l1, l2;
-  internal::manage_caching_sizes(GetAction, &l1, &l2);
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
   return l1;
 }
 
@@ -1321,19 +2126,29 @@
   * \sa setCpuCacheSize */
 inline std::ptrdiff_t l2CacheSize()
 {
-  std::ptrdiff_t l1, l2;
-  internal::manage_caching_sizes(GetAction, &l1, &l2);
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
   return l2;
 }
 
+/** \returns the currently set level 3 cpu cache size (in bytes) used to estimate the ideal blocking size paramete\
+rs.                                                                                                                
+* \sa setCpuCacheSize */
+inline std::ptrdiff_t l3CacheSize()
+{
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
+  return l3;
+}
+
 /** Set the cpu L1 and L2 cache sizes (in bytes).
   * These values are use to adjust the size of the blocks
   * for the algorithms working per blocks.
   *
   * \sa computeProductBlockingSizes */
-inline void setCpuCacheSizes(std::ptrdiff_t l1, std::ptrdiff_t l2)
+inline void setCpuCacheSizes(std::ptrdiff_t l1, std::ptrdiff_t l2, std::ptrdiff_t l3)
 {
-  internal::manage_caching_sizes(SetAction, &l1, &l2);
+  internal::manage_caching_sizes(SetAction, &l1, &l2, &l3);
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/products/GeneralMatrixMatrix.h b/Eigen/src/Core/products/GeneralMatrixMatrix.h
index 3f5ffcf..6440e1d 100644
--- a/Eigen/src/Core/products/GeneralMatrixMatrix.h
+++ b/Eigen/src/Core/products/GeneralMatrixMatrix.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_GENERAL_MATRIX_MATRIX_H
 #define EIGEN_GENERAL_MATRIX_MATRIX_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -23,7 +23,9 @@
   typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs>
 struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef gebp_traits<RhsScalar,LhsScalar> Traits;
+
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   static EIGEN_STRONG_INLINE void run(
     Index rows, Index cols, Index depth,
     const LhsScalar* lhs, Index lhsStride,
@@ -51,42 +53,44 @@
 struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor>
 {
 
-typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+typedef gebp_traits<LhsScalar,RhsScalar> Traits;
+
+typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 static void run(Index rows, Index cols, Index depth,
   const LhsScalar* _lhs, Index lhsStride,
   const RhsScalar* _rhs, Index rhsStride,
-  ResScalar* res, Index resStride,
+  ResScalar* _res, Index resStride,
   ResScalar alpha,
   level3_blocking<LhsScalar,RhsScalar>& blocking,
   GemmParallelInfo<Index>* info = 0)
 {
-  const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-  const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
-
-  typedef gebp_traits<LhsScalar,RhsScalar> Traits;
+  typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper;
+  typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper;
+  typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+  LhsMapper lhs(_lhs,lhsStride);
+  RhsMapper rhs(_rhs,rhsStride);
+  ResMapper res(_res, resStride);
 
   Index kc = blocking.kc();                   // cache block size along the K direction
   Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
-  //Index nc = blocking.nc(); // cache block size along the N direction
+  Index nc = (std::min)(cols,blocking.nc());  // cache block size along the N direction
 
-  gemm_pack_lhs<LhsScalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-  gemm_pack_rhs<RhsScalar, Index, Traits::nr, RhsStorageOrder> pack_rhs;
-  gebp_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
+  gemm_pack_lhs<LhsScalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+  gemm_pack_rhs<RhsScalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
+  gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
 
 #ifdef EIGEN_HAS_OPENMP
   if(info)
   {
     // this is the parallel version!
-    Index tid = omp_get_thread_num();
-    Index threads = omp_get_num_threads();
-    
-    std::size_t sizeA = kc*mc;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, 0);
-    ei_declare_aligned_stack_constructed_variable(RhsScalar, w, sizeW, 0);
-    
-    RhsScalar* blockB = blocking.blockB();
-    eigen_internal_assert(blockB!=0);
+    int tid = omp_get_thread_num();
+    int threads = omp_get_num_threads();
+
+    LhsScalar* blockA = blocking.blockA();
+    eigen_internal_assert(blockA!=0);
+
+    std::size_t sizeB = kc*nc;
+    ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, 0);
 
     // For each horizontal panel of the rhs, and corresponding vertical panel of the lhs...
     for(Index k=0; k<depth; k+=kc)
@@ -94,54 +98,56 @@
       const Index actual_kc = (std::min)(k+kc,depth)-k; // => rows of B', and cols of the A'
 
       // In order to reduce the chance that a thread has to wait for the other,
-      // let's start by packing A'.
-      pack_lhs(blockA, &lhs(0,k), lhsStride, actual_kc, mc);
+      // let's start by packing B'.
+      pack_rhs(blockB, rhs.getSubMapper(k,0), actual_kc, nc);
 
-      // Pack B_k to B' in a parallel fashion:
-      // each thread packs the sub block B_k,j to B'_j where j is the thread id.
+      // Pack A_k to A' in a parallel fashion:
+      // each thread packs the sub block A_k,i to A'_i where i is the thread id.
 
-      // However, before copying to B'_j, we have to make sure that no other thread is still using it,
+      // However, before copying to A'_i, we have to make sure that no other thread is still using it,
       // i.e., we test that info[tid].users equals 0.
       // Then, we set info[tid].users to the number of threads to mark that all other threads are going to use it.
       while(info[tid].users!=0) {}
       info[tid].users += threads;
 
-      pack_rhs(blockB+info[tid].rhs_start*actual_kc, &rhs(k,info[tid].rhs_start), rhsStride, actual_kc, info[tid].rhs_length);
+      pack_lhs(blockA+info[tid].lhs_start*actual_kc, lhs.getSubMapper(info[tid].lhs_start,k), actual_kc, info[tid].lhs_length);
 
-      // Notify the other threads that the part B'_j is ready to go.
+      // Notify the other threads that the part A'_i is ready to go.
       info[tid].sync = k;
 
-      // Computes C_i += A' * B' per B'_j
-      for(Index shift=0; shift<threads; ++shift)
+      // Computes C_i += A' * B' per A'_i
+      for(int shift=0; shift<threads; ++shift)
       {
-        Index j = (tid+shift)%threads;
+        int i = (tid+shift)%threads;
 
-        // At this point we have to make sure that B'_j has been updated by the thread j,
+        // At this point we have to make sure that A'_i has been updated by the thread i,
         // we use testAndSetOrdered to mimic a volatile access.
         // However, no need to wait for the B' part which has been updated by the current thread!
-        if(shift>0)
-          while(info[j].sync!=k) {}
+        if (shift>0) {
+          while(info[i].sync!=k) {
+          }
+        }
 
-        gebp(res+info[j].rhs_start*resStride, resStride, blockA, blockB+info[j].rhs_start*actual_kc, mc, actual_kc, info[j].rhs_length, alpha, -1,-1,0,0, w);
+        gebp(res.getSubMapper(info[i].lhs_start, 0), blockA+info[i].lhs_start*actual_kc, blockB, info[i].lhs_length, actual_kc, nc, alpha);
       }
 
-      // Then keep going as usual with the remaining A'
-      for(Index i=mc; i<rows; i+=mc)
+      // Then keep going as usual with the remaining B'
+      for(Index j=nc; j<cols; j+=nc)
       {
-        const Index actual_mc = (std::min)(i+mc,rows)-i;
+        const Index actual_nc = (std::min)(j+nc,cols)-j;
 
-        // pack A_i,k to A'
-        pack_lhs(blockA, &lhs(i,k), lhsStride, actual_kc, actual_mc);
+        // pack B_k,j to B'
+        pack_rhs(blockB, rhs.getSubMapper(k,j), actual_kc, actual_nc);
 
-        // C_i += A' * B'
-        gebp(res+i, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1,-1,0,0, w);
+        // C_j += A' * B'
+        gebp(res.getSubMapper(0, j), blockA, blockB, rows, actual_kc, actual_nc, alpha);
       }
 
-      // Release all the sub blocks B'_j of B' for the current thread,
+      // Release all the sub blocks A'_i of A' for the current thread,
       // i.e., we simply decrement the number of users by 1
-      for(Index j=0; j<threads; ++j)
+      for(Index i=0; i<threads; ++i)
         #pragma omp atomic
-        --(info[j].users);
+        info[i].users -= 1;
     }
   }
   else
@@ -151,38 +157,42 @@
 
     // this is the sequential version!
     std::size_t sizeA = kc*mc;
-    std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
+    std::size_t sizeB = kc*nc;
 
     ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(RhsScalar, blockW, sizeW, blocking.blockW());
+
+    const bool pack_rhs_once = mc!=rows && kc==depth && nc==cols;
 
     // For each horizontal panel of the rhs, and corresponding panel of the lhs...
-    // (==GEMM_VAR1)
-    for(Index k2=0; k2<depth; k2+=kc)
+    for(Index i2=0; i2<rows; i2+=mc)
     {
-      const Index actual_kc = (std::min)(k2+kc,depth)-k2;
+      const Index actual_mc = (std::min)(i2+mc,rows)-i2;
 
-      // OK, here we have selected one horizontal panel of rhs and one vertical panel of lhs.
-      // => Pack rhs's panel into a sequential chunk of memory (L2 caching)
-      // Note that this panel will be read as many times as the number of blocks in the lhs's
-      // vertical panel which is, in practice, a very low number.
-      pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, cols);
-
-      // For each mc x kc block of the lhs's vertical panel...
-      // (==GEPP_VAR1)
-      for(Index i2=0; i2<rows; i2+=mc)
+      for(Index k2=0; k2<depth; k2+=kc)
       {
-        const Index actual_mc = (std::min)(i2+mc,rows)-i2;
+        const Index actual_kc = (std::min)(k2+kc,depth)-k2;
 
-        // We pack the lhs's block into a sequential chunk of memory (L1 caching)
-        // Note that this block will be read a very high number of times, which is equal to the number of
-        // micro vertical panel of the large rhs's panel (e.g., cols/4 times).
-        pack_lhs(blockA, &lhs(i2,k2), lhsStride, actual_kc, actual_mc);
+        // OK, here we have selected one horizontal panel of rhs and one vertical panel of lhs.
+        // => Pack lhs's panel into a sequential chunk of memory (L2/L3 caching)
+        // Note that this panel will be read as many times as the number of blocks in the rhs's
+        // horizontal panel which is, in practice, a very low number.
+        pack_lhs(blockA, lhs.getSubMapper(i2,k2), actual_kc, actual_mc);
 
-        // Everything is packed, we can now call the block * panel kernel:
-        gebp(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1, -1, 0, 0, blockW);
+        // For each kc x nc block of the rhs's horizontal panel...
+        for(Index j2=0; j2<cols; j2+=nc)
+        {
+          const Index actual_nc = (std::min)(j2+nc,cols)-j2;
+
+          // We pack the rhs's block into a sequential chunk of memory (L2 caching)
+          // Note that this block will be read a very high number of times, which is equal to the number of
+          // micro horizontal panel of the large rhs's panel (e.g., rows/12 times).
+          if((!pack_rhs_once) || i2==0)
+            pack_rhs(blockB, rhs.getSubMapper(k2,j2), actual_kc, actual_nc);
+
+          // Everything is packed, we can now call the panel * block kernel:
+          gebp(res.getSubMapper(i2, j2), blockA, blockB, actual_mc, actual_kc, actual_nc, alpha);
+        }
       }
     }
   }
@@ -191,26 +201,21 @@
 };
 
 /*********************************************************************************
-*  Specialization of GeneralProduct<> for "large" GEMM, i.e.,
+*  Specialization of generic_product_impl for "large" GEMM, i.e.,
 *  implementation of the high level wrapper to general_matrix_matrix_product
 **********************************************************************************/
 
-template<typename Lhs, typename Rhs>
-struct traits<GeneralProduct<Lhs,Rhs,GemmProduct> >
- : traits<ProductBase<GeneralProduct<Lhs,Rhs,GemmProduct>, Lhs, Rhs> >
-{};
-
 template<typename Scalar, typename Index, typename Gemm, typename Lhs, typename Rhs, typename Dest, typename BlockingType>
 struct gemm_functor
 {
-  gemm_functor(const Lhs& lhs, const Rhs& rhs, Dest& dest, const Scalar& actualAlpha,
-                  BlockingType& blocking)
+  gemm_functor(const Lhs& lhs, const Rhs& rhs, Dest& dest, const Scalar& actualAlpha, BlockingType& blocking)
     : m_lhs(lhs), m_rhs(rhs), m_dest(dest), m_actualAlpha(actualAlpha), m_blocking(blocking)
   {}
 
-  void initParallelSession() const
+  void initParallelSession(Index num_threads) const
   {
-    m_blocking.allocateB();
+    m_blocking.initParallel(m_lhs.rows(), m_rhs.cols(), m_lhs.cols(), num_threads);
+    m_blocking.allocateA();
   }
 
   void operator() (Index row, Index rows, Index col=0, Index cols=-1, GemmParallelInfo<Index>* info=0) const
@@ -219,12 +224,14 @@
       cols = m_rhs.cols();
 
     Gemm::run(rows, cols, m_lhs.cols(),
-              /*(const Scalar*)*/&m_lhs.coeffRef(row,0), m_lhs.outerStride(),
-              /*(const Scalar*)*/&m_rhs.coeffRef(0,col), m_rhs.outerStride(),
+              &m_lhs.coeffRef(row,0), m_lhs.outerStride(),
+              &m_rhs.coeffRef(0,col), m_rhs.outerStride(),
               (Scalar*)&(m_dest.coeffRef(row,col)), m_dest.outerStride(),
               m_actualAlpha, m_blocking, info);
   }
 
+  typedef typename Gemm::Traits Traits;
+
   protected:
     const Lhs& m_lhs;
     const Rhs& m_rhs;
@@ -245,29 +252,27 @@
   protected:
     LhsScalar* m_blockA;
     RhsScalar* m_blockB;
-    RhsScalar* m_blockW;
 
-    DenseIndex m_mc;
-    DenseIndex m_nc;
-    DenseIndex m_kc;
+    Index m_mc;
+    Index m_nc;
+    Index m_kc;
 
   public:
 
     level3_blocking()
-      : m_blockA(0), m_blockB(0), m_blockW(0), m_mc(0), m_nc(0), m_kc(0)
+      : m_blockA(0), m_blockB(0), m_mc(0), m_nc(0), m_kc(0)
     {}
 
-    inline DenseIndex mc() const { return m_mc; }
-    inline DenseIndex nc() const { return m_nc; }
-    inline DenseIndex kc() const { return m_kc; }
+    inline Index mc() const { return m_mc; }
+    inline Index nc() const { return m_nc; }
+    inline Index kc() const { return m_kc; }
 
     inline LhsScalar* blockA() { return m_blockA; }
     inline RhsScalar* blockB() { return m_blockB; }
-    inline RhsScalar* blockW() { return m_blockW; }
 };
 
 template<int StorageOrder, typename _LhsScalar, typename _RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor>
-class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, true>
+class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, true /* == FiniteAtCompileTime */>
   : public level3_blocking<
       typename conditional<StorageOrder==RowMajor,_RhsScalar,_LhsScalar>::type,
       typename conditional<StorageOrder==RowMajor,_LhsScalar,_RhsScalar>::type>
@@ -282,29 +287,38 @@
     typedef gebp_traits<LhsScalar,RhsScalar> Traits;
     enum {
       SizeA = ActualRows * MaxDepth,
-      SizeB = ActualCols * MaxDepth,
-      SizeW = MaxDepth * Traits::WorkSpaceFactor
+      SizeB = ActualCols * MaxDepth
     };
 
-    EIGEN_ALIGN16 LhsScalar m_staticA[SizeA];
-    EIGEN_ALIGN16 RhsScalar m_staticB[SizeB];
-    EIGEN_ALIGN16 RhsScalar m_staticW[SizeW];
+#if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES
+    EIGEN_ALIGN_MAX LhsScalar m_staticA[SizeA];
+    EIGEN_ALIGN_MAX RhsScalar m_staticB[SizeB];
+#else
+    EIGEN_ALIGN_MAX char m_staticA[SizeA * sizeof(LhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1];
+    EIGEN_ALIGN_MAX char m_staticB[SizeB * sizeof(RhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1];
+#endif
 
   public:
 
-    gemm_blocking_space(DenseIndex /*rows*/, DenseIndex /*cols*/, DenseIndex /*depth*/)
+    gemm_blocking_space(Index /*rows*/, Index /*cols*/, Index /*depth*/, Index /*num_threads*/, bool /*full_rows = false*/)
     {
       this->m_mc = ActualRows;
       this->m_nc = ActualCols;
       this->m_kc = MaxDepth;
+#if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES
       this->m_blockA = m_staticA;
       this->m_blockB = m_staticB;
-      this->m_blockW = m_staticW;
+#else
+      this->m_blockA = reinterpret_cast<LhsScalar*>((internal::UIntPtr(m_staticA) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1));
+      this->m_blockB = reinterpret_cast<RhsScalar*>((internal::UIntPtr(m_staticB) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1));
+#endif
     }
 
+    void initParallel(Index, Index, Index, Index)
+    {}
+
     inline void allocateA() {}
     inline void allocateB() {}
-    inline void allocateW() {}
     inline void allocateAll() {}
 };
 
@@ -321,22 +335,42 @@
     typedef typename conditional<Transpose,_LhsScalar,_RhsScalar>::type RhsScalar;
     typedef gebp_traits<LhsScalar,RhsScalar> Traits;
 
-    DenseIndex m_sizeA;
-    DenseIndex m_sizeB;
-    DenseIndex m_sizeW;
+    Index m_sizeA;
+    Index m_sizeB;
 
   public:
 
-    gemm_blocking_space(DenseIndex rows, DenseIndex cols, DenseIndex depth)
+    gemm_blocking_space(Index rows, Index cols, Index depth, Index num_threads, bool l3_blocking)
     {
       this->m_mc = Transpose ? cols : rows;
       this->m_nc = Transpose ? rows : cols;
       this->m_kc = depth;
 
-      computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, this->m_nc);
+      if(l3_blocking)
+      {
+        computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, this->m_nc, num_threads);
+      }
+      else  // no l3 blocking
+      {
+        Index n = this->m_nc;
+        computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, n, num_threads);
+      }
+
       m_sizeA = this->m_mc * this->m_kc;
       m_sizeB = this->m_kc * this->m_nc;
-      m_sizeW = this->m_kc*Traits::WorkSpaceFactor;
+    }
+
+    void initParallel(Index rows, Index cols, Index depth, Index num_threads)
+    {
+      this->m_mc = Transpose ? cols : rows;
+      this->m_nc = Transpose ? rows : cols;
+      this->m_kc = depth;
+
+      eigen_internal_assert(this->m_blockA==0 && this->m_blockB==0);
+      Index m = this->m_mc;
+      computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, m, this->m_nc, num_threads);
+      m_sizeA = this->m_mc * this->m_kc;
+      m_sizeB = this->m_kc * this->m_nc;
     }
 
     void allocateA()
@@ -351,77 +385,108 @@
         this->m_blockB = aligned_new<RhsScalar>(m_sizeB);
     }
 
-    void allocateW()
-    {
-      if(this->m_blockW==0)
-        this->m_blockW = aligned_new<RhsScalar>(m_sizeW);
-    }
-
     void allocateAll()
     {
       allocateA();
       allocateB();
-      allocateW();
     }
 
     ~gemm_blocking_space()
     {
       aligned_delete(this->m_blockA, m_sizeA);
       aligned_delete(this->m_blockB, m_sizeB);
-      aligned_delete(this->m_blockW, m_sizeW);
     }
 };
 
 } // end namespace internal
 
+namespace internal {
+
 template<typename Lhs, typename Rhs>
-class GeneralProduct<Lhs, Rhs, GemmProduct>
-  : public ProductBase<GeneralProduct<Lhs,Rhs,GemmProduct>, Lhs, Rhs>
+struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct> >
 {
-    enum {
-      MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime)
-    };
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct)
-    
-    typedef typename  Lhs::Scalar LhsScalar;
-    typedef typename  Rhs::Scalar RhsScalar;
-    typedef           Scalar      ResScalar;
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  typedef typename Lhs::Scalar LhsScalar;
+  typedef typename Rhs::Scalar RhsScalar;
 
-    GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
+  typedef internal::blas_traits<Lhs> LhsBlasTraits;
+  typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+  typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
+
+  typedef internal::blas_traits<Rhs> RhsBlasTraits;
+  typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
+
+  enum {
+    MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime)
+  };
+
+  typedef generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> lazyproduct;
+
+  template<typename Dst>
+  static void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
+      lazyproduct::evalTo(dst, lhs, rhs);
+    else
     {
-      typedef internal::scalar_product_op<LhsScalar,RhsScalar> BinOp;
-      EIGEN_CHECK_BINARY_COMPATIBILIY(BinOp,LhsScalar,RhsScalar);
+      dst.setZero();
+      scaleAndAddTo(dst, lhs, rhs, Scalar(1));
     }
+  }
 
-    template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
-    {
-      eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+  template<typename Dst>
+  static void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
+      lazyproduct::addTo(dst, lhs, rhs);
+    else
+      scaleAndAddTo(dst,lhs, rhs, Scalar(1));
+  }
 
-      typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-      typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+  template<typename Dst>
+  static void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    if((rhs.rows()+dst.rows()+dst.cols())<20 && rhs.rows()>0)
+      lazyproduct::subTo(dst, lhs, rhs);
+    else
+      scaleAndAddTo(dst, lhs, rhs, Scalar(-1));
+  }
 
-      Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                                 * RhsBlasTraits::extractScalarFactor(m_rhs);
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& a_lhs, const Rhs& a_rhs, const Scalar& alpha)
+  {
+    eigen_assert(dst.rows()==a_lhs.rows() && dst.cols()==a_rhs.cols());
+    if(a_lhs.cols()==0 || a_lhs.rows()==0 || a_rhs.cols()==0)
+      return;
 
-      typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar,
-              Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType;
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
 
-      typedef internal::gemm_functor<
-        Scalar, Index,
-        internal::general_matrix_matrix_product<
-          Index,
-          LhsScalar, (_ActualLhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
-          RhsScalar, (_ActualRhsType::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
-          (Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
-        _ActualLhsType, _ActualRhsType, Dest, BlockingType> GemmFunctor;
+    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
+                               * RhsBlasTraits::extractScalarFactor(a_rhs);
 
-      BlockingType blocking(dst.rows(), dst.cols(), lhs.cols());
+    typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar,
+            Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType;
 
-      internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)>(GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), this->rows(), this->cols(), Dest::Flags&RowMajorBit);
-    }
+    typedef internal::gemm_functor<
+      Scalar, Index,
+      internal::general_matrix_matrix_product<
+        Index,
+        LhsScalar, (ActualLhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate),
+        RhsScalar, (ActualRhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate),
+        (Dest::Flags&RowMajorBit) ? RowMajor : ColMajor>,
+      ActualLhsTypeCleaned, ActualRhsTypeCleaned, Dest, BlockingType> GemmFunctor;
+
+    BlockingType blocking(dst.rows(), dst.cols(), lhs.cols(), 1, true);
+    internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)>
+        (GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), a_lhs.rows(), a_rhs.cols(), a_lhs.cols(), Dest::Flags&RowMajorBit);
+  }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_GENERAL_MATRIX_MATRIX_H
diff --git a/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h b/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h
index 5c37639..e844e37 100644
--- a/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h
+++ b/Eigen/src/Core/products/GeneralMatrixMatrixTriangular.h
@@ -20,7 +20,7 @@
 /**********************************************************************
 * This file implements a general A * B product while
 * evaluating only one triangular part of the product.
-* This is more general version of self adjoint product (C += A A^T)
+* This is a more general version of self adjoint product (C += A A^T)
 * as the level 3 SYRK Blas routine.
 **********************************************************************/
 
@@ -40,15 +40,16 @@
                           typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int  UpLo, int Version>
 struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,UpLo,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* lhs, Index lhsStride,
-                                      const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resStride, const ResScalar& alpha)
+                                      const RhsScalar* rhs, Index rhsStride, ResScalar* res, Index resStride,
+                                      const ResScalar& alpha, level3_blocking<RhsScalar,LhsScalar>& blocking)
   {
     general_matrix_matrix_triangular_product<Index,
         RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs,
         LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs,
         ColMajor, UpLo==Lower?Upper:Lower>
-      ::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha);
+      ::run(size,depth,rhs,rhsStride,lhs,lhsStride,res,resStride,alpha,blocking);
   }
 };
 
@@ -56,32 +57,36 @@
                           typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, int  UpLo, int Version>
 struct general_matrix_matrix_triangular_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,UpLo,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const LhsScalar* _lhs, Index lhsStride,
-                                      const RhsScalar* _rhs, Index rhsStride, ResScalar* res, Index resStride, const ResScalar& alpha)
+                                      const RhsScalar* _rhs, Index rhsStride, ResScalar* _res, Index resStride,
+                                      const ResScalar& alpha, level3_blocking<LhsScalar,RhsScalar>& blocking)
   {
-    const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
-
     typedef gebp_traits<LhsScalar,RhsScalar> Traits;
 
-    Index kc = depth; // cache block size along the K direction
-    Index mc = size;  // cache block size along the M direction
-    Index nc = size;  // cache block size along the N direction
-    computeProductBlockingSizes<LhsScalar,RhsScalar>(kc, mc, nc);
+    typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
+
+    Index kc = blocking.kc();
+    Index mc = (std::min)(size,blocking.mc());
+
     // !!! mc must be a multiple of nr:
     if(mc > Traits::nr)
       mc = (mc/Traits::nr)*Traits::nr;
 
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    std::size_t sizeB = sizeW + kc*size;
-    ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, kc*mc, 0);
-    ei_declare_aligned_stack_constructed_variable(RhsScalar, allocatedBlockB, sizeB, 0);
-    RhsScalar* blockB = allocatedBlockB + sizeW;
-    
-    gemm_pack_lhs<LhsScalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<RhsScalar, Index, Traits::nr, RhsStorageOrder> pack_rhs;
-    gebp_kernel <LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
+    std::size_t sizeA = kc*mc;
+    std::size_t sizeB = kc*size;
+
+    ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, blocking.blockA());
+    ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, blocking.blockB());
+
+    gemm_pack_lhs<LhsScalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    gemm_pack_rhs<RhsScalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
+    gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp;
     tribb_kernel<LhsScalar, RhsScalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs, UpLo> sybb;
 
     for(Index k2=0; k2<depth; k2+=kc)
@@ -89,29 +94,30 @@
       const Index actual_kc = (std::min)(k2+kc,depth)-k2;
 
       // note that the actual rhs is the transpose/adjoint of mat
-      pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, size);
+      pack_rhs(blockB, rhs.getSubMapper(k2,0), actual_kc, size);
 
       for(Index i2=0; i2<size; i2+=mc)
       {
         const Index actual_mc = (std::min)(i2+mc,size)-i2;
 
-        pack_lhs(blockA, &lhs(i2, k2), lhsStride, actual_kc, actual_mc);
+        pack_lhs(blockA, lhs.getSubMapper(i2, k2), actual_kc, actual_mc);
 
         // the selected actual_mc * size panel of res is split into three different part:
         //  1 - before the diagonal => processed with gebp or skipped
         //  2 - the actual_mc x actual_mc symmetric block => processed with a special kernel
         //  3 - after the diagonal => processed with gebp or skipped
         if (UpLo==Lower)
-          gebp(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, (std::min)(size,i2), alpha,
-               -1, -1, 0, 0, allocatedBlockB);
+          gebp(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc,
+               (std::min)(size,i2), alpha, -1, -1, 0, 0);
 
-        sybb(res+resStride*i2 + i2, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha, allocatedBlockB);
+
+        sybb(_res+resStride*i2 + i2, resStride, blockA, blockB + actual_kc*i2, actual_mc, actual_kc, alpha);
 
         if (UpLo==Upper)
         {
           Index j2 = i2+actual_mc;
-          gebp(res+resStride*j2+i2, resStride, blockA, blockB+actual_kc*j2, actual_mc, actual_kc, (std::max)(Index(0), size-j2), alpha,
-               -1, -1, 0, 0, allocatedBlockB);
+          gebp(res.getSubMapper(i2, j2), blockA, blockB+actual_kc*j2, actual_mc,
+               actual_kc, (std::max)(Index(0), size-j2), alpha, -1, -1, 0, 0);
         }
       }
     }
@@ -132,14 +138,17 @@
 {
   typedef gebp_traits<LhsScalar,RhsScalar,ConjLhs,ConjRhs> Traits;
   typedef typename Traits::ResScalar ResScalar;
-  
+
   enum {
-    BlockSize  = EIGEN_PLAIN_ENUM_MAX(mr,nr)
+    BlockSize  = meta_least_common_multiple<EIGEN_PLAIN_ENUM_MAX(mr,nr),EIGEN_PLAIN_ENUM_MIN(mr,nr)>::ret
   };
-  void operator()(ResScalar* res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha, RhsScalar* workspace)
+  void operator()(ResScalar* _res, Index resStride, const LhsScalar* blockA, const RhsScalar* blockB, Index size, Index depth, const ResScalar& alpha)
   {
-    gebp_kernel<LhsScalar, RhsScalar, Index, mr, nr, ConjLhs, ConjRhs> gebp_kernel;
-    Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer;
+    typedef blas_data_mapper<ResScalar, Index, ColMajor> ResMapper;
+    ResMapper res(_res, resStride);
+    gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, mr, nr, ConjLhs, ConjRhs> gebp_kernel;
+
+    Matrix<ResScalar,BlockSize,BlockSize,ColMajor> buffer((internal::constructor_without_unaligned_array_assert()));
 
     // let's process the block per panel of actual_mc x BlockSize,
     // again, each is split into three parts, etc.
@@ -149,20 +158,20 @@
       const RhsScalar* actual_b = blockB+j*depth;
 
       if(UpLo==Upper)
-        gebp_kernel(res+j*resStride, resStride, blockA, actual_b, j, depth, actualBlockSize, alpha,
-                    -1, -1, 0, 0, workspace);
+        gebp_kernel(res.getSubMapper(0, j), blockA, actual_b, j, depth, actualBlockSize, alpha,
+                    -1, -1, 0, 0);
 
       // selfadjoint micro block
       {
         Index i = j;
         buffer.setZero();
         // 1 - apply the kernel on the temporary buffer
-        gebp_kernel(buffer.data(), BlockSize, blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
-                    -1, -1, 0, 0, workspace);
+        gebp_kernel(ResMapper(buffer.data(), BlockSize), blockA+depth*i, actual_b, actualBlockSize, depth, actualBlockSize, alpha,
+                    -1, -1, 0, 0);
         // 2 - triangular accumulation
         for(Index j1=0; j1<actualBlockSize; ++j1)
         {
-          ResScalar* r = res + (j+j1)*resStride + i;
+          ResScalar* r = &res(i, j + j1);
           for(Index i1=UpLo==Lower ? j1 : 0;
               UpLo==Lower ? i1<actualBlockSize : i1<=j1; ++i1)
             r[i1] += buffer(i1,j1);
@@ -172,8 +181,8 @@
       if(UpLo==Lower)
       {
         Index i = j+actualBlockSize;
-        gebp_kernel(res+j*resStride+i, resStride, blockA+depth*i, actual_b, size-i, depth, actualBlockSize, alpha,
-                    -1, -1, 0, 0, workspace);
+        gebp_kernel(res.getSubMapper(i, j), blockA+depth*i, actual_b, size-i, 
+                    depth, actualBlockSize, alpha, -1, -1, 0, 0);
       }
     }
   }
@@ -190,10 +199,9 @@
 template<typename MatrixType, typename ProductType, int UpLo>
 struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,true>
 {
-  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
   {
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
     
     typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
     typedef internal::blas_traits<Lhs> LhsBlasTraits;
@@ -209,6 +217,9 @@
 
     Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
 
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
     enum {
       StorageOrder = (internal::traits<MatrixType>::Flags&RowMajorBit) ? RowMajor : ColMajor,
       UseLhsDirectly = _ActualLhs::InnerStrideAtCompileTime==1,
@@ -236,10 +247,8 @@
 template<typename MatrixType, typename ProductType, int UpLo>
 struct general_product_to_triangular_selector<MatrixType,ProductType,UpLo,false>
 {
-  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha)
+  static void run(MatrixType& mat, const ProductType& prod, const typename MatrixType::Scalar& alpha, bool beta)
   {
-    typedef typename MatrixType::Index Index;
-    
     typedef typename internal::remove_all<typename ProductType::LhsNested>::type Lhs;
     typedef internal::blas_traits<Lhs> LhsBlasTraits;
     typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhs;
@@ -254,23 +263,47 @@
 
     typename ProductType::Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs().derived()) * RhsBlasTraits::extractScalarFactor(prod.rhs().derived());
 
+    if(!beta)
+      mat.template triangularView<UpLo>().setZero();
+
+    enum {
+      IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0,
+      LhsIsRowMajor = _ActualLhs::Flags&RowMajorBit ? 1 : 0,
+      RhsIsRowMajor = _ActualRhs::Flags&RowMajorBit ? 1 : 0,
+      SkipDiag = (UpLo&(UnitDiag|ZeroDiag))!=0
+    };
+
+    Index size = mat.cols();
+    if(SkipDiag)
+      size--;
+    Index depth = actualLhs.cols();
+
+    typedef internal::gemm_blocking_space<IsRowMajor ? RowMajor : ColMajor,typename Lhs::Scalar,typename Rhs::Scalar,
+          MatrixType::MaxColsAtCompileTime, MatrixType::MaxColsAtCompileTime, _ActualRhs::MaxColsAtCompileTime> BlockingType;
+
+    BlockingType blocking(size, size, depth, 1, false);
+
     internal::general_matrix_matrix_triangular_product<Index,
-      typename Lhs::Scalar, _ActualLhs::Flags&RowMajorBit ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
-      typename Rhs::Scalar, _ActualRhs::Flags&RowMajorBit ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
-      MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor, UpLo>
-      ::run(mat.cols(), actualLhs.cols(),
-            &actualLhs.coeffRef(0,0), actualLhs.outerStride(), &actualRhs.coeffRef(0,0), actualRhs.outerStride(),
-            mat.data(), mat.outerStride(), actualAlpha);
+      typename Lhs::Scalar, LhsIsRowMajor ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
+      typename Rhs::Scalar, RhsIsRowMajor ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
+      IsRowMajor ? RowMajor : ColMajor, UpLo&(Lower|Upper)>
+      ::run(size, depth,
+            &actualLhs.coeffRef(SkipDiag&&(UpLo&Lower)==Lower ? 1 : 0,0), actualLhs.outerStride(),
+            &actualRhs.coeffRef(0,SkipDiag&&(UpLo&Upper)==Upper ? 1 : 0), actualRhs.outerStride(),
+            mat.data() + (SkipDiag ? (bool(IsRowMajor) != ((UpLo&Lower)==Lower) ? 1 : mat.outerStride() ) : 0), mat.outerStride(), actualAlpha, blocking);
   }
 };
 
 template<typename MatrixType, unsigned int UpLo>
-template<typename ProductDerived, typename _Lhs, typename _Rhs>
-TriangularView<MatrixType,UpLo>& TriangularView<MatrixType,UpLo>::assignProduct(const ProductBase<ProductDerived, _Lhs,_Rhs>& prod, const Scalar& alpha)
+template<typename ProductType>
+TriangularView<MatrixType,UpLo>& TriangularViewImpl<MatrixType,UpLo,Dense>::_assignProduct(const ProductType& prod, const Scalar& alpha, bool beta)
 {
-  general_product_to_triangular_selector<MatrixType, ProductDerived, UpLo, (_Lhs::ColsAtCompileTime==1) || (_Rhs::RowsAtCompileTime==1)>::run(m_matrix.const_cast_derived(), prod.derived(), alpha);
+  EIGEN_STATIC_ASSERT((UpLo&UnitDiag)==0, WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED);
+  eigen_assert(derived().nestedExpression().rows() == prod.rows() && derived().cols() == prod.cols());
   
-  return *this;
+  general_product_to_triangular_selector<MatrixType, ProductType, UpLo, internal::traits<ProductType>::InnerSize==1>::run(derived().nestedExpression().const_cast_derived(), prod, alpha, beta);
+  
+  return derived();
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_MKL.h b/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h
similarity index 66%
rename from Eigen/src/Core/products/GeneralMatrixMatrixTriangular_MKL.h
rename to Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h
index 3deed06..f6f9ebe 100644
--- a/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_MKL.h
+++ b/Eigen/src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h
@@ -25,15 +25,15 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to BLAS F77
  *   Level 3 BLAS SYRK/HERK implementation.
  ********************************************************************************
 */
 
-#ifndef EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_MKL_H
-#define EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_MKL_H
+#ifndef EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
+#define EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -44,34 +44,35 @@
 
 
 // try to go to BLAS specialization
-#define EIGEN_MKL_RANKUPDATE_SPECIALIZE(Scalar) \
+#define EIGEN_BLAS_RANKUPDATE_SPECIALIZE(Scalar) \
 template <typename Index, int LhsStorageOrder, bool ConjugateLhs, \
                           int RhsStorageOrder, bool ConjugateRhs, int  UpLo> \
 struct general_matrix_matrix_triangular_product<Index,Scalar,LhsStorageOrder,ConjugateLhs, \
                Scalar,RhsStorageOrder,ConjugateRhs,ColMajor,UpLo,Specialized> { \
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const Scalar* lhs, Index lhsStride, \
-                          const Scalar* rhs, Index rhsStride, Scalar* res, Index resStride, Scalar alpha) \
+                          const Scalar* rhs, Index rhsStride, Scalar* res, Index resStride, Scalar alpha, level3_blocking<Scalar, Scalar>& blocking) \
   { \
-    if (lhs==rhs) { \
+    if ( lhs==rhs && ((UpLo&(Lower|Upper))==UpLo) ) { \
       general_matrix_matrix_rankupdate<Index,Scalar,LhsStorageOrder,ConjugateLhs,ColMajor,UpLo> \
-      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha); \
+      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha,blocking); \
     } else { \
       general_matrix_matrix_triangular_product<Index, \
         Scalar, LhsStorageOrder, ConjugateLhs, \
         Scalar, RhsStorageOrder, ConjugateRhs, \
         ColMajor, UpLo, BuiltIn> \
-      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha); \
+      ::run(size,depth,lhs,lhsStride,rhs,rhsStride,res,resStride,alpha,blocking); \
     } \
   } \
 };
 
-EIGEN_MKL_RANKUPDATE_SPECIALIZE(double)
-//EIGEN_MKL_RANKUPDATE_SPECIALIZE(dcomplex)
-EIGEN_MKL_RANKUPDATE_SPECIALIZE(float)
-//EIGEN_MKL_RANKUPDATE_SPECIALIZE(scomplex)
+EIGEN_BLAS_RANKUPDATE_SPECIALIZE(double)
+EIGEN_BLAS_RANKUPDATE_SPECIALIZE(float)
+// TODO handle complex cases
+// EIGEN_BLAS_RANKUPDATE_SPECIALIZE(dcomplex)
+// EIGEN_BLAS_RANKUPDATE_SPECIALIZE(scomplex)
 
 // SYRK for float/double
-#define EIGEN_MKL_RANKUPDATE_R(EIGTYPE, MKLTYPE, MKLFUNC) \
+#define EIGEN_BLAS_RANKUPDATE_R(EIGTYPE, BLASTYPE, BLASFUNC) \
 template <typename Index, int AStorageOrder, bool ConjugateA, int  UpLo> \
 struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,ColMajor,UpLo> { \
   enum { \
@@ -80,23 +81,19 @@
     conjA = ((AStorageOrder==ColMajor) && ConjugateA) ? 1 : 0 \
   }; \
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const EIGTYPE* lhs, Index lhsStride, \
-                          const EIGTYPE* rhs, Index rhsStride, EIGTYPE* res, Index resStride, EIGTYPE alpha) \
+                          const EIGTYPE* /*rhs*/, Index /*rhsStride*/, EIGTYPE* res, Index resStride, EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
   { \
   /* typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs;*/ \
 \
-   MKL_INT lda=lhsStride, ldc=resStride, n=size, k=depth; \
-   char uplo=(IsLower) ? 'L' : 'U', trans=(AStorageOrder==RowMajor) ? 'T':'N'; \
-   MKLTYPE alpha_, beta_; \
-\
-/* Set alpha_ & beta_ */ \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(beta_, EIGTYPE(1)); \
-   MKLFUNC(&uplo, &trans, &n, &k, &alpha_, lhs, &lda, &beta_, res, &ldc); \
+   BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'T':'N'); \
+   EIGTYPE beta(1); \
+   BLASFUNC(&uplo, &trans, &n, &k, (const BLASTYPE*)&numext::real_ref(alpha), lhs, &lda, (const BLASTYPE*)&numext::real_ref(beta), res, &ldc); \
   } \
 };
 
 // HERK for complex data
-#define EIGEN_MKL_RANKUPDATE_C(EIGTYPE, MKLTYPE, RTYPE, MKLFUNC) \
+#define EIGEN_BLAS_RANKUPDATE_C(EIGTYPE, BLASTYPE, RTYPE, BLASFUNC) \
 template <typename Index, int AStorageOrder, bool ConjugateA, int  UpLo> \
 struct general_matrix_matrix_rankupdate<Index,EIGTYPE,AStorageOrder,ConjugateA,ColMajor,UpLo> { \
   enum { \
@@ -105,18 +102,15 @@
     conjA = (((AStorageOrder==ColMajor) && ConjugateA) || ((AStorageOrder==RowMajor) && !ConjugateA)) ? 1 : 0 \
   }; \
   static EIGEN_STRONG_INLINE void run(Index size, Index depth,const EIGTYPE* lhs, Index lhsStride, \
-                          const EIGTYPE* rhs, Index rhsStride, EIGTYPE* res, Index resStride, EIGTYPE alpha) \
+                          const EIGTYPE* /*rhs*/, Index /*rhsStride*/, EIGTYPE* res, Index resStride, EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
   { \
    typedef Matrix<EIGTYPE, Dynamic, Dynamic, AStorageOrder> MatrixType; \
 \
-   MKL_INT lda=lhsStride, ldc=resStride, n=size, k=depth; \
-   char uplo=(IsLower) ? 'L' : 'U', trans=(AStorageOrder==RowMajor) ? 'C':'N'; \
+   BlasIndex lda=convert_index<BlasIndex>(lhsStride), ldc=convert_index<BlasIndex>(resStride), n=convert_index<BlasIndex>(size), k=convert_index<BlasIndex>(depth); \
+   char uplo=((IsLower) ? 'L' : 'U'), trans=((AStorageOrder==RowMajor) ? 'C':'N'); \
    RTYPE alpha_, beta_; \
    const EIGTYPE* a_ptr; \
 \
-/* Set alpha_ & beta_ */ \
-/*   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); */\
-/*   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(beta_, EIGTYPE(1));*/ \
    alpha_ = alpha.real(); \
    beta_ = 1.0; \
 /* Copy with conjugation in some cases*/ \
@@ -127,20 +121,25 @@
      lda = a.outerStride(); \
      a_ptr = a.data(); \
    } else a_ptr=lhs; \
-   MKLFUNC(&uplo, &trans, &n, &k, &alpha_, (MKLTYPE*)a_ptr, &lda, &beta_, (MKLTYPE*)res, &ldc); \
+   BLASFUNC(&uplo, &trans, &n, &k, &alpha_, (BLASTYPE*)a_ptr, &lda, &beta_, (BLASTYPE*)res, &ldc); \
   } \
 };
 
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_RANKUPDATE_R(double, double, dsyrk)
+EIGEN_BLAS_RANKUPDATE_R(float,  float,  ssyrk)
+#else
+EIGEN_BLAS_RANKUPDATE_R(double, double, dsyrk_)
+EIGEN_BLAS_RANKUPDATE_R(float,  float,  ssyrk_)
+#endif
 
-EIGEN_MKL_RANKUPDATE_R(double, double, dsyrk)
-EIGEN_MKL_RANKUPDATE_R(float,  float,  ssyrk)
-
-//EIGEN_MKL_RANKUPDATE_C(dcomplex, MKL_Complex16, double, zherk)
-//EIGEN_MKL_RANKUPDATE_C(scomplex, MKL_Complex8,  double, cherk)
+// TODO hanlde complex cases
+// EIGEN_BLAS_RANKUPDATE_C(dcomplex, double, double, zherk_)
+// EIGEN_BLAS_RANKUPDATE_C(scomplex, float,  float, cherk_)
 
 
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_MKL_H
+#endif // EIGEN_GENERAL_MATRIX_MATRIX_TRIANGULAR_BLAS_H
diff --git a/Eigen/src/Core/products/GeneralMatrixMatrix_MKL.h b/Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h
similarity index 72%
rename from Eigen/src/Core/products/GeneralMatrixMatrix_MKL.h
rename to Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h
index 060af32..b0f6b0d 100644
--- a/Eigen/src/Core/products/GeneralMatrixMatrix_MKL.h
+++ b/Eigen/src/Core/products/GeneralMatrixMatrix_BLAS.h
@@ -25,13 +25,13 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to BLAS F77
  *   General matrix-matrix product functionality based on ?GEMM.
  ********************************************************************************
 */
 
-#ifndef EIGEN_GENERAL_MATRIX_MATRIX_MKL_H
-#define EIGEN_GENERAL_MATRIX_MATRIX_MKL_H
+#ifndef EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
+#define EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
 
 namespace Eigen { 
 
@@ -46,13 +46,15 @@
 
 // gemm specialization
 
-#define GEMM_SPECIALIZATION(EIGTYPE, EIGPREFIX, MKLTYPE, MKLPREFIX) \
+#define GEMM_SPECIALIZATION(EIGTYPE, EIGPREFIX, BLASTYPE, BLASFUNC) \
 template< \
   typename Index, \
   int LhsStorageOrder, bool ConjugateLhs, \
   int RhsStorageOrder, bool ConjugateRhs> \
 struct general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor> \
 { \
+typedef gebp_traits<EIGTYPE,EIGTYPE> Traits; \
+\
 static void run(Index rows, Index cols, Index depth, \
   const EIGTYPE* _lhs, Index lhsStride, \
   const EIGTYPE* _rhs, Index rhsStride, \
@@ -64,55 +66,57 @@
   using std::conj; \
 \
   char transa, transb; \
-  MKL_INT m, n, k, lda, ldb, ldc; \
+  BlasIndex m, n, k, lda, ldb, ldc; \
   const EIGTYPE *a, *b; \
-  MKLTYPE alpha_, beta_; \
+  EIGTYPE beta(1); \
   MatrixX##EIGPREFIX a_tmp, b_tmp; \
-  EIGTYPE myone(1);\
 \
 /* Set transpose options */ \
   transa = (LhsStorageOrder==RowMajor) ? ((ConjugateLhs) ? 'C' : 'T') : 'N'; \
   transb = (RhsStorageOrder==RowMajor) ? ((ConjugateRhs) ? 'C' : 'T') : 'N'; \
 \
 /* Set m, n, k */ \
-  m = (MKL_INT)rows;  \
-  n = (MKL_INT)cols;  \
-  k = (MKL_INT)depth; \
-\
-/* Set alpha_ & beta_ */ \
-  assign_scalar_eig2mkl(alpha_, alpha); \
-  assign_scalar_eig2mkl(beta_, myone); \
+  m = convert_index<BlasIndex>(rows);  \
+  n = convert_index<BlasIndex>(cols);  \
+  k = convert_index<BlasIndex>(depth); \
 \
 /* Set lda, ldb, ldc */ \
-  lda = (MKL_INT)lhsStride; \
-  ldb = (MKL_INT)rhsStride; \
-  ldc = (MKL_INT)resStride; \
+  lda = convert_index<BlasIndex>(lhsStride); \
+  ldb = convert_index<BlasIndex>(rhsStride); \
+  ldc = convert_index<BlasIndex>(resStride); \
 \
 /* Set a, b, c */ \
   if ((LhsStorageOrder==ColMajor) && (ConjugateLhs)) { \
     Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,m,k,OuterStride<>(lhsStride)); \
     a_tmp = lhs.conjugate(); \
     a = a_tmp.data(); \
-    lda = a_tmp.outerStride(); \
+    lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
   } else a = _lhs; \
 \
   if ((RhsStorageOrder==ColMajor) && (ConjugateRhs)) { \
     Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,k,n,OuterStride<>(rhsStride)); \
     b_tmp = rhs.conjugate(); \
     b = b_tmp.data(); \
-    ldb = b_tmp.outerStride(); \
+    ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
   } else b = _rhs; \
 \
-  MKLPREFIX##gemm(&transa, &transb, &m, &n, &k, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)b, &ldb, &beta_, (MKLTYPE*)res, &ldc); \
+  BLASFUNC(&transa, &transb, &m, &n, &k, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
 }};
 
-GEMM_SPECIALIZATION(double,   d,  double,        d)
-GEMM_SPECIALIZATION(float,    f,  float,         s)
-GEMM_SPECIALIZATION(dcomplex, cd, MKL_Complex16, z)
-GEMM_SPECIALIZATION(scomplex, cf, MKL_Complex8,  c)
+#ifdef EIGEN_USE_MKL
+GEMM_SPECIALIZATION(double,   d,  double, dgemm)
+GEMM_SPECIALIZATION(float,    f,  float,  sgemm)
+GEMM_SPECIALIZATION(dcomplex, cd, MKL_Complex16, zgemm)
+GEMM_SPECIALIZATION(scomplex, cf, MKL_Complex8,  cgemm)
+#else
+GEMM_SPECIALIZATION(double,   d,  double, dgemm_)
+GEMM_SPECIALIZATION(float,    f,  float,  sgemm_)
+GEMM_SPECIALIZATION(dcomplex, cd, double, zgemm_)
+GEMM_SPECIALIZATION(scomplex, cf, float,  cgemm_)
+#endif
 
 } // end namespase internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_GENERAL_MATRIX_MATRIX_MKL_H
+#endif // EIGEN_GENERAL_MATRIX_MATRIX_BLAS_H
diff --git a/Eigen/src/Core/products/GeneralMatrixVector.h b/Eigen/src/Core/products/GeneralMatrixVector.h
index 0938770..a597c1f 100644
--- a/Eigen/src/Core/products/GeneralMatrixVector.h
+++ b/Eigen/src/Core/products/GeneralMatrixVector.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_GENERAL_MATRIX_VECTOR_H
 #define EIGEN_GENERAL_MATRIX_VECTOR_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -26,11 +26,39 @@
  *  |real |cplx |real | alpha is converted to a cplx when calling the run function, no vectorization
  *  |cplx |real |cplx | invalid, the caller has to do tmp: = A * B; C += alpha*tmp
  *  |cplx |real |real | optimal case, vectorization possible via real-cplx mul
+ *
+ * Accesses to the matrix coefficients follow the following logic:
+ *
+ * - if all columns have the same alignment then
+ *   - if the columns have the same alignment as the result vector, then easy! (-> AllAligned case)
+ *   - otherwise perform unaligned loads only (-> NoneAligned case)
+ * - otherwise
+ *   - if even columns have the same alignment then
+ *     // odd columns are guaranteed to have the same alignment too
+ *     - if even or odd columns have the same alignment as the result, then
+ *       // for a register size of 2 scalars, this is guarantee to be the case (e.g., SSE with double)
+ *       - perform half aligned and half unaligned loads (-> EvenAligned case)
+ *     - otherwise perform unaligned loads only (-> NoneAligned case)
+ *   - otherwise, if the register size is 4 scalars (e.g., SSE with float) then
+ *     - one over 4 consecutive columns is guaranteed to be aligned with the result vector,
+ *       perform simple aligned loads for this column and aligned loads plus re-alignment for the other. (-> FirstAligned case)
+ *       // this re-alignment is done by the palign function implemented for SSE in Eigen/src/Core/arch/SSE/PacketMath.h
+ *   - otherwise,
+ *     // if we get here, this means the register size is greater than 4 (e.g., AVX with floats),
+ *     // we currently fall back to the NoneAligned case
+ *
+ * The same reasoning apply for the transposed case.
+ *
+ * The last case (PacketSize>4) could probably be improved by generalizing the FirstAligned case, but since we do not support AVX yet...
+ * One might also wonder why in the EvenAligned case we perform unaligned loads instead of using the aligned-loads plus re-alignment
+ * strategy as in the FirstAligned case. The reason is that we observed that unaligned loads on a 8 byte boundary are not too slow
+ * compared to unaligned loads on a 4 byte boundary.
+ *
  */
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-struct general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+struct general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>
 {
-typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
 enum {
   Vectorizable = packet_traits<LhsScalar>::Vectorizable && packet_traits<RhsScalar>::Vectorizable
@@ -50,31 +78,35 @@
 
 EIGEN_DONT_INLINE static void run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index resIncr, RhsScalar alpha);
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
+        ResScalar* res, Index resIncr,
+  RhsScalar alpha);
 };
 
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>::run(
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>::run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index resIncr, RhsScalar alpha)
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
+        ResScalar* res, Index resIncr,
+  RhsScalar alpha)
 {
-  EIGEN_UNUSED_VARIABLE(resIncr)
+  EIGEN_UNUSED_VARIABLE(resIncr);
   eigen_internal_assert(resIncr==1);
   #ifdef _EIGEN_ACCUMULATE_PACKETS
   #error _EIGEN_ACCUMULATE_PACKETS has already been defined
   #endif
-  #define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) \
+  #define _EIGEN_ACCUMULATE_PACKETS(Alignment0,Alignment13,Alignment2) \
     pstore(&res[j], \
       padd(pload<ResPacket>(&res[j]), \
         padd( \
-          padd(pcj.pmul(EIGEN_CAT(ploa , A0)<LhsPacket>(&lhs0[j]),    ptmp0), \
-                  pcj.pmul(EIGEN_CAT(ploa , A13)<LhsPacket>(&lhs1[j]),   ptmp1)), \
-          padd(pcj.pmul(EIGEN_CAT(ploa , A2)<LhsPacket>(&lhs2[j]),    ptmp2), \
-                  pcj.pmul(EIGEN_CAT(ploa , A13)<LhsPacket>(&lhs3[j]),   ptmp3)) )))
+      padd(pcj.pmul(lhs0.template load<LhsPacket, Alignment0>(j),    ptmp0), \
+      pcj.pmul(lhs1.template load<LhsPacket, Alignment13>(j),   ptmp1)),   \
+      padd(pcj.pmul(lhs2.template load<LhsPacket, Alignment2>(j),    ptmp2), \
+      pcj.pmul(lhs3.template load<LhsPacket, Alignment13>(j),   ptmp3)) )))
+
+  typedef typename LhsMapper::VectorMapper LhsScalars;
 
   conj_helper<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> cj;
   conj_helper<LhsPacket,RhsPacket,ConjugateLhs,ConjugateRhs> pcj;
@@ -88,10 +120,12 @@
   const Index ResPacketAlignedMask = ResPacketSize-1;
 //  const Index PeelAlignedMask = ResPacketSize*peels-1;
   const Index size = rows;
-  
+
+  const Index lhsStride = lhs.stride();
+
   // How many coeffs of the result do we have to skip to be aligned.
   // Here we assume data are at least aligned on the base scalar type.
-  Index alignedStart = internal::first_aligned(res,size);
+  Index alignedStart = internal::first_default_aligned(res,size);
   Index alignedSize = ResPacketSize>1 ? alignedStart + ((size-alignedStart) & ~ResPacketAlignedMask) : 0;
   const Index peeledSize = alignedSize - RhsPacketSize*peels - RhsPacketSize + 1;
 
@@ -101,19 +135,26 @@
                        : FirstAligned;
 
   // we cannot assume the first element is aligned because of sub-matrices
-  const Index lhsAlignmentOffset = internal::first_aligned(lhs,size);
+  const Index lhsAlignmentOffset = lhs.firstAligned(size);
 
   // find how many columns do we have to skip to be aligned with the result (if possible)
   Index skipColumns = 0;
   // if the data cannot be aligned (TODO add some compile time tests when possible, e.g. for floats)
-  if( (size_t(lhs)%sizeof(LhsScalar)) || (size_t(res)%sizeof(ResScalar)) )
+  if( (lhsAlignmentOffset < 0) || (lhsAlignmentOffset == size) || (UIntPtr(res)%sizeof(ResScalar)) )
   {
     alignedSize = 0;
     alignedStart = 0;
+    alignmentPattern = NoneAligned;
+  }
+  else if(LhsPacketSize > 4)
+  {
+    // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
+    // Currently, it seems to be better to perform unaligned loads anyway
+    alignmentPattern = NoneAligned;
   }
   else if (LhsPacketSize>1)
   {
-    eigen_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0 || size<LhsPacketSize);
+  //    eigen_internal_assert(size_t(firstLhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0 || size<LhsPacketSize);
 
     while (skipColumns<LhsPacketSize &&
           alignedStart != ((lhsAlignmentOffset + alignmentStep*skipColumns)%LhsPacketSize))
@@ -130,10 +171,10 @@
       // note that the skiped columns are processed later.
     }
 
-    eigen_internal_assert(  (alignmentPattern==NoneAligned)
+    /*    eigen_internal_assert(  (alignmentPattern==NoneAligned)
                       || (skipColumns + columnsAtOnce >= cols)
                       || LhsPacketSize > size
-                      || (size_t(lhs+alignedStart+lhsStride*skipColumns)%sizeof(LhsPacket))==0);
+                      || (size_t(firstLhs+alignedStart+lhsStride*skipColumns)%sizeof(LhsPacket))==0);*/
   }
   else if(Vectorizable)
   {
@@ -142,20 +183,20 @@
     alignmentPattern = AllAligned;
   }
 
-  Index offset1 = (FirstAligned && alignmentStep==1?3:1);
-  Index offset3 = (FirstAligned && alignmentStep==1?1:3);
+  const Index offset1 = (alignmentPattern==FirstAligned && alignmentStep==1)?3:1;
+  const Index offset3 = (alignmentPattern==FirstAligned && alignmentStep==1)?1:3;
 
   Index columnBound = ((cols-skipColumns)/columnsAtOnce)*columnsAtOnce + skipColumns;
   for (Index i=skipColumns; i<columnBound; i+=columnsAtOnce)
   {
-    RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs[i*rhsIncr]),
-              ptmp1 = pset1<RhsPacket>(alpha*rhs[(i+offset1)*rhsIncr]),
-              ptmp2 = pset1<RhsPacket>(alpha*rhs[(i+2)*rhsIncr]),
-              ptmp3 = pset1<RhsPacket>(alpha*rhs[(i+offset3)*rhsIncr]);
+    RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs(i, 0)),
+              ptmp1 = pset1<RhsPacket>(alpha*rhs(i+offset1, 0)),
+              ptmp2 = pset1<RhsPacket>(alpha*rhs(i+2, 0)),
+              ptmp3 = pset1<RhsPacket>(alpha*rhs(i+offset3, 0));
 
     // this helps a lot generating better binary code
-    const LhsScalar *lhs0 = lhs + i*lhsStride,     *lhs1 = lhs + (i+offset1)*lhsStride,
-                    *lhs2 = lhs + (i+2)*lhsStride, *lhs3 = lhs + (i+offset3)*lhsStride;
+    const LhsScalars lhs0 = lhs.getVectorMapper(0, i+0),   lhs1 = lhs.getVectorMapper(0, i+offset1),
+                     lhs2 = lhs.getVectorMapper(0, i+2),   lhs3 = lhs.getVectorMapper(0, i+offset3);
 
     if (Vectorizable)
     {
@@ -163,10 +204,10 @@
       // process initial unaligned coeffs
       for (Index j=0; j<alignedStart; ++j)
       {
-        res[j] = cj.pmadd(lhs0[j], pfirst(ptmp0), res[j]);
-        res[j] = cj.pmadd(lhs1[j], pfirst(ptmp1), res[j]);
-        res[j] = cj.pmadd(lhs2[j], pfirst(ptmp2), res[j]);
-        res[j] = cj.pmadd(lhs3[j], pfirst(ptmp3), res[j]);
+        res[j] = cj.pmadd(lhs0(j), pfirst(ptmp0), res[j]);
+        res[j] = cj.pmadd(lhs1(j), pfirst(ptmp1), res[j]);
+        res[j] = cj.pmadd(lhs2(j), pfirst(ptmp2), res[j]);
+        res[j] = cj.pmadd(lhs3(j), pfirst(ptmp3), res[j]);
       }
 
       if (alignedSize>alignedStart)
@@ -175,11 +216,11 @@
         {
           case AllAligned:
             for (Index j = alignedStart; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,d,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Aligned,Aligned);
             break;
           case EvenAligned:
             for (Index j = alignedStart; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Aligned);
             break;
           case FirstAligned:
           {
@@ -189,28 +230,28 @@
               LhsPacket A00, A01, A02, A03, A10, A11, A12, A13;
               ResPacket T0, T1;
 
-              A01 = pload<LhsPacket>(&lhs1[alignedStart-1]);
-              A02 = pload<LhsPacket>(&lhs2[alignedStart-2]);
-              A03 = pload<LhsPacket>(&lhs3[alignedStart-3]);
+              A01 = lhs1.template load<LhsPacket, Aligned>(alignedStart-1);
+              A02 = lhs2.template load<LhsPacket, Aligned>(alignedStart-2);
+              A03 = lhs3.template load<LhsPacket, Aligned>(alignedStart-3);
 
               for (; j<peeledSize; j+=peels*ResPacketSize)
               {
-                A11 = pload<LhsPacket>(&lhs1[j-1+LhsPacketSize]);  palign<1>(A01,A11);
-                A12 = pload<LhsPacket>(&lhs2[j-2+LhsPacketSize]);  palign<2>(A02,A12);
-                A13 = pload<LhsPacket>(&lhs3[j-3+LhsPacketSize]);  palign<3>(A03,A13);
+                A11 = lhs1.template load<LhsPacket, Aligned>(j-1+LhsPacketSize);  palign<1>(A01,A11);
+                A12 = lhs2.template load<LhsPacket, Aligned>(j-2+LhsPacketSize);  palign<2>(A02,A12);
+                A13 = lhs3.template load<LhsPacket, Aligned>(j-3+LhsPacketSize);  palign<3>(A03,A13);
 
-                A00 = pload<LhsPacket>(&lhs0[j]);
-                A10 = pload<LhsPacket>(&lhs0[j+LhsPacketSize]);
+                A00 = lhs0.template load<LhsPacket, Aligned>(j);
+                A10 = lhs0.template load<LhsPacket, Aligned>(j+LhsPacketSize);
                 T0  = pcj.pmadd(A00, ptmp0, pload<ResPacket>(&res[j]));
                 T1  = pcj.pmadd(A10, ptmp0, pload<ResPacket>(&res[j+ResPacketSize]));
 
                 T0  = pcj.pmadd(A01, ptmp1, T0);
-                A01 = pload<LhsPacket>(&lhs1[j-1+2*LhsPacketSize]);  palign<1>(A11,A01);
+                A01 = lhs1.template load<LhsPacket, Aligned>(j-1+2*LhsPacketSize);  palign<1>(A11,A01);
                 T0  = pcj.pmadd(A02, ptmp2, T0);
-                A02 = pload<LhsPacket>(&lhs2[j-2+2*LhsPacketSize]);  palign<2>(A12,A02);
+                A02 = lhs2.template load<LhsPacket, Aligned>(j-2+2*LhsPacketSize);  palign<2>(A12,A02);
                 T0  = pcj.pmadd(A03, ptmp3, T0);
                 pstore(&res[j],T0);
-                A03 = pload<LhsPacket>(&lhs3[j-3+2*LhsPacketSize]);  palign<3>(A13,A03);
+                A03 = lhs3.template load<LhsPacket, Aligned>(j-3+2*LhsPacketSize);  palign<3>(A13,A03);
                 T1  = pcj.pmadd(A11, ptmp1, T1);
                 T1  = pcj.pmadd(A12, ptmp2, T1);
                 T1  = pcj.pmadd(A13, ptmp3, T1);
@@ -218,12 +259,12 @@
               }
             }
             for (; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Unaligned);
             break;
           }
           default:
             for (Index j = alignedStart; j<alignedSize; j+=ResPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(du,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Unaligned,Unaligned,Unaligned);
             break;
         }
       }
@@ -232,10 +273,10 @@
     /* process remaining coeffs (or all if there is no explicit vectorization) */
     for (Index j=alignedSize; j<size; ++j)
     {
-      res[j] = cj.pmadd(lhs0[j], pfirst(ptmp0), res[j]);
-      res[j] = cj.pmadd(lhs1[j], pfirst(ptmp1), res[j]);
-      res[j] = cj.pmadd(lhs2[j], pfirst(ptmp2), res[j]);
-      res[j] = cj.pmadd(lhs3[j], pfirst(ptmp3), res[j]);
+      res[j] = cj.pmadd(lhs0(j), pfirst(ptmp0), res[j]);
+      res[j] = cj.pmadd(lhs1(j), pfirst(ptmp1), res[j]);
+      res[j] = cj.pmadd(lhs2(j), pfirst(ptmp2), res[j]);
+      res[j] = cj.pmadd(lhs3(j), pfirst(ptmp3), res[j]);
     }
   }
 
@@ -246,27 +287,27 @@
   {
     for (Index k=start; k<end; ++k)
     {
-      RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs[k*rhsIncr]);
-      const LhsScalar* lhs0 = lhs + k*lhsStride;
+      RhsPacket ptmp0 = pset1<RhsPacket>(alpha*rhs(k, 0));
+      const LhsScalars lhs0 = lhs.getVectorMapper(0, k);
 
       if (Vectorizable)
       {
         /* explicit vectorization */
         // process first unaligned result's coeffs
         for (Index j=0; j<alignedStart; ++j)
-          res[j] += cj.pmul(lhs0[j], pfirst(ptmp0));
+          res[j] += cj.pmul(lhs0(j), pfirst(ptmp0));
         // process aligned result's coeffs
-        if ((size_t(lhs0+alignedStart)%sizeof(LhsPacket))==0)
+        if (lhs0.template aligned<LhsPacket>(alignedStart))
           for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
-            pstore(&res[i], pcj.pmadd(pload<LhsPacket>(&lhs0[i]), ptmp0, pload<ResPacket>(&res[i])));
+            pstore(&res[i], pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(i), ptmp0, pload<ResPacket>(&res[i])));
         else
           for (Index i = alignedStart;i<alignedSize;i+=ResPacketSize)
-            pstore(&res[i], pcj.pmadd(ploadu<LhsPacket>(&lhs0[i]), ptmp0, pload<ResPacket>(&res[i])));
+            pstore(&res[i], pcj.pmadd(lhs0.template load<LhsPacket, Unaligned>(i), ptmp0, pload<ResPacket>(&res[i])));
       }
 
       // process remaining scalars (or all if no explicit vectorization)
       for (Index i=alignedSize; i<size; ++i)
-        res[i] += cj.pmul(lhs0[i], pfirst(ptmp0));
+        res[i] += cj.pmul(lhs0(i), pfirst(ptmp0));
     }
     if (skipColumns)
     {
@@ -290,10 +331,10 @@
  *  - alpha is always a complex (or converted to a complex)
  *  - no vectorization
  */
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-struct general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+struct general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>
 {
-typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
 
 enum {
   Vectorizable = packet_traits<LhsScalar>::Vectorizable && packet_traits<RhsScalar>::Vectorizable
@@ -310,73 +351,84 @@
 typedef typename conditional<Vectorizable,_LhsPacket,LhsScalar>::type LhsPacket;
 typedef typename conditional<Vectorizable,_RhsPacket,RhsScalar>::type RhsPacket;
 typedef typename conditional<Vectorizable,_ResPacket,ResScalar>::type ResPacket;
-  
+
 EIGEN_DONT_INLINE static void run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
-  ResScalar* res, Index resIncr,
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
+        ResScalar* res, Index resIncr,
   ResScalar alpha);
 };
 
-template<typename Index, typename LhsScalar, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version>
-EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjugateLhs,RhsScalar,ConjugateRhs,Version>::run(
+template<typename Index, typename LhsScalar, typename LhsMapper, bool ConjugateLhs, typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version>
+EIGEN_DONT_INLINE void general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjugateLhs,RhsScalar,RhsMapper,ConjugateRhs,Version>::run(
   Index rows, Index cols,
-  const LhsScalar* lhs, Index lhsStride,
-  const RhsScalar* rhs, Index rhsIncr,
+  const LhsMapper& lhs,
+  const RhsMapper& rhs,
   ResScalar* res, Index resIncr,
   ResScalar alpha)
 {
-  EIGEN_UNUSED_VARIABLE(rhsIncr);
-  eigen_internal_assert(rhsIncr==1);
+  eigen_internal_assert(rhs.stride()==1);
+
   #ifdef _EIGEN_ACCUMULATE_PACKETS
   #error _EIGEN_ACCUMULATE_PACKETS has already been defined
   #endif
 
-  #define _EIGEN_ACCUMULATE_PACKETS(A0,A13,A2) {\
-    RhsPacket b = pload<RhsPacket>(&rhs[j]); \
-    ptmp0 = pcj.pmadd(EIGEN_CAT(ploa,A0) <LhsPacket>(&lhs0[j]), b, ptmp0); \
-    ptmp1 = pcj.pmadd(EIGEN_CAT(ploa,A13)<LhsPacket>(&lhs1[j]), b, ptmp1); \
-    ptmp2 = pcj.pmadd(EIGEN_CAT(ploa,A2) <LhsPacket>(&lhs2[j]), b, ptmp2); \
-    ptmp3 = pcj.pmadd(EIGEN_CAT(ploa,A13)<LhsPacket>(&lhs3[j]), b, ptmp3); }
+  #define _EIGEN_ACCUMULATE_PACKETS(Alignment0,Alignment13,Alignment2) {\
+    RhsPacket b = rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0);  \
+    ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Alignment0>(j), b, ptmp0); \
+    ptmp1 = pcj.pmadd(lhs1.template load<LhsPacket, Alignment13>(j), b, ptmp1); \
+    ptmp2 = pcj.pmadd(lhs2.template load<LhsPacket, Alignment2>(j), b, ptmp2); \
+    ptmp3 = pcj.pmadd(lhs3.template load<LhsPacket, Alignment13>(j), b, ptmp3); }
 
   conj_helper<LhsScalar,RhsScalar,ConjugateLhs,ConjugateRhs> cj;
   conj_helper<LhsPacket,RhsPacket,ConjugateLhs,ConjugateRhs> pcj;
 
+  typedef typename LhsMapper::VectorMapper LhsScalars;
+
   enum { AllAligned=0, EvenAligned=1, FirstAligned=2, NoneAligned=3 };
   const Index rowsAtOnce = 4;
   const Index peels = 2;
   const Index RhsPacketAlignedMask = RhsPacketSize-1;
   const Index LhsPacketAlignedMask = LhsPacketSize-1;
-//   const Index PeelAlignedMask = RhsPacketSize*peels-1;
   const Index depth = cols;
+  const Index lhsStride = lhs.stride();
 
   // How many coeffs of the result do we have to skip to be aligned.
   // Here we assume data are at least aligned on the base scalar type
   // if that's not the case then vectorization is discarded, see below.
-  Index alignedStart = internal::first_aligned(rhs, depth);
+  Index alignedStart = rhs.firstAligned(depth);
   Index alignedSize = RhsPacketSize>1 ? alignedStart + ((depth-alignedStart) & ~RhsPacketAlignedMask) : 0;
   const Index peeledSize = alignedSize - RhsPacketSize*peels - RhsPacketSize + 1;
 
   const Index alignmentStep = LhsPacketSize>1 ? (LhsPacketSize - lhsStride % LhsPacketSize) & LhsPacketAlignedMask : 0;
   Index alignmentPattern = alignmentStep==0 ? AllAligned
-                         : alignmentStep==(LhsPacketSize/2) ? EvenAligned
-                         : FirstAligned;
+                           : alignmentStep==(LhsPacketSize/2) ? EvenAligned
+                           : FirstAligned;
 
   // we cannot assume the first element is aligned because of sub-matrices
-  const Index lhsAlignmentOffset = internal::first_aligned(lhs,depth);
+  const Index lhsAlignmentOffset = lhs.firstAligned(depth);
+  const Index rhsAlignmentOffset = rhs.firstAligned(rows);
 
   // find how many rows do we have to skip to be aligned with rhs (if possible)
   Index skipRows = 0;
   // if the data cannot be aligned (TODO add some compile time tests when possible, e.g. for floats)
-  if( (sizeof(LhsScalar)!=sizeof(RhsScalar)) || (size_t(lhs)%sizeof(LhsScalar)) || (size_t(rhs)%sizeof(RhsScalar)) )
+  if( (sizeof(LhsScalar)!=sizeof(RhsScalar)) ||
+      (lhsAlignmentOffset < 0) || (lhsAlignmentOffset == depth) ||
+      (rhsAlignmentOffset < 0) || (rhsAlignmentOffset == rows) )
   {
     alignedSize = 0;
     alignedStart = 0;
+    alignmentPattern = NoneAligned;
+  }
+  else if(LhsPacketSize > 4)
+  {
+    // TODO: extend the code to support aligned loads whenever possible when LhsPacketSize > 4.
+    alignmentPattern = NoneAligned;
   }
   else if (LhsPacketSize>1)
   {
-    eigen_internal_assert(size_t(lhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0  || depth<LhsPacketSize);
+  //    eigen_internal_assert(size_t(firstLhs+lhsAlignmentOffset)%sizeof(LhsPacket)==0  || depth<LhsPacketSize);
 
     while (skipRows<LhsPacketSize &&
            alignedStart != ((lhsAlignmentOffset + alignmentStep*skipRows)%LhsPacketSize))
@@ -392,11 +444,11 @@
       skipRows = (std::min)(skipRows,Index(rows));
       // note that the skiped columns are processed later.
     }
-    eigen_internal_assert(  alignmentPattern==NoneAligned
+    /*    eigen_internal_assert(  alignmentPattern==NoneAligned
                       || LhsPacketSize==1
                       || (skipRows + rowsAtOnce >= rows)
                       || LhsPacketSize > depth
-                      || (size_t(lhs+alignedStart+lhsStride*skipRows)%sizeof(LhsPacket))==0);
+                      || (size_t(firstLhs+alignedStart+lhsStride*skipRows)%sizeof(LhsPacket))==0);*/
   }
   else if(Vectorizable)
   {
@@ -405,18 +457,19 @@
     alignmentPattern = AllAligned;
   }
 
-  Index offset1 = (FirstAligned && alignmentStep==1?3:1);
-  Index offset3 = (FirstAligned && alignmentStep==1?1:3);
+  const Index offset1 = (alignmentPattern==FirstAligned && alignmentStep==1)?3:1;
+  const Index offset3 = (alignmentPattern==FirstAligned && alignmentStep==1)?1:3;
 
   Index rowBound = ((rows-skipRows)/rowsAtOnce)*rowsAtOnce + skipRows;
   for (Index i=skipRows; i<rowBound; i+=rowsAtOnce)
   {
-    EIGEN_ALIGN16 ResScalar tmp0 = ResScalar(0);
+    // FIXME: what is the purpose of this EIGEN_ALIGN_DEFAULT ??
+    EIGEN_ALIGN_MAX ResScalar tmp0 = ResScalar(0);
     ResScalar tmp1 = ResScalar(0), tmp2 = ResScalar(0), tmp3 = ResScalar(0);
 
     // this helps the compiler generating good binary code
-    const LhsScalar *lhs0 = lhs + i*lhsStride,     *lhs1 = lhs + (i+offset1)*lhsStride,
-                    *lhs2 = lhs + (i+2)*lhsStride, *lhs3 = lhs + (i+offset3)*lhsStride;
+    const LhsScalars lhs0 = lhs.getVectorMapper(i+0, 0),    lhs1 = lhs.getVectorMapper(i+offset1, 0),
+                     lhs2 = lhs.getVectorMapper(i+2, 0),    lhs3 = lhs.getVectorMapper(i+offset3, 0);
 
     if (Vectorizable)
     {
@@ -428,9 +481,9 @@
       // FIXME this loop get vectorized by the compiler !
       for (Index j=0; j<alignedStart; ++j)
       {
-        RhsScalar b = rhs[j];
-        tmp0 += cj.pmul(lhs0[j],b); tmp1 += cj.pmul(lhs1[j],b);
-        tmp2 += cj.pmul(lhs2[j],b); tmp3 += cj.pmul(lhs3[j],b);
+        RhsScalar b = rhs(j, 0);
+        tmp0 += cj.pmul(lhs0(j),b); tmp1 += cj.pmul(lhs1(j),b);
+        tmp2 += cj.pmul(lhs2(j),b); tmp3 += cj.pmul(lhs3(j),b);
       }
 
       if (alignedSize>alignedStart)
@@ -439,11 +492,11 @@
         {
           case AllAligned:
             for (Index j = alignedStart; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,d,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Aligned,Aligned);
             break;
           case EvenAligned:
             for (Index j = alignedStart; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,d);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Aligned);
             break;
           case FirstAligned:
           {
@@ -457,39 +510,39 @@
                * than basic unaligned loads.
                */
               LhsPacket A01, A02, A03, A11, A12, A13;
-              A01 = pload<LhsPacket>(&lhs1[alignedStart-1]);
-              A02 = pload<LhsPacket>(&lhs2[alignedStart-2]);
-              A03 = pload<LhsPacket>(&lhs3[alignedStart-3]);
+              A01 = lhs1.template load<LhsPacket, Aligned>(alignedStart-1);
+              A02 = lhs2.template load<LhsPacket, Aligned>(alignedStart-2);
+              A03 = lhs3.template load<LhsPacket, Aligned>(alignedStart-3);
 
               for (; j<peeledSize; j+=peels*RhsPacketSize)
               {
-                RhsPacket b = pload<RhsPacket>(&rhs[j]);
-                A11 = pload<LhsPacket>(&lhs1[j-1+LhsPacketSize]);  palign<1>(A01,A11);
-                A12 = pload<LhsPacket>(&lhs2[j-2+LhsPacketSize]);  palign<2>(A02,A12);
-                A13 = pload<LhsPacket>(&lhs3[j-3+LhsPacketSize]);  palign<3>(A03,A13);
+                RhsPacket b = rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0);
+                A11 = lhs1.template load<LhsPacket, Aligned>(j-1+LhsPacketSize);  palign<1>(A01,A11);
+                A12 = lhs2.template load<LhsPacket, Aligned>(j-2+LhsPacketSize);  palign<2>(A02,A12);
+                A13 = lhs3.template load<LhsPacket, Aligned>(j-3+LhsPacketSize);  palign<3>(A03,A13);
 
-                ptmp0 = pcj.pmadd(pload<LhsPacket>(&lhs0[j]), b, ptmp0);
+                ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(j), b, ptmp0);
                 ptmp1 = pcj.pmadd(A01, b, ptmp1);
-                A01 = pload<LhsPacket>(&lhs1[j-1+2*LhsPacketSize]);  palign<1>(A11,A01);
+                A01 = lhs1.template load<LhsPacket, Aligned>(j-1+2*LhsPacketSize);  palign<1>(A11,A01);
                 ptmp2 = pcj.pmadd(A02, b, ptmp2);
-                A02 = pload<LhsPacket>(&lhs2[j-2+2*LhsPacketSize]);  palign<2>(A12,A02);
+                A02 = lhs2.template load<LhsPacket, Aligned>(j-2+2*LhsPacketSize);  palign<2>(A12,A02);
                 ptmp3 = pcj.pmadd(A03, b, ptmp3);
-                A03 = pload<LhsPacket>(&lhs3[j-3+2*LhsPacketSize]);  palign<3>(A13,A03);
+                A03 = lhs3.template load<LhsPacket, Aligned>(j-3+2*LhsPacketSize);  palign<3>(A13,A03);
 
-                b = pload<RhsPacket>(&rhs[j+RhsPacketSize]);
-                ptmp0 = pcj.pmadd(pload<LhsPacket>(&lhs0[j+LhsPacketSize]), b, ptmp0);
+                b = rhs.getVectorMapper(j+RhsPacketSize, 0).template load<RhsPacket, Aligned>(0);
+                ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(j+LhsPacketSize), b, ptmp0);
                 ptmp1 = pcj.pmadd(A11, b, ptmp1);
                 ptmp2 = pcj.pmadd(A12, b, ptmp2);
                 ptmp3 = pcj.pmadd(A13, b, ptmp3);
               }
             }
             for (; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(d,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Aligned,Unaligned,Unaligned);
             break;
           }
           default:
             for (Index j = alignedStart; j<alignedSize; j+=RhsPacketSize)
-              _EIGEN_ACCUMULATE_PACKETS(du,du,du);
+              _EIGEN_ACCUMULATE_PACKETS(Unaligned,Unaligned,Unaligned);
             break;
         }
         tmp0 += predux(ptmp0);
@@ -503,9 +556,9 @@
     // FIXME this loop get vectorized by the compiler !
     for (Index j=alignedSize; j<depth; ++j)
     {
-      RhsScalar b = rhs[j];
-      tmp0 += cj.pmul(lhs0[j],b); tmp1 += cj.pmul(lhs1[j],b);
-      tmp2 += cj.pmul(lhs2[j],b); tmp3 += cj.pmul(lhs3[j],b);
+      RhsScalar b = rhs(j, 0);
+      tmp0 += cj.pmul(lhs0(j),b); tmp1 += cj.pmul(lhs1(j),b);
+      tmp2 += cj.pmul(lhs2(j),b); tmp3 += cj.pmul(lhs3(j),b);
     }
     res[i*resIncr]            += alpha*tmp0;
     res[(i+offset1)*resIncr]  += alpha*tmp1;
@@ -520,30 +573,30 @@
   {
     for (Index i=start; i<end; ++i)
     {
-      EIGEN_ALIGN16 ResScalar tmp0 = ResScalar(0);
+      EIGEN_ALIGN_MAX ResScalar tmp0 = ResScalar(0);
       ResPacket ptmp0 = pset1<ResPacket>(tmp0);
-      const LhsScalar* lhs0 = lhs + i*lhsStride;
+      const LhsScalars lhs0 = lhs.getVectorMapper(i, 0);
       // process first unaligned result's coeffs
       // FIXME this loop get vectorized by the compiler !
       for (Index j=0; j<alignedStart; ++j)
-        tmp0 += cj.pmul(lhs0[j], rhs[j]);
+        tmp0 += cj.pmul(lhs0(j), rhs(j, 0));
 
       if (alignedSize>alignedStart)
       {
         // process aligned rhs coeffs
-        if ((size_t(lhs0+alignedStart)%sizeof(LhsPacket))==0)
+        if (lhs0.template aligned<LhsPacket>(alignedStart))
           for (Index j = alignedStart;j<alignedSize;j+=RhsPacketSize)
-            ptmp0 = pcj.pmadd(pload<LhsPacket>(&lhs0[j]), pload<RhsPacket>(&rhs[j]), ptmp0);
+            ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Aligned>(j), rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0), ptmp0);
         else
           for (Index j = alignedStart;j<alignedSize;j+=RhsPacketSize)
-            ptmp0 = pcj.pmadd(ploadu<LhsPacket>(&lhs0[j]), pload<RhsPacket>(&rhs[j]), ptmp0);
+            ptmp0 = pcj.pmadd(lhs0.template load<LhsPacket, Unaligned>(j), rhs.getVectorMapper(j, 0).template load<RhsPacket, Aligned>(0), ptmp0);
         tmp0 += predux(ptmp0);
       }
 
       // process remaining scalars
       // FIXME this loop get vectorized by the compiler !
       for (Index j=alignedSize; j<depth; ++j)
-        tmp0 += cj.pmul(lhs0[j], rhs[j]);
+        tmp0 += cj.pmul(lhs0(j), rhs(j, 0));
       res[i*resIncr] += alpha*tmp0;
     }
     if (skipRows)
diff --git a/Eigen/src/Core/products/GeneralMatrixVector_BLAS.h b/Eigen/src/Core/products/GeneralMatrixVector_BLAS.h
new file mode 100644
index 0000000..6e36c2b
--- /dev/null
+++ b/Eigen/src/Core/products/GeneralMatrixVector_BLAS.h
@@ -0,0 +1,136 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to BLAS F77
+ *   General matrix-vector product functionality based on ?GEMV.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_GENERAL_MATRIX_VECTOR_BLAS_H
+#define EIGEN_GENERAL_MATRIX_VECTOR_BLAS_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/**********************************************************************
+* This file implements general matrix-vector multiplication using BLAS
+* gemv function via partial specialization of
+* general_matrix_vector_product::run(..) method for float, double,
+* std::complex<float> and std::complex<double> types
+**********************************************************************/
+
+// gemv specialization
+
+template<typename Index, typename LhsScalar, int StorageOrder, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs>
+struct general_matrix_vector_product_gemv;
+
+#define EIGEN_BLAS_GEMV_SPECIALIZE(Scalar) \
+template<typename Index, bool ConjugateLhs, bool ConjugateRhs> \
+struct general_matrix_vector_product<Index,Scalar,const_blas_data_mapper<Scalar,Index,ColMajor>,ColMajor,ConjugateLhs,Scalar,const_blas_data_mapper<Scalar,Index,RowMajor>,ConjugateRhs,Specialized> { \
+static void run( \
+  Index rows, Index cols, \
+  const const_blas_data_mapper<Scalar,Index,ColMajor> &lhs, \
+  const const_blas_data_mapper<Scalar,Index,RowMajor> &rhs, \
+  Scalar* res, Index resIncr, Scalar alpha) \
+{ \
+  if (ConjugateLhs) { \
+    general_matrix_vector_product<Index,Scalar,const_blas_data_mapper<Scalar,Index,ColMajor>,ColMajor,ConjugateLhs,Scalar,const_blas_data_mapper<Scalar,Index,RowMajor>,ConjugateRhs,BuiltIn>::run( \
+      rows, cols, lhs, rhs, res, resIncr, alpha); \
+  } else { \
+    general_matrix_vector_product_gemv<Index,Scalar,ColMajor,ConjugateLhs,Scalar,ConjugateRhs>::run( \
+      rows, cols, lhs.data(), lhs.stride(), rhs.data(), rhs.stride(), res, resIncr, alpha); \
+  } \
+} \
+}; \
+template<typename Index, bool ConjugateLhs, bool ConjugateRhs> \
+struct general_matrix_vector_product<Index,Scalar,const_blas_data_mapper<Scalar,Index,RowMajor>,RowMajor,ConjugateLhs,Scalar,const_blas_data_mapper<Scalar,Index,ColMajor>,ConjugateRhs,Specialized> { \
+static void run( \
+  Index rows, Index cols, \
+  const const_blas_data_mapper<Scalar,Index,RowMajor> &lhs, \
+  const const_blas_data_mapper<Scalar,Index,ColMajor> &rhs, \
+  Scalar* res, Index resIncr, Scalar alpha) \
+{ \
+    general_matrix_vector_product_gemv<Index,Scalar,RowMajor,ConjugateLhs,Scalar,ConjugateRhs>::run( \
+      rows, cols, lhs.data(), lhs.stride(), rhs.data(), rhs.stride(), res, resIncr, alpha); \
+} \
+}; \
+
+EIGEN_BLAS_GEMV_SPECIALIZE(double)
+EIGEN_BLAS_GEMV_SPECIALIZE(float)
+EIGEN_BLAS_GEMV_SPECIALIZE(dcomplex)
+EIGEN_BLAS_GEMV_SPECIALIZE(scomplex)
+
+#define EIGEN_BLAS_GEMV_SPECIALIZATION(EIGTYPE,BLASTYPE,BLASFUNC) \
+template<typename Index, int LhsStorageOrder, bool ConjugateLhs, bool ConjugateRhs> \
+struct general_matrix_vector_product_gemv<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,ConjugateRhs> \
+{ \
+typedef Matrix<EIGTYPE,Dynamic,1,ColMajor> GEMVVector;\
+\
+static void run( \
+  Index rows, Index cols, \
+  const EIGTYPE* lhs, Index lhsStride, \
+  const EIGTYPE* rhs, Index rhsIncr, \
+  EIGTYPE* res, Index resIncr, EIGTYPE alpha) \
+{ \
+  BlasIndex m=convert_index<BlasIndex>(rows), n=convert_index<BlasIndex>(cols), \
+            lda=convert_index<BlasIndex>(lhsStride), incx=convert_index<BlasIndex>(rhsIncr), incy=convert_index<BlasIndex>(resIncr); \
+  const EIGTYPE beta(1); \
+  const EIGTYPE *x_ptr; \
+  char trans=(LhsStorageOrder==ColMajor) ? 'N' : (ConjugateLhs) ? 'C' : 'T'; \
+  if (LhsStorageOrder==RowMajor) { \
+    m = convert_index<BlasIndex>(cols); \
+    n = convert_index<BlasIndex>(rows); \
+  }\
+  GEMVVector x_tmp; \
+  if (ConjugateRhs) { \
+    Map<const GEMVVector, 0, InnerStride<> > map_x(rhs,cols,1,InnerStride<>(incx)); \
+    x_tmp=map_x.conjugate(); \
+    x_ptr=x_tmp.data(); \
+    incx=1; \
+  } else x_ptr=rhs; \
+  BLASFUNC(&trans, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)lhs, &lda, (const BLASTYPE*)x_ptr, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &incy); \
+}\
+};
+
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_GEMV_SPECIALIZATION(double,   double, dgemv)
+EIGEN_BLAS_GEMV_SPECIALIZATION(float,    float,  sgemv)
+EIGEN_BLAS_GEMV_SPECIALIZATION(dcomplex, MKL_Complex16, zgemv)
+EIGEN_BLAS_GEMV_SPECIALIZATION(scomplex, MKL_Complex8 , cgemv)
+#else
+EIGEN_BLAS_GEMV_SPECIALIZATION(double,   double, dgemv_)
+EIGEN_BLAS_GEMV_SPECIALIZATION(float,    float,  sgemv_)
+EIGEN_BLAS_GEMV_SPECIALIZATION(dcomplex, double, zgemv_)
+EIGEN_BLAS_GEMV_SPECIALIZATION(scomplex, float,  cgemv_)
+#endif
+
+} // end namespase internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_GENERAL_MATRIX_VECTOR_BLAS_H
diff --git a/Eigen/src/Core/products/GeneralMatrixVector_MKL.h b/Eigen/src/Core/products/GeneralMatrixVector_MKL.h
deleted file mode 100644
index 1cb9fe6..0000000
--- a/Eigen/src/Core/products/GeneralMatrixVector_MKL.h
+++ /dev/null
@@ -1,131 +0,0 @@
-/*
- Copyright (c) 2011, Intel Corporation. All rights reserved.
-
- Redistribution and use in source and binary forms, with or without modification,
- are permitted provided that the following conditions are met:
-
- * Redistributions of source code must retain the above copyright notice, this
-   list of conditions and the following disclaimer.
- * Redistributions in binary form must reproduce the above copyright notice,
-   this list of conditions and the following disclaimer in the documentation
-   and/or other materials provided with the distribution.
- * Neither the name of Intel Corporation nor the names of its contributors may
-   be used to endorse or promote products derived from this software without
-   specific prior written permission.
-
- THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
- ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
- WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
- DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
- ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
- (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
- LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
- ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
- ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
- *   General matrix-vector product functionality based on ?GEMV.
- ********************************************************************************
-*/
-
-#ifndef EIGEN_GENERAL_MATRIX_VECTOR_MKL_H
-#define EIGEN_GENERAL_MATRIX_VECTOR_MKL_H
-
-namespace Eigen { 
-
-namespace internal {
-
-/**********************************************************************
-* This file implements general matrix-vector multiplication using BLAS
-* gemv function via partial specialization of
-* general_matrix_vector_product::run(..) method for float, double,
-* std::complex<float> and std::complex<double> types
-**********************************************************************/
-
-// gemv specialization
-
-template<typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs>
-struct general_matrix_vector_product_gemv :
-  general_matrix_vector_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,ConjugateRhs,BuiltIn> {};
-
-#define EIGEN_MKL_GEMV_SPECIALIZE(Scalar) \
-template<typename Index, bool ConjugateLhs, bool ConjugateRhs> \
-struct general_matrix_vector_product<Index,Scalar,ColMajor,ConjugateLhs,Scalar,ConjugateRhs,Specialized> { \
-static void run( \
-  Index rows, Index cols, \
-  const Scalar* lhs, Index lhsStride, \
-  const Scalar* rhs, Index rhsIncr, \
-  Scalar* res, Index resIncr, Scalar alpha) \
-{ \
-  if (ConjugateLhs) { \
-    general_matrix_vector_product<Index,Scalar,ColMajor,ConjugateLhs,Scalar,ConjugateRhs,BuiltIn>::run( \
-      rows, cols, lhs, lhsStride, rhs, rhsIncr, res, resIncr, alpha); \
-  } else { \
-    general_matrix_vector_product_gemv<Index,Scalar,ColMajor,ConjugateLhs,Scalar,ConjugateRhs>::run( \
-      rows, cols, lhs, lhsStride, rhs, rhsIncr, res, resIncr, alpha); \
-  } \
-} \
-}; \
-template<typename Index, bool ConjugateLhs, bool ConjugateRhs> \
-struct general_matrix_vector_product<Index,Scalar,RowMajor,ConjugateLhs,Scalar,ConjugateRhs,Specialized> { \
-static void run( \
-  Index rows, Index cols, \
-  const Scalar* lhs, Index lhsStride, \
-  const Scalar* rhs, Index rhsIncr, \
-  Scalar* res, Index resIncr, Scalar alpha) \
-{ \
-    general_matrix_vector_product_gemv<Index,Scalar,RowMajor,ConjugateLhs,Scalar,ConjugateRhs>::run( \
-      rows, cols, lhs, lhsStride, rhs, rhsIncr, res, resIncr, alpha); \
-} \
-}; \
-
-EIGEN_MKL_GEMV_SPECIALIZE(double)
-EIGEN_MKL_GEMV_SPECIALIZE(float)
-EIGEN_MKL_GEMV_SPECIALIZE(dcomplex)
-EIGEN_MKL_GEMV_SPECIALIZE(scomplex)
-
-#define EIGEN_MKL_GEMV_SPECIALIZATION(EIGTYPE,MKLTYPE,MKLPREFIX) \
-template<typename Index, int LhsStorageOrder, bool ConjugateLhs, bool ConjugateRhs> \
-struct general_matrix_vector_product_gemv<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,ConjugateRhs> \
-{ \
-typedef Matrix<EIGTYPE,Dynamic,1,ColMajor> GEMVVector;\
-\
-static void run( \
-  Index rows, Index cols, \
-  const EIGTYPE* lhs, Index lhsStride, \
-  const EIGTYPE* rhs, Index rhsIncr, \
-  EIGTYPE* res, Index resIncr, EIGTYPE alpha) \
-{ \
-  MKL_INT m=rows, n=cols, lda=lhsStride, incx=rhsIncr, incy=resIncr; \
-  MKLTYPE alpha_, beta_; \
-  const EIGTYPE *x_ptr, myone(1); \
-  char trans=(LhsStorageOrder==ColMajor) ? 'N' : (ConjugateLhs) ? 'C' : 'T'; \
-  if (LhsStorageOrder==RowMajor) { \
-    m=cols; \
-    n=rows; \
-  }\
-  assign_scalar_eig2mkl(alpha_, alpha); \
-  assign_scalar_eig2mkl(beta_, myone); \
-  GEMVVector x_tmp; \
-  if (ConjugateRhs) { \
-    Map<const GEMVVector, 0, InnerStride<> > map_x(rhs,cols,1,InnerStride<>(incx)); \
-    x_tmp=map_x.conjugate(); \
-    x_ptr=x_tmp.data(); \
-    incx=1; \
-  } else x_ptr=rhs; \
-  MKLPREFIX##gemv(&trans, &m, &n, &alpha_, (const MKLTYPE*)lhs, &lda, (const MKLTYPE*)x_ptr, &incx, &beta_, (MKLTYPE*)res, &incy); \
-}\
-};
-
-EIGEN_MKL_GEMV_SPECIALIZATION(double,   double,        d)
-EIGEN_MKL_GEMV_SPECIALIZATION(float,    float,         s)
-EIGEN_MKL_GEMV_SPECIALIZATION(dcomplex, MKL_Complex16, z)
-EIGEN_MKL_GEMV_SPECIALIZATION(scomplex, MKL_Complex8,  c)
-
-} // end namespase internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_GENERAL_MATRIX_VECTOR_MKL_H
diff --git a/Eigen/src/Core/products/Parallelizer.h b/Eigen/src/Core/products/Parallelizer.h
index 6937ee3..c2f084c 100644
--- a/Eigen/src/Core/products/Parallelizer.h
+++ b/Eigen/src/Core/products/Parallelizer.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_PARALLELIZER_H
 #define EIGEN_PARALLELIZER_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -49,8 +49,8 @@
 {
   int nbt;
   internal::manage_multi_threading(GetAction, &nbt);
-  std::ptrdiff_t l1, l2;
-  internal::manage_caching_sizes(GetAction, &l1, &l2);
+  std::ptrdiff_t l1, l2, l3;
+  internal::manage_caching_sizes(GetAction, &l1, &l2, &l3);
 }
 
 /** \returns the max number of threads reserved for Eigen
@@ -73,17 +73,17 @@
 
 template<typename Index> struct GemmParallelInfo
 {
-  GemmParallelInfo() : sync(-1), users(0), rhs_start(0), rhs_length(0) {}
+  GemmParallelInfo() : sync(-1), users(0), lhs_start(0), lhs_length(0) {}
 
-  int volatile sync;
+  Index volatile sync;
   int volatile users;
 
-  Index rhs_start;
-  Index rhs_length;
+  Index lhs_start;
+  Index lhs_length;
 };
 
 template<bool Condition, typename Functor, typename Index>
-void parallelize_gemm(const Functor& func, Index rows, Index cols, bool transpose)
+void parallelize_gemm(const Functor& func, Index rows, Index cols, Index depth, bool transpose)
 {
   // TODO when EIGEN_USE_BLAS is defined,
   // we should still enable OMP for other scalar types
@@ -92,6 +92,7 @@
   // the matrix product when multithreading is enabled. This is a temporary
   // fix to support row-major destination matrices. This whole
   // parallelizer mechanism has to be redisigned anyway.
+  EIGEN_UNUSED_VARIABLE(depth);
   EIGEN_UNUSED_VARIABLE(transpose);
   func(0,rows, 0,cols);
 #else
@@ -102,56 +103,56 @@
   // - we are not already in a parallel code
   // - the sizes are large enough
 
-  // 1- are we already in a parallel session?
+  // compute the maximal number of threads from the size of the product:
+  // This first heuristic takes into account that the product kernel is fully optimized when working with nr columns at once.
+  Index size = transpose ? rows : cols;
+  Index pb_max_threads = std::max<Index>(1,size / Functor::Traits::nr);
+
+  // compute the maximal number of threads from the total amount of work:
+  double work = static_cast<double>(rows) * static_cast<double>(cols) *
+      static_cast<double>(depth);
+  double kMinTaskSize = 50000;  // FIXME improve this heuristic.
+  pb_max_threads = std::max<Index>(1, std::min<Index>(pb_max_threads, work / kMinTaskSize));
+
+  // compute the number of threads we are going to use
+  Index threads = std::min<Index>(nbThreads(), pb_max_threads);
+
+  // if multi-threading is explicitely disabled, not useful, or if we already are in a parallel session,
+  // then abort multi-threading
   // FIXME omp_get_num_threads()>1 only works for openmp, what if the user does not use openmp?
-  if((!Condition) || (omp_get_num_threads()>1))
-    return func(0,rows, 0,cols);
-
-  Index size = transpose ? cols : rows;
-
-  // 2- compute the maximal number of threads from the size of the product:
-  // FIXME this has to be fine tuned
-  Index max_threads = std::max<Index>(1,size / 32);
-
-  // 3 - compute the number of threads we are going to use
-  Index threads = std::min<Index>(nbThreads(), max_threads);
-
-  if(threads==1)
+  if((!Condition) || (threads==1) || (omp_get_num_threads()>1))
     return func(0,rows, 0,cols);
 
   Eigen::initParallel();
-  func.initParallelSession();
+  func.initParallelSession(threads);
 
   if(transpose)
     std::swap(rows,cols);
 
-  GemmParallelInfo<Index>* info = new GemmParallelInfo<Index>[threads];
+  ei_declare_aligned_stack_constructed_variable(GemmParallelInfo<Index>,info,threads,0);
 
   #pragma omp parallel num_threads(threads)
   {
     Index i = omp_get_thread_num();
     // Note that the actual number of threads might be lower than the number of request ones.
     Index actual_threads = omp_get_num_threads();
-    
+
     Index blockCols = (cols / actual_threads) & ~Index(0x3);
-    Index blockRows = (rows / actual_threads) & ~Index(0x7);
-    
+    Index blockRows = (rows / actual_threads);
+    blockRows = (blockRows/Functor::Traits::mr)*Functor::Traits::mr;
+
     Index r0 = i*blockRows;
     Index actualBlockRows = (i+1==actual_threads) ? rows-r0 : blockRows;
 
     Index c0 = i*blockCols;
     Index actualBlockCols = (i+1==actual_threads) ? cols-c0 : blockCols;
 
-    info[i].rhs_start = c0;
-    info[i].rhs_length = actualBlockCols;
+    info[i].lhs_start = r0;
+    info[i].lhs_length = actualBlockRows;
 
-    if(transpose)
-      func(0, cols, r0, actualBlockRows, info);
-    else
-      func(r0, actualBlockRows, 0,cols, info);
+    if(transpose) func(c0, actualBlockCols, 0, rows, info);
+    else          func(0, rows, c0, actualBlockCols, info);
   }
-
-  delete[] info;
 #endif
 }
 
diff --git a/Eigen/src/Core/products/SelfadjointMatrixMatrix.h b/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
index 99cf9e0..da6f82a 100644
--- a/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
+++ b/Eigen/src/Core/products/SelfadjointMatrixMatrix.h
@@ -15,7 +15,7 @@
 namespace internal {
 
 // pack a selfadjoint block diagonal for use with the gebp_kernel
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder>
+template<typename Scalar, typename Index, int Pack1, int Pack2_dummy, int StorageOrder>
 struct symm_pack_lhs
 {
   template<int BlockRows> inline
@@ -45,25 +45,32 @@
   }
   void operator()(Scalar* blockA, const Scalar* _lhs, Index lhsStride, Index cols, Index rows)
   {
+    enum { PacketSize = packet_traits<Scalar>::size };
     const_blas_data_mapper<Scalar,Index,StorageOrder> lhs(_lhs,lhsStride);
     Index count = 0;
-    Index peeled_mc = (rows/Pack1)*Pack1;
-    for(Index i=0; i<peeled_mc; i+=Pack1)
-    {
-      pack<Pack1>(blockA, lhs, cols, i, count);
-    }
-
-    if(rows-peeled_mc>=Pack2)
-    {
-      pack<Pack2>(blockA, lhs, cols, peeled_mc, count);
-      peeled_mc += Pack2;
-    }
+    //Index peeled_mc3 = (rows/Pack1)*Pack1;
+    
+    const Index peeled_mc3 = Pack1>=3*PacketSize ? (rows/(3*PacketSize))*(3*PacketSize) : 0;
+    const Index peeled_mc2 = Pack1>=2*PacketSize ? peeled_mc3+((rows-peeled_mc3)/(2*PacketSize))*(2*PacketSize) : 0;
+    const Index peeled_mc1 = Pack1>=1*PacketSize ? (rows/(1*PacketSize))*(1*PacketSize) : 0;
+    
+    if(Pack1>=3*PacketSize)
+      for(Index i=0; i<peeled_mc3; i+=3*PacketSize)
+        pack<3*PacketSize>(blockA, lhs, cols, i, count);
+    
+    if(Pack1>=2*PacketSize)
+      for(Index i=peeled_mc3; i<peeled_mc2; i+=2*PacketSize)
+        pack<2*PacketSize>(blockA, lhs, cols, i, count);
+    
+    if(Pack1>=1*PacketSize)
+      for(Index i=peeled_mc2; i<peeled_mc1; i+=1*PacketSize)
+        pack<1*PacketSize>(blockA, lhs, cols, i, count);
 
     // do the same with mr==1
-    for(Index i=peeled_mc; i<rows; i++)
+    for(Index i=peeled_mc1; i<rows; i++)
     {
       for(Index k=0; k<i; k++)
-        blockA[count++] = lhs(i, k);              // normal
+        blockA[count++] = lhs(i, k);                   // normal
 
       blockA[count++] = numext::real(lhs(i, i));       // real (diagonal)
 
@@ -82,7 +89,8 @@
     Index end_k = k2 + rows;
     Index count = 0;
     const_blas_data_mapper<Scalar,Index,StorageOrder> rhs(_rhs,rhsStride);
-    Index packet_cols = (cols/nr)*nr;
+    Index packet_cols8 = nr>=8 ? (cols/8) * 8 : 0;
+    Index packet_cols4 = nr>=4 ? (cols/4) * 4 : 0;
 
     // first part: normal case
     for(Index j2=0; j2<k2; j2+=nr)
@@ -91,79 +99,151 @@
       {
         blockB[count+0] = rhs(k,j2+0);
         blockB[count+1] = rhs(k,j2+1);
-        if (nr==4)
+        if (nr>=4)
         {
           blockB[count+2] = rhs(k,j2+2);
           blockB[count+3] = rhs(k,j2+3);
         }
+        if (nr>=8)
+        {
+          blockB[count+4] = rhs(k,j2+4);
+          blockB[count+5] = rhs(k,j2+5);
+          blockB[count+6] = rhs(k,j2+6);
+          blockB[count+7] = rhs(k,j2+7);
+        }
         count += nr;
       }
     }
 
     // second part: diagonal block
-    for(Index j2=k2; j2<(std::min)(k2+rows,packet_cols); j2+=nr)
+    Index end8 = nr>=8 ? (std::min)(k2+rows,packet_cols8) : k2;
+    if(nr>=8)
     {
-      // again we can split vertically in three different parts (transpose, symmetric, normal)
-      // transpose
-      for(Index k=k2; k<j2; k++)
+      for(Index j2=k2; j2<end8; j2+=8)
       {
-        blockB[count+0] = numext::conj(rhs(j2+0,k));
-        blockB[count+1] = numext::conj(rhs(j2+1,k));
-        if (nr==4)
+        // again we can split vertically in three different parts (transpose, symmetric, normal)
+        // transpose
+        for(Index k=k2; k<j2; k++)
         {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
           blockB[count+2] = numext::conj(rhs(j2+2,k));
           blockB[count+3] = numext::conj(rhs(j2+3,k));
+          blockB[count+4] = numext::conj(rhs(j2+4,k));
+          blockB[count+5] = numext::conj(rhs(j2+5,k));
+          blockB[count+6] = numext::conj(rhs(j2+6,k));
+          blockB[count+7] = numext::conj(rhs(j2+7,k));
+          count += 8;
         }
-        count += nr;
-      }
-      // symmetric
-      Index h = 0;
-      for(Index k=j2; k<j2+nr; k++)
-      {
-        // normal
-        for (Index w=0 ; w<h; ++w)
-          blockB[count+w] = rhs(k,j2+w);
-
-        blockB[count+h] = numext::real(rhs(k,k));
-
-        // transpose
-        for (Index w=h+1 ; w<nr; ++w)
-          blockB[count+w] = numext::conj(rhs(j2+w,k));
-        count += nr;
-        ++h;
-      }
-      // normal
-      for(Index k=j2+nr; k<end_k; k++)
-      {
-        blockB[count+0] = rhs(k,j2+0);
-        blockB[count+1] = rhs(k,j2+1);
-        if (nr==4)
+        // symmetric
+        Index h = 0;
+        for(Index k=j2; k<j2+8; k++)
         {
+          // normal
+          for (Index w=0 ; w<h; ++w)
+            blockB[count+w] = rhs(k,j2+w);
+
+          blockB[count+h] = numext::real(rhs(k,k));
+
+          // transpose
+          for (Index w=h+1 ; w<8; ++w)
+            blockB[count+w] = numext::conj(rhs(j2+w,k));
+          count += 8;
+          ++h;
+        }
+        // normal
+        for(Index k=j2+8; k<end_k; k++)
+        {
+          blockB[count+0] = rhs(k,j2+0);
+          blockB[count+1] = rhs(k,j2+1);
           blockB[count+2] = rhs(k,j2+2);
           blockB[count+3] = rhs(k,j2+3);
+          blockB[count+4] = rhs(k,j2+4);
+          blockB[count+5] = rhs(k,j2+5);
+          blockB[count+6] = rhs(k,j2+6);
+          blockB[count+7] = rhs(k,j2+7);
+          count += 8;
         }
-        count += nr;
+      }
+    }
+    if(nr>=4)
+    {
+      for(Index j2=end8; j2<(std::min)(k2+rows,packet_cols4); j2+=4)
+      {
+        // again we can split vertically in three different parts (transpose, symmetric, normal)
+        // transpose
+        for(Index k=k2; k<j2; k++)
+        {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
+          blockB[count+2] = numext::conj(rhs(j2+2,k));
+          blockB[count+3] = numext::conj(rhs(j2+3,k));
+          count += 4;
+        }
+        // symmetric
+        Index h = 0;
+        for(Index k=j2; k<j2+4; k++)
+        {
+          // normal
+          for (Index w=0 ; w<h; ++w)
+            blockB[count+w] = rhs(k,j2+w);
+
+          blockB[count+h] = numext::real(rhs(k,k));
+
+          // transpose
+          for (Index w=h+1 ; w<4; ++w)
+            blockB[count+w] = numext::conj(rhs(j2+w,k));
+          count += 4;
+          ++h;
+        }
+        // normal
+        for(Index k=j2+4; k<end_k; k++)
+        {
+          blockB[count+0] = rhs(k,j2+0);
+          blockB[count+1] = rhs(k,j2+1);
+          blockB[count+2] = rhs(k,j2+2);
+          blockB[count+3] = rhs(k,j2+3);
+          count += 4;
+        }
       }
     }
 
     // third part: transposed
-    for(Index j2=k2+rows; j2<packet_cols; j2+=nr)
+    if(nr>=8)
     {
-      for(Index k=k2; k<end_k; k++)
+      for(Index j2=k2+rows; j2<packet_cols8; j2+=8)
       {
-        blockB[count+0] = numext::conj(rhs(j2+0,k));
-        blockB[count+1] = numext::conj(rhs(j2+1,k));
-        if (nr==4)
+        for(Index k=k2; k<end_k; k++)
         {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
           blockB[count+2] = numext::conj(rhs(j2+2,k));
           blockB[count+3] = numext::conj(rhs(j2+3,k));
+          blockB[count+4] = numext::conj(rhs(j2+4,k));
+          blockB[count+5] = numext::conj(rhs(j2+5,k));
+          blockB[count+6] = numext::conj(rhs(j2+6,k));
+          blockB[count+7] = numext::conj(rhs(j2+7,k));
+          count += 8;
         }
-        count += nr;
+      }
+    }
+    if(nr>=4)
+    {
+      for(Index j2=(std::max)(packet_cols8,k2+rows); j2<packet_cols4; j2+=4)
+      {
+        for(Index k=k2; k<end_k; k++)
+        {
+          blockB[count+0] = numext::conj(rhs(j2+0,k));
+          blockB[count+1] = numext::conj(rhs(j2+1,k));
+          blockB[count+2] = numext::conj(rhs(j2+2,k));
+          blockB[count+3] = numext::conj(rhs(j2+3,k));
+          count += 4;
+        }
       }
     }
 
     // copy the remaining columns one at a time (=> the same with nr==1)
-    for(Index j2=packet_cols; j2<cols; ++j2)
+    for(Index j2=packet_cols4; j2<cols; ++j2)
     {
       // transpose
       Index half = (std::min)(end_k,j2);
@@ -211,7 +291,7 @@
     const Scalar* lhs, Index lhsStride,
     const Scalar* rhs, Index rhsStride,
     Scalar* res,       Index resStride,
-    const Scalar& alpha)
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     product_selfadjoint_matrix<Scalar, Index,
       EIGEN_LOGICAL_XOR(RhsSelfAdjoint,RhsStorageOrder==RowMajor) ? ColMajor : RowMajor,
@@ -219,7 +299,7 @@
       EIGEN_LOGICAL_XOR(LhsSelfAdjoint,LhsStorageOrder==RowMajor) ? ColMajor : RowMajor,
       LhsSelfAdjoint, NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsSelfAdjoint,ConjugateLhs),
       ColMajor>
-      ::run(cols, rows,  rhs, rhsStride,  lhs, lhsStride,  res, resStride,  alpha);
+      ::run(cols, rows,  rhs, rhsStride,  lhs, lhsStride,  res, resStride,  alpha, blocking);
   }
 };
 
@@ -234,7 +314,7 @@
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
     Scalar* res,        Index resStride,
-    const Scalar& alpha);
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
 };
 
 template <typename Scalar, typename Index,
@@ -244,33 +324,35 @@
     Index rows, Index cols,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
-    const Scalar& alpha)
+    Scalar* _res,        Index resStride,
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     Index size = rows;
 
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<Scalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
-
     typedef gebp_traits<Scalar,Scalar> Traits;
 
-    Index kc = size;  // cache block size along the K direction
-    Index mc = rows;  // cache block size along the M direction
-    Index nc = cols;  // cache block size along the N direction
-    computeProductBlockingSizes<Scalar,Scalar>(kc, mc, nc);
-    // kc must smaller than mc
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, (LhsStorageOrder == RowMajor) ? ColMajor : RowMajor> LhsTransposeMapper;
+    typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    LhsTransposeMapper lhs_transpose(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
+
+    Index kc = blocking.kc();                   // cache block size along the K direction
+    Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
+    // kc must be smaller than mc
     kc = (std::min)(kc,mc);
+    std::size_t sizeA = kc*mc;
+    std::size_t sizeB = kc*cols;
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
 
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    std::size_t sizeB = sizeW + kc*cols;
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, kc*mc, 0);
-    ei_declare_aligned_stack_constructed_variable(Scalar, allocatedBlockB, sizeB, 0);
-    Scalar* blockB = allocatedBlockB + sizeW;
-
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
     symm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder==RowMajor?ColMajor:RowMajor, true> pack_lhs_transposed;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder> pack_rhs;
+    gemm_pack_lhs<Scalar, Index, LhsTransposeMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder==RowMajor?ColMajor:RowMajor, true> pack_lhs_transposed;
 
     for(Index k2=0; k2<size; k2+=kc)
     {
@@ -279,7 +361,7 @@
       // we have selected one row panel of rhs and one column panel of lhs
       // pack rhs's panel into a sequential chunk of memory
       // and expand each coeff to a constant packet for further reuse
-      pack_rhs(blockB, &rhs(k2,0), rhsStride, actual_kc, cols);
+      pack_rhs(blockB, rhs.getSubMapper(k2,0), actual_kc, cols);
 
       // the select lhs's panel has to be split in three different parts:
       //  1 - the transposed panel above the diagonal block => transposed packed copy
@@ -289,9 +371,9 @@
       {
         const Index actual_mc = (std::min)(i2+mc,k2)-i2;
         // transposed packed copy
-        pack_lhs_transposed(blockA, &lhs(k2, i2), lhsStride, actual_kc, actual_mc);
+        pack_lhs_transposed(blockA, lhs_transpose.getSubMapper(i2, k2), actual_kc, actual_mc);
 
-        gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
       // the block diagonal
       {
@@ -299,16 +381,16 @@
         // symmetric packed copy
         pack_lhs(blockA, &lhs(k2,k2), lhsStride, actual_kc, actual_mc);
 
-        gebp_kernel(res+k2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(k2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
 
       for(Index i2=k2+kc; i2<size; i2+=mc)
       {
         const Index actual_mc = (std::min)(i2+mc,size)-i2;
-        gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder,false>()
-          (blockA, &lhs(i2, k2), lhsStride, actual_kc, actual_mc);
+        gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder,false>()
+          (blockA, lhs.getSubMapper(i2, k2), actual_kc, actual_mc);
 
-        gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
     }
   }
@@ -325,7 +407,7 @@
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
     Scalar* res,        Index resStride,
-    const Scalar& alpha);
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking);
 };
 
 template <typename Scalar, typename Index,
@@ -335,27 +417,27 @@
     Index rows, Index cols,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
-    const Scalar& alpha)
+    Scalar* _res,        Index resStride,
+    const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     Index size = cols;
 
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-
     typedef gebp_traits<Scalar,Scalar> Traits;
 
-    Index kc = size; // cache block size along the K direction
-    Index mc = rows;  // cache block size along the M direction
-    Index nc = cols;  // cache block size along the N direction
-    computeProductBlockingSizes<Scalar,Scalar>(kc, mc, nc);
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
-    std::size_t sizeB = sizeW + kc*cols;
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, kc*mc, 0);
-    ei_declare_aligned_stack_constructed_variable(Scalar, allocatedBlockB, sizeB, 0);
-    Scalar* blockB = allocatedBlockB + sizeW;
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    ResMapper res(_res,resStride);
 
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    Index kc = blocking.kc();                   // cache block size along the K direction
+    Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
+    std::size_t sizeA = kc*mc;
+    std::size_t sizeB = kc*cols;
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
+    ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
+
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
     symm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
 
     for(Index k2=0; k2<size; k2+=kc)
@@ -368,9 +450,9 @@
       for(Index i2=0; i2<rows; i2+=mc)
       {
         const Index actual_mc = (std::min)(i2+mc,rows)-i2;
-        pack_lhs(blockA, &lhs(i2, k2), lhsStride, actual_kc, actual_mc);
+        pack_lhs(blockA, lhs.getSubMapper(i2, k2), actual_kc, actual_mc);
 
-        gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha);
+        gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, alpha);
       }
     }
   }
@@ -382,55 +464,58 @@
 ***************************************************************************/
 
 namespace internal {
+  
 template<typename Lhs, int LhsMode, typename Rhs, int RhsMode>
-struct traits<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false> >
-  : traits<ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>, Lhs, Rhs> >
-{};
-}
-
-template<typename Lhs, int LhsMode, typename Rhs, int RhsMode>
-struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>
-  : public ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,RhsMode,false>, Lhs, Rhs >
+struct selfadjoint_product_impl<Lhs,LhsMode,false,Rhs,RhsMode,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
-
-  SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  typedef internal::blas_traits<Lhs> LhsBlasTraits;
+  typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+  typedef internal::blas_traits<Rhs> RhsBlasTraits;
+  typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  
   enum {
     LhsIsUpper = (LhsMode&(Upper|Lower))==Upper,
     LhsIsSelfAdjoint = (LhsMode&SelfAdjoint)==SelfAdjoint,
     RhsIsUpper = (RhsMode&(Upper|Lower))==Upper,
     RhsIsSelfAdjoint = (RhsMode&SelfAdjoint)==SelfAdjoint
   };
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  
+  template<typename Dest>
+  static void run(Dest &dst, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
   {
-    eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+    eigen_assert(dst.rows()==a_lhs.rows() && dst.cols()==a_rhs.cols());
 
-    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
 
-    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                               * RhsBlasTraits::extractScalarFactor(m_rhs);
+    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
+                               * RhsBlasTraits::extractScalarFactor(a_rhs);
+
+    typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
+              Lhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxColsAtCompileTime,1> BlockingType;
+
+    BlockingType blocking(lhs.rows(), rhs.cols(), lhs.cols(), 1, false);
 
     internal::product_selfadjoint_matrix<Scalar, Index,
-      EIGEN_LOGICAL_XOR(LhsIsUpper,
-                        internal::traits<Lhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, LhsIsSelfAdjoint,
+      EIGEN_LOGICAL_XOR(LhsIsUpper,internal::traits<Lhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, LhsIsSelfAdjoint,
       NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(LhsIsUpper,bool(LhsBlasTraits::NeedToConjugate)),
-      EIGEN_LOGICAL_XOR(RhsIsUpper,
-                        internal::traits<Rhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, RhsIsSelfAdjoint,
+      EIGEN_LOGICAL_XOR(RhsIsUpper,internal::traits<Rhs>::Flags &RowMajorBit) ? RowMajor : ColMajor, RhsIsSelfAdjoint,
       NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(RhsIsUpper,bool(RhsBlasTraits::NeedToConjugate)),
       internal::traits<Dest>::Flags&RowMajorBit  ? RowMajor : ColMajor>
       ::run(
         lhs.rows(), rhs.cols(),                 // sizes
-        &lhs.coeffRef(0,0),    lhs.outerStride(),  // lhs info
-        &rhs.coeffRef(0,0),    rhs.outerStride(),  // rhs info
+        &lhs.coeffRef(0,0), lhs.outerStride(),  // lhs info
+        &rhs.coeffRef(0,0), rhs.outerStride(),  // rhs info
         &dst.coeffRef(0,0), dst.outerStride(),  // result info
-        actualAlpha                             // alpha
+        actualAlpha, blocking                   // alpha
       );
   }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SELFADJOINT_MATRIX_MATRIX_H
diff --git a/Eigen/src/Core/products/SelfadjointMatrixMatrix_MKL.h b/Eigen/src/Core/products/SelfadjointMatrixMatrix_BLAS.h
similarity index 65%
rename from Eigen/src/Core/products/SelfadjointMatrixMatrix_MKL.h
rename to Eigen/src/Core/products/SelfadjointMatrixMatrix_BLAS.h
index dfa687f..9a53185 100644
--- a/Eigen/src/Core/products/SelfadjointMatrixMatrix_MKL.h
+++ b/Eigen/src/Core/products/SelfadjointMatrixMatrix_BLAS.h
@@ -25,13 +25,13 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 //
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to BLAS F77
  *   Self adjoint matrix * matrix product functionality based on ?SYMM/?HEMM.
  ********************************************************************************
 */
 
-#ifndef EIGEN_SELFADJOINT_MATRIX_MATRIX_MKL_H
-#define EIGEN_SELFADJOINT_MATRIX_MATRIX_MKL_H
+#ifndef EIGEN_SELFADJOINT_MATRIX_MATRIX_BLAS_H
+#define EIGEN_SELFADJOINT_MATRIX_MATRIX_BLAS_H
 
 namespace Eigen { 
 
@@ -40,7 +40,7 @@
 
 /* Optimized selfadjoint matrix * matrix (?SYMM/?HEMM) product */
 
-#define EIGEN_MKL_SYMM_L(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_SYMM_L(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
 template <typename Index, \
           int LhsStorageOrder, bool ConjugateLhs, \
           int RhsStorageOrder, bool ConjugateRhs> \
@@ -52,28 +52,23 @@
     const EIGTYPE* _lhs, Index lhsStride, \
     const EIGTYPE* _rhs, Index rhsStride, \
     EIGTYPE* res,        Index resStride, \
-    EIGTYPE alpha) \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
   { \
     char side='L', uplo='L'; \
-    MKL_INT m, n, lda, ldb, ldc; \
+    BlasIndex m, n, lda, ldb, ldc; \
     const EIGTYPE *a, *b; \
-    MKLTYPE alpha_, beta_; \
+    EIGTYPE beta(1); \
     MatrixX##EIGPREFIX b_tmp; \
-    EIGTYPE myone(1);\
 \
 /* Set transpose options */ \
 /* Set m, n, k */ \
-    m = (MKL_INT)rows;  \
-    n = (MKL_INT)cols;  \
-\
-/* Set alpha_ & beta_ */ \
-    assign_scalar_eig2mkl(alpha_, alpha); \
-    assign_scalar_eig2mkl(beta_, myone); \
+    m = convert_index<BlasIndex>(rows);  \
+    n = convert_index<BlasIndex>(cols);  \
 \
 /* Set lda, ldb, ldc */ \
-    lda = (MKL_INT)lhsStride; \
-    ldb = (MKL_INT)rhsStride; \
-    ldc = (MKL_INT)resStride; \
+    lda = convert_index<BlasIndex>(lhsStride); \
+    ldb = convert_index<BlasIndex>(rhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
 \
 /* Set a, b, c */ \
     if (LhsStorageOrder==RowMajor) uplo='U'; \
@@ -83,16 +78,16 @@
       Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > rhs(_rhs,n,m,OuterStride<>(rhsStride)); \
       b_tmp = rhs.adjoint(); \
       b = b_tmp.data(); \
-      ldb = b_tmp.outerStride(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
     } else b = _rhs; \
 \
-    MKLPREFIX##symm(&side, &uplo, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)b, &ldb, &beta_, (MKLTYPE*)res, &ldc); \
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
 \
   } \
 };
 
 
-#define EIGEN_MKL_HEMM_L(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_HEMM_L(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
 template <typename Index, \
           int LhsStorageOrder, bool ConjugateLhs, \
           int RhsStorageOrder, bool ConjugateRhs> \
@@ -103,36 +98,31 @@
     const EIGTYPE* _lhs, Index lhsStride, \
     const EIGTYPE* _rhs, Index rhsStride, \
     EIGTYPE* res,        Index resStride, \
-    EIGTYPE alpha) \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
   { \
     char side='L', uplo='L'; \
-    MKL_INT m, n, lda, ldb, ldc; \
+    BlasIndex m, n, lda, ldb, ldc; \
     const EIGTYPE *a, *b; \
-    MKLTYPE alpha_, beta_; \
+    EIGTYPE beta(1); \
     MatrixX##EIGPREFIX b_tmp; \
     Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> a_tmp; \
-    EIGTYPE myone(1); \
 \
 /* Set transpose options */ \
 /* Set m, n, k */ \
-    m = (MKL_INT)rows; \
-    n = (MKL_INT)cols; \
-\
-/* Set alpha_ & beta_ */ \
-    assign_scalar_eig2mkl(alpha_, alpha); \
-    assign_scalar_eig2mkl(beta_, myone); \
+    m = convert_index<BlasIndex>(rows); \
+    n = convert_index<BlasIndex>(cols); \
 \
 /* Set lda, ldb, ldc */ \
-    lda = (MKL_INT)lhsStride; \
-    ldb = (MKL_INT)rhsStride; \
-    ldc = (MKL_INT)resStride; \
+    lda = convert_index<BlasIndex>(lhsStride); \
+    ldb = convert_index<BlasIndex>(rhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
 \
 /* Set a, b, c */ \
     if (((LhsStorageOrder==ColMajor) && ConjugateLhs) || ((LhsStorageOrder==RowMajor) && (!ConjugateLhs))) { \
       Map<const Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder>, 0, OuterStride<> > lhs(_lhs,m,m,OuterStride<>(lhsStride)); \
       a_tmp = lhs.conjugate(); \
       a = a_tmp.data(); \
-      lda = a_tmp.outerStride(); \
+      lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
     } else a = _lhs; \
     if (LhsStorageOrder==RowMajor) uplo='U'; \
 \
@@ -151,23 +141,29 @@
         b_tmp = rhs.transpose(); \
       } \
       b = b_tmp.data(); \
-      ldb = b_tmp.outerStride(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
     } \
 \
-    MKLPREFIX##hemm(&side, &uplo, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)b, &ldb, &beta_, (MKLTYPE*)res, &ldc); \
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
 \
   } \
 };
 
-EIGEN_MKL_SYMM_L(double, double, d, d)
-EIGEN_MKL_SYMM_L(float, float, f, s)
-EIGEN_MKL_HEMM_L(dcomplex, MKL_Complex16, cd, z)
-EIGEN_MKL_HEMM_L(scomplex, MKL_Complex8, cf, c)
-
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_SYMM_L(double, double, d, dsymm)
+EIGEN_BLAS_SYMM_L(float, float, f, ssymm)
+EIGEN_BLAS_HEMM_L(dcomplex, MKL_Complex16, cd, zhemm)
+EIGEN_BLAS_HEMM_L(scomplex, MKL_Complex8, cf, chemm)
+#else
+EIGEN_BLAS_SYMM_L(double, double, d, dsymm_)
+EIGEN_BLAS_SYMM_L(float, float, f, ssymm_)
+EIGEN_BLAS_HEMM_L(dcomplex, double, cd, zhemm_)
+EIGEN_BLAS_HEMM_L(scomplex, float, cf, chemm_)
+#endif
 
 /* Optimized matrix * selfadjoint matrix (?SYMM/?HEMM) product */
 
-#define EIGEN_MKL_SYMM_R(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_SYMM_R(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
 template <typename Index, \
           int LhsStorageOrder, bool ConjugateLhs, \
           int RhsStorageOrder, bool ConjugateRhs> \
@@ -179,27 +175,22 @@
     const EIGTYPE* _lhs, Index lhsStride, \
     const EIGTYPE* _rhs, Index rhsStride, \
     EIGTYPE* res,        Index resStride, \
-    EIGTYPE alpha) \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
   { \
     char side='R', uplo='L'; \
-    MKL_INT m, n, lda, ldb, ldc; \
+    BlasIndex m, n, lda, ldb, ldc; \
     const EIGTYPE *a, *b; \
-    MKLTYPE alpha_, beta_; \
+    EIGTYPE beta(1); \
     MatrixX##EIGPREFIX b_tmp; \
-    EIGTYPE myone(1);\
 \
 /* Set m, n, k */ \
-    m = (MKL_INT)rows;  \
-    n = (MKL_INT)cols;  \
-\
-/* Set alpha_ & beta_ */ \
-    assign_scalar_eig2mkl(alpha_, alpha); \
-    assign_scalar_eig2mkl(beta_, myone); \
+    m = convert_index<BlasIndex>(rows);  \
+    n = convert_index<BlasIndex>(cols);  \
 \
 /* Set lda, ldb, ldc */ \
-    lda = (MKL_INT)rhsStride; \
-    ldb = (MKL_INT)lhsStride; \
-    ldc = (MKL_INT)resStride; \
+    lda = convert_index<BlasIndex>(rhsStride); \
+    ldb = convert_index<BlasIndex>(lhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
 \
 /* Set a, b, c */ \
     if (RhsStorageOrder==RowMajor) uplo='U'; \
@@ -209,16 +200,16 @@
       Map<const MatrixX##EIGPREFIX, 0, OuterStride<> > lhs(_lhs,n,m,OuterStride<>(rhsStride)); \
       b_tmp = lhs.adjoint(); \
       b = b_tmp.data(); \
-      ldb = b_tmp.outerStride(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
     } else b = _lhs; \
 \
-    MKLPREFIX##symm(&side, &uplo, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)b, &ldb, &beta_, (MKLTYPE*)res, &ldc); \
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
 \
   } \
 };
 
 
-#define EIGEN_MKL_HEMM_R(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_HEMM_R(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
 template <typename Index, \
           int LhsStorageOrder, bool ConjugateLhs, \
           int RhsStorageOrder, bool ConjugateRhs> \
@@ -229,35 +220,30 @@
     const EIGTYPE* _lhs, Index lhsStride, \
     const EIGTYPE* _rhs, Index rhsStride, \
     EIGTYPE* res,        Index resStride, \
-    EIGTYPE alpha) \
+    EIGTYPE alpha, level3_blocking<EIGTYPE, EIGTYPE>& /*blocking*/) \
   { \
     char side='R', uplo='L'; \
-    MKL_INT m, n, lda, ldb, ldc; \
+    BlasIndex m, n, lda, ldb, ldc; \
     const EIGTYPE *a, *b; \
-    MKLTYPE alpha_, beta_; \
+    EIGTYPE beta(1); \
     MatrixX##EIGPREFIX b_tmp; \
     Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> a_tmp; \
-    EIGTYPE myone(1); \
 \
 /* Set m, n, k */ \
-    m = (MKL_INT)rows; \
-    n = (MKL_INT)cols; \
-\
-/* Set alpha_ & beta_ */ \
-    assign_scalar_eig2mkl(alpha_, alpha); \
-    assign_scalar_eig2mkl(beta_, myone); \
+    m = convert_index<BlasIndex>(rows); \
+    n = convert_index<BlasIndex>(cols); \
 \
 /* Set lda, ldb, ldc */ \
-    lda = (MKL_INT)rhsStride; \
-    ldb = (MKL_INT)lhsStride; \
-    ldc = (MKL_INT)resStride; \
+    lda = convert_index<BlasIndex>(rhsStride); \
+    ldb = convert_index<BlasIndex>(lhsStride); \
+    ldc = convert_index<BlasIndex>(resStride); \
 \
 /* Set a, b, c */ \
     if (((RhsStorageOrder==ColMajor) && ConjugateRhs) || ((RhsStorageOrder==RowMajor) && (!ConjugateRhs))) { \
       Map<const Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder>, 0, OuterStride<> > rhs(_rhs,n,n,OuterStride<>(rhsStride)); \
       a_tmp = rhs.conjugate(); \
       a = a_tmp.data(); \
-      lda = a_tmp.outerStride(); \
+      lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
     } else a = _rhs; \
     if (RhsStorageOrder==RowMajor) uplo='U'; \
 \
@@ -276,20 +262,26 @@
         b_tmp = lhs.transpose(); \
       } \
       b = b_tmp.data(); \
-      ldb = b_tmp.outerStride(); \
+      ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
     } \
 \
-    MKLPREFIX##hemm(&side, &uplo, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)b, &ldb, &beta_, (MKLTYPE*)res, &ldc); \
+    BLASFUNC(&side, &uplo, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)b, &ldb, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &ldc); \
   } \
 };
 
-EIGEN_MKL_SYMM_R(double, double, d, d)
-EIGEN_MKL_SYMM_R(float, float, f, s)
-EIGEN_MKL_HEMM_R(dcomplex, MKL_Complex16, cd, z)
-EIGEN_MKL_HEMM_R(scomplex, MKL_Complex8, cf, c)
-
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_SYMM_R(double, double, d, dsymm)
+EIGEN_BLAS_SYMM_R(float, float, f, ssymm)
+EIGEN_BLAS_HEMM_R(dcomplex, MKL_Complex16, cd, zhemm)
+EIGEN_BLAS_HEMM_R(scomplex, MKL_Complex8, cf, chemm)
+#else
+EIGEN_BLAS_SYMM_R(double, double, d, dsymm_)
+EIGEN_BLAS_SYMM_R(float, float, f, ssymm_)
+EIGEN_BLAS_HEMM_R(dcomplex, double, cd, zhemm_)
+EIGEN_BLAS_HEMM_R(scomplex, float, cf, chemm_)
+#endif
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_SELFADJOINT_MATRIX_MATRIX_MKL_H
+#endif // EIGEN_SELFADJOINT_MATRIX_MATRIX_BLAS_H
diff --git a/Eigen/src/Core/products/SelfadjointMatrixVector.h b/Eigen/src/Core/products/SelfadjointMatrixVector.h
index f698f67..3fd180e 100644
--- a/Eigen/src/Core/products/SelfadjointMatrixVector.h
+++ b/Eigen/src/Core/products/SelfadjointMatrixVector.h
@@ -30,7 +30,7 @@
 static EIGEN_DONT_INLINE void run(
   Index size,
   const Scalar*  lhs, Index lhsStride,
-  const Scalar* _rhs, Index rhsIncr,
+  const Scalar*  rhs,
   Scalar* res,
   Scalar alpha);
 };
@@ -39,11 +39,12 @@
 EIGEN_DONT_INLINE void selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,Version>::run(
   Index size,
   const Scalar*  lhs, Index lhsStride,
-  const Scalar* _rhs, Index rhsIncr,
+  const Scalar*  rhs,
   Scalar* res,
   Scalar alpha)
 {
   typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename NumTraits<Scalar>::Real RealScalar;
   const Index PacketSize = sizeof(Packet)/sizeof(Scalar);
 
   enum {
@@ -54,23 +55,13 @@
 
   conj_helper<Scalar,Scalar,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs,  IsRowMajor), ConjugateRhs> cj0;
   conj_helper<Scalar,Scalar,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs, !IsRowMajor), ConjugateRhs> cj1;
-  conj_helper<Scalar,Scalar,NumTraits<Scalar>::IsComplex, ConjugateRhs> cjd;
+  conj_helper<RealScalar,Scalar,false, ConjugateRhs> cjd;
 
   conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs,  IsRowMajor), ConjugateRhs> pcj0;
   conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex && EIGEN_LOGICAL_XOR(ConjugateLhs, !IsRowMajor), ConjugateRhs> pcj1;
 
   Scalar cjAlpha = ConjugateRhs ? numext::conj(alpha) : alpha;
 
-  // FIXME this copy is now handled outside product_selfadjoint_vector, so it could probably be removed.
-  // if the rhs is not sequentially stored in memory we copy it to a temporary buffer,
-  // this is because we need to extract packets
-  ei_declare_aligned_stack_constructed_variable(Scalar,rhs,size,rhsIncr==1 ? const_cast<Scalar*>(_rhs) : 0);  
-  if (rhsIncr!=1)
-  {
-    const Scalar* it = _rhs;
-    for (Index i=0; i<size; ++i, it+=rhsIncr)
-      rhs[i] = *it;
-  }
 
   Index bound = (std::max)(Index(0),size-8) & 0xfffffffe;
   if (FirstTriangular)
@@ -92,12 +83,11 @@
     Scalar t3(0);
     Packet ptmp3 = pset1<Packet>(t3);
 
-    size_t starti = FirstTriangular ? 0 : j+2;
-    size_t endi   = FirstTriangular ? j : size;
-    size_t alignedStart = (starti) + internal::first_aligned(&res[starti], endi-starti);
-    size_t alignedEnd = alignedStart + ((endi-alignedStart)/(PacketSize))*(PacketSize);
+    Index starti = FirstTriangular ? 0 : j+2;
+    Index endi   = FirstTriangular ? j : size;
+    Index alignedStart = (starti) + internal::first_default_aligned(&res[starti], endi-starti);
+    Index alignedEnd = alignedStart + ((endi-alignedStart)/(PacketSize))*(PacketSize);
 
-    // TODO make sure this product is a real * complex and that the rhs is properly conjugated if needed
     res[j]   += cjd.pmul(numext::real(A0[j]), t0);
     res[j+1] += cjd.pmul(numext::real(A1[j+1]), t1);
     if(FirstTriangular)
@@ -111,11 +101,11 @@
       t2 += cj1.pmul(A0[j+1], rhs[j+1]);
     }
 
-    for (size_t i=starti; i<alignedStart; ++i)
+    for (Index i=starti; i<alignedStart; ++i)
     {
-      res[i] += t0 * A0[i] + t1 * A1[i];
-      t2 += numext::conj(A0[i]) * rhs[i];
-      t3 += numext::conj(A1[i]) * rhs[i];
+      res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
+      t2 += cj1.pmul(A0[i], rhs[i]);
+      t3 += cj1.pmul(A1[i], rhs[i]);
     }
     // Yes this an optimization for gcc 4.3 and 4.4 (=> huge speed up)
     // gcc 4.2 does this optimization automatically.
@@ -123,7 +113,7 @@
     const Scalar* EIGEN_RESTRICT a1It  = A1  + alignedStart;
     const Scalar* EIGEN_RESTRICT rhsIt = rhs + alignedStart;
           Scalar* EIGEN_RESTRICT resIt = res + alignedStart;
-    for (size_t i=alignedStart; i<alignedEnd; i+=PacketSize)
+    for (Index i=alignedStart; i<alignedEnd; i+=PacketSize)
     {
       Packet A0i = ploadu<Packet>(a0It);  a0It  += PacketSize;
       Packet A1i = ploadu<Packet>(a1It);  a1It  += PacketSize;
@@ -135,7 +125,7 @@
       ptmp3 = pcj1.pmadd(A1i,  Bi, ptmp3);
       pstore(resIt,Xi); resIt += PacketSize;
     }
-    for (size_t i=alignedEnd; i<endi; i++)
+    for (Index i=alignedEnd; i<endi; i++)
     {
       res[i] += cj0.pmul(A0[i], t0) + cj0.pmul(A1[i],t1);
       t2 += cj1.pmul(A0[i], rhs[i]);
@@ -151,7 +141,6 @@
 
     Scalar t1 = cjAlpha * rhs[j];
     Scalar t2(0);
-    // TODO make sure this product is a real * complex and that the rhs is properly conjugated if needed
     res[j] += cjd.pmul(numext::real(A0[j]), t1);
     for (Index i=FirstTriangular ? 0 : j+1; i<(FirstTriangular ? j : size); i++)
     {
@@ -169,45 +158,44 @@
 ***************************************************************************/
 
 namespace internal {
-template<typename Lhs, int LhsMode, typename Rhs>
-struct traits<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true> >
-  : traits<ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>, Lhs, Rhs> >
-{};
-}
 
 template<typename Lhs, int LhsMode, typename Rhs>
-struct SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>
-  : public ProductBase<SelfadjointProductMatrix<Lhs,LhsMode,false,Rhs,0,true>, Lhs, Rhs >
+struct selfadjoint_product_impl<Lhs,LhsMode,false,Rhs,0,true>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  typedef internal::blas_traits<Lhs> LhsBlasTraits;
+  typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+  typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
+  
+  typedef internal::blas_traits<Rhs> RhsBlasTraits;
+  typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+  typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
 
-  enum {
-    LhsUpLo = LhsMode&(Upper|Lower)
-  };
+  enum { LhsUpLo = LhsMode&(Upper|Lower) };
 
-  SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+  template<typename Dest>
+  static void run(Dest& dest, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
   {
     typedef typename Dest::Scalar ResScalar;
-    typedef typename Base::RhsScalar RhsScalar;
-    typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
+    typedef typename Rhs::Scalar RhsScalar;
+    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
     
-    eigen_assert(dest.rows()==m_lhs.rows() && dest.cols()==m_rhs.cols());
+    eigen_assert(dest.rows()==a_lhs.rows() && dest.cols()==a_rhs.cols());
 
-    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
 
-    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                               * RhsBlasTraits::extractScalarFactor(m_rhs);
+    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs)
+                               * RhsBlasTraits::extractScalarFactor(a_rhs);
 
     enum {
       EvalToDest = (Dest::InnerStrideAtCompileTime==1),
-      UseRhs = (_ActualRhsType::InnerStrideAtCompileTime==1)
+      UseRhs = (ActualRhsTypeCleaned::InnerStrideAtCompileTime==1)
     };
     
     internal::gemv_static_vector_if<ResScalar,Dest::SizeAtCompileTime,Dest::MaxSizeAtCompileTime,!EvalToDest> static_dest;
-    internal::gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!UseRhs> static_rhs;
+    internal::gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!UseRhs> static_rhs;
 
     ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
                                                   EvalToDest ? dest.data() : static_dest.data());
@@ -218,7 +206,7 @@
     if(!EvalToDest)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = dest.size();
+      Index size = dest.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
       MappedDest(actualDestPtr, dest.size()) = dest;
@@ -227,18 +215,19 @@
     if(!UseRhs)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = rhs.size();
+      Index size = rhs.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
-      Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, rhs.size()) = rhs;
+      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, rhs.size()) = rhs;
     }
       
       
-    internal::selfadjoint_matrix_vector_product<Scalar, Index, (internal::traits<_ActualLhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, int(LhsUpLo), bool(LhsBlasTraits::NeedToConjugate), bool(RhsBlasTraits::NeedToConjugate)>::run
+    internal::selfadjoint_matrix_vector_product<Scalar, Index, (internal::traits<ActualLhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor,
+                                                int(LhsUpLo), bool(LhsBlasTraits::NeedToConjugate), bool(RhsBlasTraits::NeedToConjugate)>::run
       (
         lhs.rows(),                             // size
         &lhs.coeffRef(0,0),  lhs.outerStride(), // lhs info
-        actualRhsPtr, 1,                        // rhs info
+        actualRhsPtr,                           // rhs info
         actualDestPtr,                          // result info
         actualAlpha                             // scale factor
       );
@@ -248,34 +237,24 @@
   }
 };
 
-namespace internal {
 template<typename Lhs, typename Rhs, int RhsMode>
-struct traits<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false> >
-  : traits<ProductBase<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>, Lhs, Rhs> >
-{};
-}
-
-template<typename Lhs, typename Rhs, int RhsMode>
-struct SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>
-  : public ProductBase<SelfadjointProductMatrix<Lhs,0,true,Rhs,RhsMode,false>, Lhs, Rhs >
+struct selfadjoint_product_impl<Lhs,0,true,Rhs,RhsMode,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(SelfadjointProductMatrix)
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  enum { RhsUpLo = RhsMode&(Upper|Lower)  };
 
-  enum {
-    RhsUpLo = RhsMode&(Upper|Lower)
-  };
-
-  SelfadjointProductMatrix(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+  template<typename Dest>
+  static void run(Dest& dest, const Lhs &a_lhs, const Rhs &a_rhs, const Scalar& alpha)
   {
     // let's simply transpose the product
     Transpose<Dest> destT(dest);
-    SelfadjointProductMatrix<Transpose<const Rhs>, int(RhsUpLo)==Upper ? Lower : Upper, false,
-                             Transpose<const Lhs>, 0, true>(m_rhs.transpose(), m_lhs.transpose()).scaleAndAddTo(destT, alpha);
+    selfadjoint_product_impl<Transpose<const Rhs>, int(RhsUpLo)==Upper ? Lower : Upper, false,
+                             Transpose<const Lhs>, 0, true>::run(destT, a_rhs.transpose(), a_lhs.transpose(), alpha);
   }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SELFADJOINT_MATRIX_VECTOR_H
diff --git a/Eigen/src/Core/products/SelfadjointMatrixVector_MKL.h b/Eigen/src/Core/products/SelfadjointMatrixVector_BLAS.h
similarity index 70%
rename from Eigen/src/Core/products/SelfadjointMatrixVector_MKL.h
rename to Eigen/src/Core/products/SelfadjointMatrixVector_BLAS.h
index 86684b6..1238345 100644
--- a/Eigen/src/Core/products/SelfadjointMatrixVector_MKL.h
+++ b/Eigen/src/Core/products/SelfadjointMatrixVector_BLAS.h
@@ -25,13 +25,13 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to BLAS F77
  *   Selfadjoint matrix-vector product functionality based on ?SYMV/HEMV.
  ********************************************************************************
 */
 
-#ifndef EIGEN_SELFADJOINT_MATRIX_VECTOR_MKL_H
-#define EIGEN_SELFADJOINT_MATRIX_VECTOR_MKL_H
+#ifndef EIGEN_SELFADJOINT_MATRIX_VECTOR_BLAS_H
+#define EIGEN_SELFADJOINT_MATRIX_VECTOR_BLAS_H
 
 namespace Eigen { 
 
@@ -47,31 +47,31 @@
 struct selfadjoint_matrix_vector_product_symv :
   selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,BuiltIn> {};
 
-#define EIGEN_MKL_SYMV_SPECIALIZE(Scalar) \
+#define EIGEN_BLAS_SYMV_SPECIALIZE(Scalar) \
 template<typename Index, int StorageOrder, int UpLo, bool ConjugateLhs, bool ConjugateRhs> \
 struct selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,Specialized> { \
 static void run( \
   Index size, const Scalar*  lhs, Index lhsStride, \
-  const Scalar* _rhs, Index rhsIncr, Scalar* res, Scalar alpha) { \
+  const Scalar* _rhs, Scalar* res, Scalar alpha) { \
     enum {\
       IsColMajor = StorageOrder==ColMajor \
     }; \
     if (IsColMajor == ConjugateLhs) {\
       selfadjoint_matrix_vector_product<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs,BuiltIn>::run( \
-        size, lhs, lhsStride, _rhs, rhsIncr, res, alpha);  \
+        size, lhs, lhsStride, _rhs, res, alpha);  \
     } else {\
       selfadjoint_matrix_vector_product_symv<Scalar,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs>::run( \
-        size, lhs, lhsStride, _rhs, rhsIncr, res, alpha);  \
+        size, lhs, lhsStride, _rhs, res, alpha);  \
     }\
   } \
 }; \
 
-EIGEN_MKL_SYMV_SPECIALIZE(double)
-EIGEN_MKL_SYMV_SPECIALIZE(float)
-EIGEN_MKL_SYMV_SPECIALIZE(dcomplex)
-EIGEN_MKL_SYMV_SPECIALIZE(scomplex)
+EIGEN_BLAS_SYMV_SPECIALIZE(double)
+EIGEN_BLAS_SYMV_SPECIALIZE(float)
+EIGEN_BLAS_SYMV_SPECIALIZE(dcomplex)
+EIGEN_BLAS_SYMV_SPECIALIZE(scomplex)
 
-#define EIGEN_MKL_SYMV_SPECIALIZATION(EIGTYPE,MKLTYPE,MKLFUNC) \
+#define EIGEN_BLAS_SYMV_SPECIALIZATION(EIGTYPE,BLASTYPE,BLASFUNC) \
 template<typename Index, int StorageOrder, int UpLo, bool ConjugateLhs, bool ConjugateRhs> \
 struct selfadjoint_matrix_vector_product_symv<EIGTYPE,Index,StorageOrder,UpLo,ConjugateLhs,ConjugateRhs> \
 { \
@@ -79,36 +79,40 @@
 \
 static void run( \
 Index size, const EIGTYPE*  lhs, Index lhsStride, \
-const EIGTYPE* _rhs, Index rhsIncr, EIGTYPE* res, EIGTYPE alpha) \
+const EIGTYPE* _rhs, EIGTYPE* res, EIGTYPE alpha) \
 { \
   enum {\
     IsRowMajor = StorageOrder==RowMajor ? 1 : 0, \
     IsLower = UpLo == Lower ? 1 : 0 \
   }; \
-  MKL_INT n=size, lda=lhsStride, incx=rhsIncr, incy=1; \
-  MKLTYPE alpha_, beta_; \
-  const EIGTYPE *x_ptr, myone(1); \
+  BlasIndex n=convert_index<BlasIndex>(size), lda=convert_index<BlasIndex>(lhsStride), incx=1, incy=1; \
+  EIGTYPE beta(1); \
+  const EIGTYPE *x_ptr; \
   char uplo=(IsRowMajor) ? (IsLower ? 'U' : 'L') : (IsLower ? 'L' : 'U'); \
-  assign_scalar_eig2mkl(alpha_, alpha); \
-  assign_scalar_eig2mkl(beta_, myone); \
   SYMVVector x_tmp; \
   if (ConjugateRhs) { \
-    Map<const SYMVVector, 0, InnerStride<> > map_x(_rhs,size,1,InnerStride<>(incx)); \
+    Map<const SYMVVector, 0 > map_x(_rhs,size,1); \
     x_tmp=map_x.conjugate(); \
     x_ptr=x_tmp.data(); \
-    incx=1; \
   } else x_ptr=_rhs; \
-  MKLFUNC(&uplo, &n, &alpha_, (const MKLTYPE*)lhs, &lda, (const MKLTYPE*)x_ptr, &incx, &beta_, (MKLTYPE*)res, &incy); \
+  BLASFUNC(&uplo, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)lhs, &lda, (const BLASTYPE*)x_ptr, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)res, &incy); \
 }\
 };
 
-EIGEN_MKL_SYMV_SPECIALIZATION(double,   double,        dsymv)
-EIGEN_MKL_SYMV_SPECIALIZATION(float,    float,         ssymv)
-EIGEN_MKL_SYMV_SPECIALIZATION(dcomplex, MKL_Complex16, zhemv)
-EIGEN_MKL_SYMV_SPECIALIZATION(scomplex, MKL_Complex8,  chemv)
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_SYMV_SPECIALIZATION(double,   double, dsymv)
+EIGEN_BLAS_SYMV_SPECIALIZATION(float,    float,  ssymv)
+EIGEN_BLAS_SYMV_SPECIALIZATION(dcomplex, MKL_Complex16, zhemv)
+EIGEN_BLAS_SYMV_SPECIALIZATION(scomplex, MKL_Complex8,  chemv)
+#else
+EIGEN_BLAS_SYMV_SPECIALIZATION(double,   double, dsymv_)
+EIGEN_BLAS_SYMV_SPECIALIZATION(float,    float,  ssymv_)
+EIGEN_BLAS_SYMV_SPECIALIZATION(dcomplex, double, zhemv_)
+EIGEN_BLAS_SYMV_SPECIALIZATION(scomplex, float,  chemv_)
+#endif
 
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_SELFADJOINT_MATRIX_VECTOR_MKL_H
+#endif // EIGEN_SELFADJOINT_MATRIX_VECTOR_BLAS_H
diff --git a/Eigen/src/Core/products/SelfadjointProduct.h b/Eigen/src/Core/products/SelfadjointProduct.h
index 6ca4ae6..f038d68 100644
--- a/Eigen/src/Core/products/SelfadjointProduct.h
+++ b/Eigen/src/Core/products/SelfadjointProduct.h
@@ -53,7 +53,6 @@
   static void run(MatrixType& mat, const OtherType& other, const typename MatrixType::Scalar& alpha)
   {
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
     typedef internal::blas_traits<OtherType> OtherBlasTraits;
     typedef typename OtherBlasTraits::DirectLinearAccessType ActualOtherType;
     typedef typename internal::remove_all<ActualOtherType>::type _ActualOtherType;
@@ -86,7 +85,6 @@
   static void run(MatrixType& mat, const OtherType& other, const typename MatrixType::Scalar& alpha)
   {
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
     typedef internal::blas_traits<OtherType> OtherBlasTraits;
     typedef typename OtherBlasTraits::DirectLinearAccessType ActualOtherType;
     typedef typename internal::remove_all<ActualOtherType>::type _ActualOtherType;
@@ -94,15 +92,27 @@
 
     Scalar actualAlpha = alpha * OtherBlasTraits::extractScalarFactor(other.derived());
 
-    enum { IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0 };
+    enum {
+      IsRowMajor = (internal::traits<MatrixType>::Flags&RowMajorBit) ? 1 : 0,
+      OtherIsRowMajor = _ActualOtherType::Flags&RowMajorBit ? 1 : 0
+    };
+
+    Index size = mat.cols();
+    Index depth = actualOther.cols();
+
+    typedef internal::gemm_blocking_space<IsRowMajor ? RowMajor : ColMajor,Scalar,Scalar,
+              MatrixType::MaxColsAtCompileTime, MatrixType::MaxColsAtCompileTime, _ActualOtherType::MaxColsAtCompileTime> BlockingType;
+
+    BlockingType blocking(size, size, depth, 1, false);
+
 
     internal::general_matrix_matrix_triangular_product<Index,
-      Scalar, _ActualOtherType::Flags&RowMajorBit ? RowMajor : ColMajor,   OtherBlasTraits::NeedToConjugate  && NumTraits<Scalar>::IsComplex,
-      Scalar, _ActualOtherType::Flags&RowMajorBit ? ColMajor : RowMajor, (!OtherBlasTraits::NeedToConjugate) && NumTraits<Scalar>::IsComplex,
-      MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor, UpLo>
-      ::run(mat.cols(), actualOther.cols(),
+      Scalar, OtherIsRowMajor ? RowMajor : ColMajor,   OtherBlasTraits::NeedToConjugate  && NumTraits<Scalar>::IsComplex,
+      Scalar, OtherIsRowMajor ? ColMajor : RowMajor, (!OtherBlasTraits::NeedToConjugate) && NumTraits<Scalar>::IsComplex,
+      IsRowMajor ? RowMajor : ColMajor, UpLo>
+      ::run(size, depth,
             &actualOther.coeffRef(0,0), actualOther.outerStride(), &actualOther.coeffRef(0,0), actualOther.outerStride(),
-            mat.data(), mat.outerStride(), actualAlpha);
+            mat.data(), mat.outerStride(), actualAlpha, blocking);
   }
 };
 
diff --git a/Eigen/src/Core/products/SelfadjointRank2Update.h b/Eigen/src/Core/products/SelfadjointRank2Update.h
index 8594a97..2ae3641 100644
--- a/Eigen/src/Core/products/SelfadjointRank2Update.h
+++ b/Eigen/src/Core/products/SelfadjointRank2Update.h
@@ -79,11 +79,11 @@
   if (IsRowMajor)
     actualAlpha = numext::conj(actualAlpha);
 
-  internal::selfadjoint_rank2_update_selector<Scalar, Index,
-    typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ UBlasTraits::NeedToConjugate,_ActualUType>::type>::type,
-    typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ VBlasTraits::NeedToConjugate,_ActualVType>::type>::type,
+  typedef typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ UBlasTraits::NeedToConjugate,_ActualUType>::type>::type UType;
+  typedef typename internal::remove_all<typename internal::conj_expr_if<IsRowMajor ^ VBlasTraits::NeedToConjugate,_ActualVType>::type>::type VType;
+  internal::selfadjoint_rank2_update_selector<Scalar, Index, UType, VType,
     (IsRowMajor ? int(UpLo==Upper ? Lower : Upper) : UpLo)>
-    ::run(_expression().const_cast_derived().data(),_expression().outerStride(),actualU,actualV,actualAlpha);
+    ::run(_expression().const_cast_derived().data(),_expression().outerStride(),UType(actualU),VType(actualV),actualAlpha);
 
   return *this;
 }
diff --git a/Eigen/src/Core/products/TriangularMatrixMatrix.h b/Eigen/src/Core/products/TriangularMatrixMatrix.h
index 8110507..f784507 100644
--- a/Eigen/src/Core/products/TriangularMatrixMatrix.h
+++ b/Eigen/src/Core/products/TriangularMatrixMatrix.h
@@ -108,7 +108,7 @@
     Index _rows, Index _cols, Index _depth,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
+    Scalar* _res,        Index resStride,
     const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
     // strip zeros
@@ -117,30 +117,42 @@
     Index depth     = IsLower ? diagSize : _depth;
     Index cols      = _cols;
     
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<Scalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
 
     Index kc = blocking.kc();                   // cache block size along the K direction
     Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
+    // The small panel size must not be larger than blocking size.
+    // Usually this should never be the case because SmallPanelWidth^2 is very small
+    // compared to L2 cache size, but let's be safe:
+    Index panelWidth = (std::min)(Index(SmallPanelWidth),(std::min)(kc,mc));
 
     std::size_t sizeA = kc*mc;
     std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
-    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,LhsStorageOrder> triangularBuffer;
+    // To work around an "error: member reference base type 'Matrix<...>
+    // (Eigen::internal::constructor_without_unaligned_array_assert (*)())' is
+    // not a structure or union" compilation error in nvcc (tested V8.0.61),
+    // create a dummy internal::constructor_without_unaligned_array_assert
+    // object to pass to the Matrix constructor.
+    internal::constructor_without_unaligned_array_assert a;
+    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,LhsStorageOrder> triangularBuffer(a);
     triangularBuffer.setZero();
     if((Mode&ZeroDiag)==ZeroDiag)
       triangularBuffer.diagonal().setZero();
     else
       triangularBuffer.diagonal().setOnes();
 
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder> pack_rhs;
 
     for(Index k2=IsLower ? depth : 0;
         IsLower ? k2>0 : k2<depth;
@@ -156,7 +168,7 @@
         k2 = k2+actual_kc-kc;
       }
 
-      pack_rhs(blockB, &rhs(actual_k2,0), rhsStride, actual_kc, cols);
+      pack_rhs(blockB, rhs.getSubMapper(actual_k2,0), actual_kc, cols);
 
       // the selected lhs's panel has to be split in three different parts:
       //  1 - the part which is zero => skip it
@@ -167,9 +179,9 @@
       if(IsLower || actual_k2<rows)
       {
         // for each small vertical panels of lhs
-        for (Index k1=0; k1<actual_kc; k1+=SmallPanelWidth)
+        for (Index k1=0; k1<actual_kc; k1+=panelWidth)
         {
-          Index actualPanelWidth = std::min<Index>(actual_kc-k1, SmallPanelWidth);
+          Index actualPanelWidth = std::min<Index>(actual_kc-k1, panelWidth);
           Index lengthTarget = IsLower ? actual_kc-k1-actualPanelWidth : k1;
           Index startBlock   = actual_k2+k1;
           Index blockBOffset = k1;
@@ -184,20 +196,22 @@
             for (Index i=IsLower ? k+1 : 0; IsLower ? i<actualPanelWidth : i<k; ++i)
               triangularBuffer.coeffRef(i,k) = lhs(startBlock+i,startBlock+k);
           }
-          pack_lhs(blockA, triangularBuffer.data(), triangularBuffer.outerStride(), actualPanelWidth, actualPanelWidth);
+          pack_lhs(blockA, LhsMapper(triangularBuffer.data(), triangularBuffer.outerStride()), actualPanelWidth, actualPanelWidth);
 
-          gebp_kernel(res+startBlock, resStride, blockA, blockB, actualPanelWidth, actualPanelWidth, cols, alpha,
-                      actualPanelWidth, actual_kc, 0, blockBOffset, blockW);
+          gebp_kernel(res.getSubMapper(startBlock, 0), blockA, blockB,
+                      actualPanelWidth, actualPanelWidth, cols, alpha,
+                      actualPanelWidth, actual_kc, 0, blockBOffset);
 
           // GEBP with remaining micro panel
           if (lengthTarget>0)
           {
             Index startTarget  = IsLower ? actual_k2+k1+actualPanelWidth : actual_k2;
 
-            pack_lhs(blockA, &lhs(startTarget,startBlock), lhsStride, actualPanelWidth, lengthTarget);
+            pack_lhs(blockA, lhs.getSubMapper(startTarget,startBlock), actualPanelWidth, lengthTarget);
 
-            gebp_kernel(res+startTarget, resStride, blockA, blockB, lengthTarget, actualPanelWidth, cols, alpha,
-                        actualPanelWidth, actual_kc, 0, blockBOffset, blockW);
+            gebp_kernel(res.getSubMapper(startTarget, 0), blockA, blockB,
+                        lengthTarget, actualPanelWidth, cols, alpha,
+                        actualPanelWidth, actual_kc, 0, blockBOffset);
           }
         }
       }
@@ -208,10 +222,11 @@
         for(Index i2=start; i2<end; i2+=mc)
         {
           const Index actual_mc = (std::min)(i2+mc,end)-i2;
-          gemm_pack_lhs<Scalar, Index, Traits::mr,Traits::LhsProgress, LhsStorageOrder,false>()
-            (blockA, &lhs(i2, actual_k2), lhsStride, actual_kc, actual_mc);
+          gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr,Traits::LhsProgress, LhsStorageOrder,false>()
+            (blockA, lhs.getSubMapper(i2, actual_k2), actual_kc, actual_mc);
 
-          gebp_kernel(res+i2, resStride, blockA, blockB, actual_mc, actual_kc, cols, alpha, -1, -1, 0, 0, blockW);
+          gebp_kernel(res.getSubMapper(i2, 0), blockA, blockB, actual_mc,
+                      actual_kc, cols, alpha, -1, -1, 0, 0);
         }
       }
     }
@@ -249,40 +264,44 @@
     Index _rows, Index _cols, Index _depth,
     const Scalar* _lhs, Index lhsStride,
     const Scalar* _rhs, Index rhsStride,
-    Scalar* res,        Index resStride,
+    Scalar* _res,        Index resStride,
     const Scalar& alpha, level3_blocking<Scalar,Scalar>& blocking)
   {
+    const Index PacketBytes = packet_traits<Scalar>::size*sizeof(Scalar);
     // strip zeros
     Index diagSize  = (std::min)(_cols,_depth);
     Index rows      = _rows;
     Index depth     = IsLower ? _depth : diagSize;
     Index cols      = IsLower ? diagSize : _cols;
     
-    const_blas_data_mapper<Scalar, Index, LhsStorageOrder> lhs(_lhs,lhsStride);
-    const_blas_data_mapper<Scalar, Index, RhsStorageOrder> rhs(_rhs,rhsStride);
+    typedef const_blas_data_mapper<Scalar, Index, LhsStorageOrder> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, RhsStorageOrder> RhsMapper;
+    typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor> ResMapper;
+    LhsMapper lhs(_lhs,lhsStride);
+    RhsMapper rhs(_rhs,rhsStride);
+    ResMapper res(_res, resStride);
 
     Index kc = blocking.kc();                   // cache block size along the K direction
     Index mc = (std::min)(rows,blocking.mc());  // cache block size along the M direction
 
     std::size_t sizeA = kc*mc;
-    std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
+    std::size_t sizeB = kc*cols+EIGEN_MAX_ALIGN_BYTES/sizeof(Scalar);
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
-    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,RhsStorageOrder> triangularBuffer;
+    internal::constructor_without_unaligned_array_assert a;
+    Matrix<Scalar,SmallPanelWidth,SmallPanelWidth,RhsStorageOrder> triangularBuffer(a);
     triangularBuffer.setZero();
     if((Mode&ZeroDiag)==ZeroDiag)
       triangularBuffer.diagonal().setZero();
     else
       triangularBuffer.diagonal().setOnes();
 
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder,false,true> pack_rhs_panel;
+    gebp_kernel<Scalar, Scalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, LhsStorageOrder> pack_lhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder> pack_rhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr,RhsStorageOrder,false,true> pack_rhs_panel;
 
     for(Index k2=IsLower ? 0 : depth;
         IsLower ? k2<depth  : k2>0;
@@ -304,8 +323,9 @@
       Index ts = (IsLower && actual_k2>=cols) ? 0 : actual_kc;
 
       Scalar* geb = blockB+ts*ts;
+      geb = geb + internal::first_aligned<PacketBytes>(geb,PacketBytes/sizeof(Scalar));
 
-      pack_rhs(geb, &rhs(actual_k2,IsLower ? 0 : k2), rhsStride, actual_kc, rs);
+      pack_rhs(geb, rhs.getSubMapper(actual_k2,IsLower ? 0 : k2), actual_kc, rs);
 
       // pack the triangular part of the rhs padding the unrolled blocks with zeros
       if(ts>0)
@@ -318,7 +338,7 @@
           Index panelLength = IsLower ? actual_kc-j2-actualPanelWidth : j2;
           // general part
           pack_rhs_panel(blockB+j2*actual_kc,
-                         &rhs(actual_k2+panelOffset, actual_j2), rhsStride,
+                         rhs.getSubMapper(actual_k2+panelOffset, actual_j2),
                          panelLength, actualPanelWidth,
                          actual_kc, panelOffset);
 
@@ -332,7 +352,7 @@
           }
 
           pack_rhs_panel(blockB+j2*actual_kc,
-                         triangularBuffer.data(), triangularBuffer.outerStride(),
+                         RhsMapper(triangularBuffer.data(), triangularBuffer.outerStride()),
                          actualPanelWidth, actualPanelWidth,
                          actual_kc, j2);
         }
@@ -341,7 +361,7 @@
       for (Index i2=0; i2<rows; i2+=mc)
       {
         const Index actual_mc = (std::min)(mc,rows-i2);
-        pack_lhs(blockA, &lhs(i2, actual_k2), lhsStride, actual_kc, actual_mc);
+        pack_lhs(blockA, lhs.getSubMapper(i2, actual_k2), actual_kc, actual_mc);
 
         // triangular kernel
         if(ts>0)
@@ -352,19 +372,18 @@
             Index panelLength = IsLower ? actual_kc-j2 : j2+actualPanelWidth;
             Index blockOffset = IsLower ? j2 : 0;
 
-            gebp_kernel(res+i2+(actual_k2+j2)*resStride, resStride,
+            gebp_kernel(res.getSubMapper(i2, actual_k2 + j2),
                         blockA, blockB+j2*actual_kc,
                         actual_mc, panelLength, actualPanelWidth,
                         alpha,
                         actual_kc, actual_kc,  // strides
-                        blockOffset, blockOffset,// offsets
-                        blockW); // workspace
+                        blockOffset, blockOffset);// offsets
           }
         }
-        gebp_kernel(res+i2+(IsLower ? 0 : k2)*resStride, resStride,
+        gebp_kernel(res.getSubMapper(i2, IsLower ? 0 : k2),
                     blockA, geb, actual_mc, actual_kc, rs,
                     alpha,
-                    -1, -1, 0, 0, blockW);
+                    -1, -1, 0, 0);
       }
     }
   }
@@ -373,28 +392,31 @@
 * Wrapper to product_triangular_matrix_matrix
 ***************************************************************************/
 
-template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false> >
-  : traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>, Lhs, Rhs> >
-{};
-
 } // end namespace internal
 
+namespace internal {
 template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
-  : public ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,false>, Lhs, Rhs >
+struct triangular_product_impl<Mode,LhsIsTriangular,Lhs,false,Rhs,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
-
-  TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  template<typename Dest> static void run(Dest& dst, const Lhs &a_lhs, const Rhs &a_rhs, const typename Dest::Scalar& alpha)
   {
-    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(m_lhs);
-    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(m_rhs);
+    typedef typename Lhs::Scalar  LhsScalar;
+    typedef typename Rhs::Scalar  RhsScalar;
+    typedef typename Dest::Scalar Scalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
+    
+    typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs);
 
-    Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(m_lhs)
-                               * RhsBlasTraits::extractScalarFactor(m_rhs);
+    LhsScalar lhs_alpha = LhsBlasTraits::extractScalarFactor(a_lhs);
+    RhsScalar rhs_alpha = RhsBlasTraits::extractScalarFactor(a_rhs);
+    Scalar actualAlpha = alpha * lhs_alpha * rhs_alpha;
 
     typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,Scalar,Scalar,
               Lhs::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime, Lhs::MaxColsAtCompileTime,4> BlockingType;
@@ -405,23 +427,40 @@
     Index stripedDepth = LhsIsTriangular ? ((!IsLower) ? lhs.cols() : (std::min)(lhs.cols(),lhs.rows()))
                                          : ((IsLower)  ? rhs.rows() : (std::min)(rhs.rows(),rhs.cols()));
 
-    BlockingType blocking(stripedRows, stripedCols, stripedDepth);
+    BlockingType blocking(stripedRows, stripedCols, stripedDepth, 1, false);
 
     internal::product_triangular_matrix_matrix<Scalar, Index,
       Mode, LhsIsTriangular,
-      (internal::traits<_ActualLhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
-      (internal::traits<_ActualRhsType>::Flags&RowMajorBit) ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
+      (internal::traits<ActualLhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, LhsBlasTraits::NeedToConjugate,
+      (internal::traits<ActualRhsTypeCleaned>::Flags&RowMajorBit) ? RowMajor : ColMajor, RhsBlasTraits::NeedToConjugate,
       (internal::traits<Dest          >::Flags&RowMajorBit) ? RowMajor : ColMajor>
       ::run(
         stripedRows, stripedCols, stripedDepth,   // sizes
-        &lhs.coeffRef(0,0),    lhs.outerStride(), // lhs info
-        &rhs.coeffRef(0,0),    rhs.outerStride(), // rhs info
+        &lhs.coeffRef(0,0), lhs.outerStride(),    // lhs info
+        &rhs.coeffRef(0,0), rhs.outerStride(),    // rhs info
         &dst.coeffRef(0,0), dst.outerStride(),    // result info
         actualAlpha, blocking
       );
+
+    // Apply correction if the diagonal is unit and a scalar factor was nested:
+    if ((Mode&UnitDiag)==UnitDiag)
+    {
+      if (LhsIsTriangular && lhs_alpha!=LhsScalar(1))
+      {
+        Index diagSize = (std::min)(lhs.rows(),lhs.cols());
+        dst.topRows(diagSize) -= ((lhs_alpha-LhsScalar(1))*a_rhs).topRows(diagSize);
+      }
+      else if ((!LhsIsTriangular) && rhs_alpha!=RhsScalar(1))
+      {
+        Index diagSize = (std::min)(rhs.rows(),rhs.cols());
+        dst.leftCols(diagSize) -= (rhs_alpha-RhsScalar(1))*a_lhs.leftCols(diagSize);
+      }
+    }
   }
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_TRIANGULAR_MATRIX_MATRIX_H
diff --git a/Eigen/src/Core/products/TriangularMatrixMatrix_MKL.h b/Eigen/src/Core/products/TriangularMatrixMatrix_BLAS.h
similarity index 73%
rename from Eigen/src/Core/products/TriangularMatrixMatrix_MKL.h
rename to Eigen/src/Core/products/TriangularMatrixMatrix_BLAS.h
index ba41a1c..a25197a 100644
--- a/Eigen/src/Core/products/TriangularMatrixMatrix_MKL.h
+++ b/Eigen/src/Core/products/TriangularMatrixMatrix_BLAS.h
@@ -25,13 +25,13 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to BLAS F77
  *   Triangular matrix * matrix product functionality based on ?TRMM.
  ********************************************************************************
 */
 
-#ifndef EIGEN_TRIANGULAR_MATRIX_MATRIX_MKL_H
-#define EIGEN_TRIANGULAR_MATRIX_MATRIX_MKL_H
+#ifndef EIGEN_TRIANGULAR_MATRIX_MATRIX_BLAS_H
+#define EIGEN_TRIANGULAR_MATRIX_MATRIX_BLAS_H
 
 namespace Eigen { 
 
@@ -50,7 +50,7 @@
 
 
 // try to go to BLAS specialization
-#define EIGEN_MKL_TRMM_SPECIALIZE(Scalar, LhsIsTriangular) \
+#define EIGEN_BLAS_TRMM_SPECIALIZE(Scalar, LhsIsTriangular) \
 template <typename Index, int Mode, \
           int LhsStorageOrder, bool ConjugateLhs, \
           int RhsStorageOrder, bool ConjugateRhs> \
@@ -65,17 +65,17 @@
   } \
 };
 
-EIGEN_MKL_TRMM_SPECIALIZE(double, true)
-EIGEN_MKL_TRMM_SPECIALIZE(double, false)
-EIGEN_MKL_TRMM_SPECIALIZE(dcomplex, true)
-EIGEN_MKL_TRMM_SPECIALIZE(dcomplex, false)
-EIGEN_MKL_TRMM_SPECIALIZE(float, true)
-EIGEN_MKL_TRMM_SPECIALIZE(float, false)
-EIGEN_MKL_TRMM_SPECIALIZE(scomplex, true)
-EIGEN_MKL_TRMM_SPECIALIZE(scomplex, false)
+EIGEN_BLAS_TRMM_SPECIALIZE(double, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(double, false)
+EIGEN_BLAS_TRMM_SPECIALIZE(dcomplex, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(dcomplex, false)
+EIGEN_BLAS_TRMM_SPECIALIZE(float, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(float, false)
+EIGEN_BLAS_TRMM_SPECIALIZE(scomplex, true)
+EIGEN_BLAS_TRMM_SPECIALIZE(scomplex, false)
 
 // implements col-major += alpha * op(triangular) * op(general)
-#define EIGEN_MKL_TRMM_L(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_TRMM_L(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
 template <typename Index, int Mode, \
           int LhsStorageOrder, bool ConjugateLhs, \
           int RhsStorageOrder, bool ConjugateRhs> \
@@ -106,13 +106,14 @@
    typedef Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> MatrixLhs; \
    typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs; \
 \
-/* Non-square case - doesn't fit to MKL ?TRMM. Fall to default triangular product or call MKL ?GEMM*/ \
+/* Non-square case - doesn't fit to BLAS ?TRMM. Fall to default triangular product or call BLAS ?GEMM*/ \
    if (rows != depth) { \
 \
-     int nthr = mkl_domain_get_max_threads(MKL_BLAS); \
+     /* FIXME handle mkl_domain_get_max_threads */ \
+     /*int nthr = mkl_domain_get_max_threads(EIGEN_BLAS_DOMAIN_BLAS);*/ int nthr = 1;\
 \
      if (((nthr==1) && (((std::max)(rows,depth)-diagSize)/(double)diagSize < 0.5))) { \
-     /* Most likely no benefit to call TRMM or GEMM from MKL*/ \
+     /* Most likely no benefit to call TRMM or GEMM from BLAS */ \
        product_triangular_matrix_matrix<EIGTYPE,Index,Mode,true, \
        LhsStorageOrder,ConjugateLhs, RhsStorageOrder, ConjugateRhs, ColMajor, BuiltIn>::run( \
            _rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha, blocking); \
@@ -121,27 +122,23 @@
      /* Make sense to call GEMM */ \
        Map<const MatrixLhs, 0, OuterStride<> > lhsMap(_lhs,rows,depth,OuterStride<>(lhsStride)); \
        MatrixLhs aa_tmp=lhsMap.template triangularView<Mode>(); \
-       MKL_INT aStride = aa_tmp.outerStride(); \
-       gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> gemm_blocking(_rows,_cols,_depth); \
+       BlasIndex aStride = convert_index<BlasIndex>(aa_tmp.outerStride()); \
+       gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> gemm_blocking(_rows,_cols,_depth, 1, true); \
        general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor>::run( \
        rows, cols, depth, aa_tmp.data(), aStride, _rhs, rhsStride, res, resStride, alpha, gemm_blocking, 0); \
 \
-     /*std::cout << "TRMM_L: A is not square! Go to MKL GEMM implementation! " << nthr<<" \n";*/ \
+     /*std::cout << "TRMM_L: A is not square! Go to BLAS GEMM implementation! " << nthr<<" \n";*/ \
      } \
      return; \
    } \
    char side = 'L', transa, uplo, diag = 'N'; \
    EIGTYPE *b; \
    const EIGTYPE *a; \
-   MKL_INT m, n, lda, ldb; \
-   MKLTYPE alpha_; \
-\
-/* Set alpha_*/ \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \
+   BlasIndex m, n, lda, ldb; \
 \
 /* Set m, n */ \
-   m = (MKL_INT)diagSize; \
-   n = (MKL_INT)cols; \
+   m = convert_index<BlasIndex>(diagSize); \
+   n = convert_index<BlasIndex>(cols); \
 \
 /* Set trans */ \
    transa = (LhsStorageOrder==RowMajor) ? ((ConjugateLhs) ? 'C' : 'T') : 'N'; \
@@ -152,7 +149,7 @@
 \
    if (ConjugateRhs) b_tmp = rhs.conjugate(); else b_tmp = rhs; \
    b = b_tmp.data(); \
-   ldb = b_tmp.outerStride(); \
+   ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
 \
 /* Set uplo */ \
    uplo = IsLower ? 'L' : 'U'; \
@@ -168,14 +165,14 @@
      else if (IsUnitDiag) \
        a_tmp.diagonal().setOnes();\
      a = a_tmp.data(); \
-     lda = a_tmp.outerStride(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
    } else { \
      a = _lhs; \
-     lda = lhsStride; \
+     lda = convert_index<BlasIndex>(lhsStride); \
    } \
-   /*std::cout << "TRMM_L: A is square! Go to MKL TRMM implementation! \n";*/ \
+   /*std::cout << "TRMM_L: A is square! Go to BLAS TRMM implementation! \n";*/ \
 /* call ?trmm*/ \
-   MKLPREFIX##trmm(&side, &uplo, &transa, &diag, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (MKLTYPE*)b, &ldb); \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)b, &ldb); \
 \
 /* Add op(a_triangular)*b into res*/ \
    Map<MatrixX##EIGPREFIX, 0, OuterStride<> > res_tmp(res,rows,cols,OuterStride<>(resStride)); \
@@ -183,13 +180,20 @@
   } \
 };
 
-EIGEN_MKL_TRMM_L(double, double, d, d)
-EIGEN_MKL_TRMM_L(dcomplex, MKL_Complex16, cd, z)
-EIGEN_MKL_TRMM_L(float, float, f, s)
-EIGEN_MKL_TRMM_L(scomplex, MKL_Complex8, cf, c)
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMM_L(double, double, d, dtrmm)
+EIGEN_BLAS_TRMM_L(dcomplex, MKL_Complex16, cd, ztrmm)
+EIGEN_BLAS_TRMM_L(float, float, f, strmm)
+EIGEN_BLAS_TRMM_L(scomplex, MKL_Complex8, cf, ctrmm)
+#else
+EIGEN_BLAS_TRMM_L(double, double, d, dtrmm_)
+EIGEN_BLAS_TRMM_L(dcomplex, double, cd, ztrmm_)
+EIGEN_BLAS_TRMM_L(float, float, f, strmm_)
+EIGEN_BLAS_TRMM_L(scomplex, float, cf, ctrmm_)
+#endif
 
 // implements col-major += alpha * op(general) * op(triangular)
-#define EIGEN_MKL_TRMM_R(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_TRMM_R(EIGTYPE, BLASTYPE, EIGPREFIX, BLASFUNC) \
 template <typename Index, int Mode, \
           int LhsStorageOrder, bool ConjugateLhs, \
           int RhsStorageOrder, bool ConjugateRhs> \
@@ -220,13 +224,13 @@
    typedef Matrix<EIGTYPE, Dynamic, Dynamic, LhsStorageOrder> MatrixLhs; \
    typedef Matrix<EIGTYPE, Dynamic, Dynamic, RhsStorageOrder> MatrixRhs; \
 \
-/* Non-square case - doesn't fit to MKL ?TRMM. Fall to default triangular product or call MKL ?GEMM*/ \
+/* Non-square case - doesn't fit to BLAS ?TRMM. Fall to default triangular product or call BLAS ?GEMM*/ \
    if (cols != depth) { \
 \
-     int nthr = mkl_domain_get_max_threads(MKL_BLAS); \
+     int nthr = 1 /*mkl_domain_get_max_threads(EIGEN_BLAS_DOMAIN_BLAS)*/; \
 \
      if ((nthr==1) && (((std::max)(cols,depth)-diagSize)/(double)diagSize < 0.5)) { \
-     /* Most likely no benefit to call TRMM or GEMM from MKL*/ \
+     /* Most likely no benefit to call TRMM or GEMM from BLAS*/ \
        product_triangular_matrix_matrix<EIGTYPE,Index,Mode,false, \
        LhsStorageOrder,ConjugateLhs, RhsStorageOrder, ConjugateRhs, ColMajor, BuiltIn>::run( \
            _rows, _cols, _depth, _lhs, lhsStride, _rhs, rhsStride, res, resStride, alpha, blocking); \
@@ -235,27 +239,23 @@
      /* Make sense to call GEMM */ \
        Map<const MatrixRhs, 0, OuterStride<> > rhsMap(_rhs,depth,cols, OuterStride<>(rhsStride)); \
        MatrixRhs aa_tmp=rhsMap.template triangularView<Mode>(); \
-       MKL_INT aStride = aa_tmp.outerStride(); \
-       gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> gemm_blocking(_rows,_cols,_depth); \
+       BlasIndex aStride = convert_index<BlasIndex>(aa_tmp.outerStride()); \
+       gemm_blocking_space<ColMajor,EIGTYPE,EIGTYPE,Dynamic,Dynamic,Dynamic> gemm_blocking(_rows,_cols,_depth, 1, true); \
        general_matrix_matrix_product<Index,EIGTYPE,LhsStorageOrder,ConjugateLhs,EIGTYPE,RhsStorageOrder,ConjugateRhs,ColMajor>::run( \
        rows, cols, depth, _lhs, lhsStride, aa_tmp.data(), aStride, res, resStride, alpha, gemm_blocking, 0); \
 \
-     /*std::cout << "TRMM_R: A is not square! Go to MKL GEMM implementation! " << nthr<<" \n";*/ \
+     /*std::cout << "TRMM_R: A is not square! Go to BLAS GEMM implementation! " << nthr<<" \n";*/ \
      } \
      return; \
    } \
    char side = 'R', transa, uplo, diag = 'N'; \
    EIGTYPE *b; \
    const EIGTYPE *a; \
-   MKL_INT m, n, lda, ldb; \
-   MKLTYPE alpha_; \
-\
-/* Set alpha_*/ \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \
+   BlasIndex m, n, lda, ldb; \
 \
 /* Set m, n */ \
-   m = (MKL_INT)rows; \
-   n = (MKL_INT)diagSize; \
+   m = convert_index<BlasIndex>(rows); \
+   n = convert_index<BlasIndex>(diagSize); \
 \
 /* Set trans */ \
    transa = (RhsStorageOrder==RowMajor) ? ((ConjugateRhs) ? 'C' : 'T') : 'N'; \
@@ -266,7 +266,7 @@
 \
    if (ConjugateLhs) b_tmp = lhs.conjugate(); else b_tmp = lhs; \
    b = b_tmp.data(); \
-   ldb = b_tmp.outerStride(); \
+   ldb = convert_index<BlasIndex>(b_tmp.outerStride()); \
 \
 /* Set uplo */ \
    uplo = IsLower ? 'L' : 'U'; \
@@ -282,14 +282,14 @@
      else if (IsUnitDiag) \
        a_tmp.diagonal().setOnes();\
      a = a_tmp.data(); \
-     lda = a_tmp.outerStride(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
    } else { \
      a = _rhs; \
-     lda = rhsStride; \
+     lda = convert_index<BlasIndex>(rhsStride); \
    } \
-   /*std::cout << "TRMM_R: A is square! Go to MKL TRMM implementation! \n";*/ \
+   /*std::cout << "TRMM_R: A is square! Go to BLAS TRMM implementation! \n";*/ \
 /* call ?trmm*/ \
-   MKLPREFIX##trmm(&side, &uplo, &transa, &diag, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (MKLTYPE*)b, &ldb); \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)b, &ldb); \
 \
 /* Add op(a_triangular)*b into res*/ \
    Map<MatrixX##EIGPREFIX, 0, OuterStride<> > res_tmp(res,rows,cols,OuterStride<>(resStride)); \
@@ -297,13 +297,19 @@
   } \
 };
 
-EIGEN_MKL_TRMM_R(double, double, d, d)
-EIGEN_MKL_TRMM_R(dcomplex, MKL_Complex16, cd, z)
-EIGEN_MKL_TRMM_R(float, float, f, s)
-EIGEN_MKL_TRMM_R(scomplex, MKL_Complex8, cf, c)
-
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMM_R(double, double, d, dtrmm)
+EIGEN_BLAS_TRMM_R(dcomplex, MKL_Complex16, cd, ztrmm)
+EIGEN_BLAS_TRMM_R(float, float, f, strmm)
+EIGEN_BLAS_TRMM_R(scomplex, MKL_Complex8, cf, ctrmm)
+#else
+EIGEN_BLAS_TRMM_R(double, double, d, dtrmm_)
+EIGEN_BLAS_TRMM_R(dcomplex, double, cd, ztrmm_)
+EIGEN_BLAS_TRMM_R(float, float, f, strmm_)
+EIGEN_BLAS_TRMM_R(scomplex, float, cf, ctrmm_)
+#endif
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_TRIANGULAR_MATRIX_MATRIX_MKL_H
+#endif // EIGEN_TRIANGULAR_MATRIX_MATRIX_BLAS_H
diff --git a/Eigen/src/Core/products/TriangularMatrixVector.h b/Eigen/src/Core/products/TriangularMatrixVector.h
index 6117d5a..76bfa15 100644
--- a/Eigen/src/Core/products/TriangularMatrixVector.h
+++ b/Eigen/src/Core/products/TriangularMatrixVector.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_TRIANGULARMATRIXVECTOR_H
 #define EIGEN_TRIANGULARMATRIXVECTOR_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -20,20 +20,20 @@
 template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs, int Version>
 struct triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,ColMajor,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   enum {
     IsLower = ((Mode&Lower)==Lower),
     HasUnitDiag = (Mode & UnitDiag)==UnitDiag,
     HasZeroDiag = (Mode & ZeroDiag)==ZeroDiag
   };
   static EIGEN_DONT_INLINE  void run(Index _rows, Index _cols, const LhsScalar* _lhs, Index lhsStride,
-                                     const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const ResScalar& alpha);
+                                     const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const RhsScalar& alpha);
 };
 
 template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs, int Version>
 EIGEN_DONT_INLINE void triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,ColMajor,Version>
   ::run(Index _rows, Index _cols, const LhsScalar* _lhs, Index lhsStride,
-        const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const ResScalar& alpha)
+        const RhsScalar* _rhs, Index rhsIncr, ResScalar* _res, Index resIncr, const RhsScalar& alpha)
   {
     static const Index PanelWidth = EIGEN_TUNE_TRIANGULAR_PANEL_WIDTH;
     Index size = (std::min)(_rows,_cols);
@@ -43,7 +43,7 @@
     typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,ColMajor>, 0, OuterStride<> > LhsMap;
     const LhsMap lhs(_lhs,rows,cols,OuterStride<>(lhsStride));
     typename conj_expr_if<ConjLhs,LhsMap>::type cjLhs(lhs);
-    
+
     typedef Map<const Matrix<RhsScalar,Dynamic,1>, 0, InnerStride<> > RhsMap;
     const RhsMap rhs(_rhs,cols,InnerStride<>(rhsIncr));
     typename conj_expr_if<ConjRhs,RhsMap>::type cjRhs(rhs);
@@ -51,6 +51,9 @@
     typedef Map<Matrix<ResScalar,Dynamic,1> > ResMap;
     ResMap res(_res,rows);
 
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+
     for (Index pi=0; pi<size; pi+=PanelWidth)
     {
       Index actualPanelWidth = (std::min)(PanelWidth, size-pi);
@@ -68,19 +71,19 @@
       if (r>0)
       {
         Index s = IsLower ? pi+actualPanelWidth : 0;
-        general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjLhs,RhsScalar,ConjRhs,BuiltIn>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs,BuiltIn>::run(
             r, actualPanelWidth,
-            &lhs.coeffRef(s,pi), lhsStride,
-            &rhs.coeffRef(pi), rhsIncr,
+            LhsMapper(&lhs.coeffRef(s,pi), lhsStride),
+            RhsMapper(&rhs.coeffRef(pi), rhsIncr),
             &res.coeffRef(s), resIncr, alpha);
       }
     }
     if((!IsLower) && cols>size)
     {
-      general_matrix_vector_product<Index,LhsScalar,ColMajor,ConjLhs,RhsScalar,ConjRhs>::run(
+      general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs>::run(
           rows, cols-size,
-          &lhs.coeffRef(0,size), lhsStride,
-          &rhs.coeffRef(size), rhsIncr,
+          LhsMapper(&lhs.coeffRef(0,size), lhsStride),
+          RhsMapper(&rhs.coeffRef(size), rhsIncr),
           _res, resIncr, alpha);
     }
   }
@@ -88,7 +91,7 @@
 template<typename Index, int Mode, typename LhsScalar, bool ConjLhs, typename RhsScalar, bool ConjRhs,int Version>
 struct triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,RowMajor,Version>
 {
-  typedef typename scalar_product_traits<LhsScalar, RhsScalar>::ReturnType ResScalar;
+  typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar;
   enum {
     IsLower = ((Mode&Lower)==Lower),
     HasUnitDiag = (Mode & UnitDiag)==UnitDiag,
@@ -118,7 +121,10 @@
 
     typedef Map<Matrix<ResScalar,Dynamic,1>, 0, InnerStride<> > ResMap;
     ResMap res(_res,rows,InnerStride<>(resIncr));
-    
+
+    typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,RowMajor> RhsMapper;
+
     for (Index pi=0; pi<diagSize; pi+=PanelWidth)
     {
       Index actualPanelWidth = (std::min)(PanelWidth, diagSize-pi);
@@ -136,19 +142,19 @@
       if (r>0)
       {
         Index s = IsLower ? 0 : pi + actualPanelWidth;
-        general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjLhs,RhsScalar,ConjRhs,BuiltIn>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs,BuiltIn>::run(
             actualPanelWidth, r,
-            &lhs.coeffRef(pi,s), lhsStride,
-            &rhs.coeffRef(s), rhsIncr,
+            LhsMapper(&lhs.coeffRef(pi,s), lhsStride),
+            RhsMapper(&rhs.coeffRef(s), rhsIncr),
             &res.coeffRef(pi), resIncr, alpha);
       }
     }
     if(IsLower && rows>diagSize)
     {
-      general_matrix_vector_product<Index,LhsScalar,RowMajor,ConjLhs,RhsScalar,ConjRhs>::run(
+      general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,ConjLhs,RhsScalar,RhsMapper,ConjRhs>::run(
             rows-diagSize, cols,
-            &lhs.coeffRef(diagSize,0), lhsStride,
-            &rhs.coeffRef(0), rhsIncr,
+            LhsMapper(&lhs.coeffRef(diagSize,0), lhsStride),
+            RhsMapper(&rhs.coeffRef(0), rhsIncr),
             &res.coeffRef(diagSize), resIncr, alpha);
     }
   }
@@ -157,83 +163,67 @@
 * Wrapper to product_triangular_vector
 ***************************************************************************/
 
-template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,true> >
- : traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,false,Rhs,true>, Lhs, Rhs> >
-{};
-
-template<int Mode, bool LhsIsTriangular, typename Lhs, typename Rhs>
-struct traits<TriangularProduct<Mode,LhsIsTriangular,Lhs,true,Rhs,false> >
- : traits<ProductBase<TriangularProduct<Mode,LhsIsTriangular,Lhs,true,Rhs,false>, Lhs, Rhs> >
-{};
-
-
-template<int StorageOrder>
+template<int Mode,int StorageOrder>
 struct trmv_selector;
 
 } // end namespace internal
 
+namespace internal {
+
 template<int Mode, typename Lhs, typename Rhs>
-struct TriangularProduct<Mode,true,Lhs,false,Rhs,true>
-  : public ProductBase<TriangularProduct<Mode,true,Lhs,false,Rhs,true>, Lhs, Rhs >
+struct triangular_product_impl<Mode,true,Lhs,false,Rhs,true>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
-
-  TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  template<typename Dest> static void run(Dest& dst, const Lhs &lhs, const Rhs &rhs, const typename Dest::Scalar& alpha)
   {
-    eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+    eigen_assert(dst.rows()==lhs.rows() && dst.cols()==rhs.cols());
   
-    internal::trmv_selector<(int(internal::traits<Lhs>::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(*this, dst, alpha);
+    internal::trmv_selector<Mode,(int(internal::traits<Lhs>::Flags)&RowMajorBit) ? RowMajor : ColMajor>::run(lhs, rhs, dst, alpha);
   }
 };
 
 template<int Mode, typename Lhs, typename Rhs>
-struct TriangularProduct<Mode,false,Lhs,true,Rhs,false>
-  : public ProductBase<TriangularProduct<Mode,false,Lhs,true,Rhs,false>, Lhs, Rhs >
+struct triangular_product_impl<Mode,false,Lhs,true,Rhs,false>
 {
-  EIGEN_PRODUCT_PUBLIC_INTERFACE(TriangularProduct)
-
-  TriangularProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) {}
-
-  template<typename Dest> void scaleAndAddTo(Dest& dst, const Scalar& alpha) const
+  template<typename Dest> static void run(Dest& dst, const Lhs &lhs, const Rhs &rhs, const typename Dest::Scalar& alpha)
   {
-    eigen_assert(dst.rows()==m_lhs.rows() && dst.cols()==m_rhs.cols());
+    eigen_assert(dst.rows()==lhs.rows() && dst.cols()==rhs.cols());
 
-    typedef TriangularProduct<(Mode & (UnitDiag|ZeroDiag)) | ((Mode & Lower) ? Upper : Lower),true,Transpose<const Rhs>,false,Transpose<const Lhs>,true> TriangularProductTranspose;
     Transpose<Dest> dstT(dst);
-    internal::trmv_selector<(int(internal::traits<Rhs>::Flags)&RowMajorBit) ? ColMajor : RowMajor>::run(
-      TriangularProductTranspose(m_rhs.transpose(),m_lhs.transpose()), dstT, alpha);
+    internal::trmv_selector<(Mode & (UnitDiag|ZeroDiag)) | ((Mode & Lower) ? Upper : Lower),
+                            (int(internal::traits<Rhs>::Flags)&RowMajorBit) ? ColMajor : RowMajor>
+            ::run(rhs.transpose(),lhs.transpose(), dstT, alpha);
   }
 };
 
+} // end namespace internal
+
 namespace internal {
 
 // TODO: find a way to factorize this piece of code with gemv_selector since the logic is exactly the same.
   
-template<> struct trmv_selector<ColMajor>
+template<int Mode> struct trmv_selector<Mode,ColMajor>
 {
-  template<int Mode, typename Lhs, typename Rhs, typename Dest>
-  static void run(const TriangularProduct<Mode,true,Lhs,false,Rhs,true>& prod, Dest& dest, const typename TriangularProduct<Mode,true,Lhs,false,Rhs,true>::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef TriangularProduct<Mode,true,Lhs,false,Rhs,true> ProductType;
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::LhsScalar   LhsScalar;
-    typedef typename ProductType::RhsScalar   RhsScalar;
-    typedef typename ProductType::Scalar      ResScalar;
-    typedef typename ProductType::RealScalar  RealScalar;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
-    typedef Map<Matrix<ResScalar,Dynamic,1>, Aligned> MappedDest;
+    typedef typename Lhs::Scalar      LhsScalar;
+    typedef typename Rhs::Scalar      RhsScalar;
+    typedef typename Dest::Scalar     ResScalar;
+    typedef typename Dest::RealScalar RealScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    
+    typedef Map<Matrix<ResScalar,Dynamic,1>, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits<ResScalar>::size)> MappedDest;
 
-    typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
-    typename internal::add_const_on_value_type<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
+    typename internal::add_const_on_value_type<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
+    typename internal::add_const_on_value_type<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
 
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    LhsScalar lhs_alpha = LhsBlasTraits::extractScalarFactor(lhs);
+    RhsScalar rhs_alpha = RhsBlasTraits::extractScalarFactor(rhs);
+    ResScalar actualAlpha = alpha * lhs_alpha * rhs_alpha;
 
     enum {
       // FIXME find a way to allow an inner stride on the result if packet_traits<Scalar>::size==1
@@ -247,7 +237,7 @@
 
     bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0));
     bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible;
-    
+
     RhsScalar compatibleAlpha = get_factor<ResScalar,RhsScalar>::run(actualAlpha);
 
     ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(),
@@ -267,7 +257,7 @@
       else
         MappedDest(actualDestPtr, dest.size()) = dest;
     }
-    
+
     internal::triangular_matrix_vector_product
       <Index,Mode,
        LhsScalar, LhsBlasTraits::NeedToConjugate,
@@ -285,36 +275,42 @@
       else
         dest = MappedDest(actualDestPtr, dest.size());
     }
+
+    if ( ((Mode&UnitDiag)==UnitDiag) && (lhs_alpha!=LhsScalar(1)) )
+    {
+      Index diagSize = (std::min)(lhs.rows(),lhs.cols());
+      dest.head(diagSize) -= (lhs_alpha-LhsScalar(1))*rhs.head(diagSize);
+    }
   }
 };
 
-template<> struct trmv_selector<RowMajor>
+template<int Mode> struct trmv_selector<Mode,RowMajor>
 {
-  template<int Mode, typename Lhs, typename Rhs, typename Dest>
-  static void run(const TriangularProduct<Mode,true,Lhs,false,Rhs,true>& prod, Dest& dest, const typename TriangularProduct<Mode,true,Lhs,false,Rhs,true>::Scalar& alpha)
+  template<typename Lhs, typename Rhs, typename Dest>
+  static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha)
   {
-    typedef TriangularProduct<Mode,true,Lhs,false,Rhs,true> ProductType;
-    typedef typename ProductType::LhsScalar LhsScalar;
-    typedef typename ProductType::RhsScalar RhsScalar;
-    typedef typename ProductType::Scalar    ResScalar;
-    typedef typename ProductType::Index Index;
-    typedef typename ProductType::ActualLhsType ActualLhsType;
-    typedef typename ProductType::ActualRhsType ActualRhsType;
-    typedef typename ProductType::_ActualRhsType _ActualRhsType;
-    typedef typename ProductType::LhsBlasTraits LhsBlasTraits;
-    typedef typename ProductType::RhsBlasTraits RhsBlasTraits;
+    typedef typename Lhs::Scalar      LhsScalar;
+    typedef typename Rhs::Scalar      RhsScalar;
+    typedef typename Dest::Scalar     ResScalar;
+    
+    typedef internal::blas_traits<Lhs> LhsBlasTraits;
+    typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType;
+    typedef internal::blas_traits<Rhs> RhsBlasTraits;
+    typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType;
+    typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned;
 
-    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(prod.lhs());
-    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(prod.rhs());
+    typename add_const<ActualLhsType>::type actualLhs = LhsBlasTraits::extract(lhs);
+    typename add_const<ActualRhsType>::type actualRhs = RhsBlasTraits::extract(rhs);
 
-    ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs())
-                                  * RhsBlasTraits::extractScalarFactor(prod.rhs());
+    LhsScalar lhs_alpha = LhsBlasTraits::extractScalarFactor(lhs);
+    RhsScalar rhs_alpha = RhsBlasTraits::extractScalarFactor(rhs);
+    ResScalar actualAlpha = alpha * lhs_alpha * rhs_alpha;
 
     enum {
-      DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1
+      DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1
     };
 
-    gemv_static_vector_if<RhsScalar,_ActualRhsType::SizeAtCompileTime,_ActualRhsType::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
+    gemv_static_vector_if<RhsScalar,ActualRhsTypeCleaned::SizeAtCompileTime,ActualRhsTypeCleaned::MaxSizeAtCompileTime,!DirectlyUseRhs> static_rhs;
 
     ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(),
         DirectlyUseRhs ? const_cast<RhsScalar*>(actualRhs.data()) : static_rhs.data());
@@ -322,12 +318,12 @@
     if(!DirectlyUseRhs)
     {
       #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN
-      int size = actualRhs.size();
+      Index size = actualRhs.size();
       EIGEN_DENSE_STORAGE_CTOR_PLUGIN
       #endif
-      Map<typename _ActualRhsType::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
+      Map<typename ActualRhsTypeCleaned::PlainObject>(actualRhsPtr, actualRhs.size()) = actualRhs;
     }
-    
+
     internal::triangular_matrix_vector_product
       <Index,Mode,
        LhsScalar, LhsBlasTraits::NeedToConjugate,
@@ -338,6 +334,12 @@
             actualRhsPtr,1,
             dest.data(),dest.innerStride(),
             actualAlpha);
+
+    if ( ((Mode&UnitDiag)==UnitDiag) && (lhs_alpha!=LhsScalar(1)) )
+    {
+      Index diagSize = (std::min)(lhs.rows(),lhs.cols());
+      dest.head(diagSize) -= (lhs_alpha-LhsScalar(1))*rhs.head(diagSize);
+    }
   }
 };
 
diff --git a/Eigen/src/Core/products/TriangularMatrixVector_MKL.h b/Eigen/src/Core/products/TriangularMatrixVector_BLAS.h
similarity index 69%
rename from Eigen/src/Core/products/TriangularMatrixVector_MKL.h
rename to Eigen/src/Core/products/TriangularMatrixVector_BLAS.h
index 09f110d..3d47a2b 100644
--- a/Eigen/src/Core/products/TriangularMatrixVector_MKL.h
+++ b/Eigen/src/Core/products/TriangularMatrixVector_BLAS.h
@@ -25,13 +25,13 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to BLAS F77
  *   Triangular matrix-vector product functionality based on ?TRMV.
  ********************************************************************************
 */
 
-#ifndef EIGEN_TRIANGULAR_MATRIX_VECTOR_MKL_H
-#define EIGEN_TRIANGULAR_MATRIX_VECTOR_MKL_H
+#ifndef EIGEN_TRIANGULAR_MATRIX_VECTOR_BLAS_H
+#define EIGEN_TRIANGULAR_MATRIX_VECTOR_BLAS_H
 
 namespace Eigen { 
 
@@ -47,7 +47,7 @@
 struct triangular_matrix_vector_product_trmv :
   triangular_matrix_vector_product<Index,Mode,LhsScalar,ConjLhs,RhsScalar,ConjRhs,StorageOrder,BuiltIn> {};
 
-#define EIGEN_MKL_TRMV_SPECIALIZE(Scalar) \
+#define EIGEN_BLAS_TRMV_SPECIALIZE(Scalar) \
 template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \
 struct triangular_matrix_vector_product<Index,Mode,Scalar,ConjLhs,Scalar,ConjRhs,ColMajor,Specialized> { \
  static void run(Index _rows, Index _cols, const Scalar* _lhs, Index lhsStride, \
@@ -65,13 +65,13 @@
   } \
 };
 
-EIGEN_MKL_TRMV_SPECIALIZE(double)
-EIGEN_MKL_TRMV_SPECIALIZE(float)
-EIGEN_MKL_TRMV_SPECIALIZE(dcomplex)
-EIGEN_MKL_TRMV_SPECIALIZE(scomplex)
+EIGEN_BLAS_TRMV_SPECIALIZE(double)
+EIGEN_BLAS_TRMV_SPECIALIZE(float)
+EIGEN_BLAS_TRMV_SPECIALIZE(dcomplex)
+EIGEN_BLAS_TRMV_SPECIALIZE(scomplex)
 
 // implements col-major: res += alpha * op(triangular) * vector
-#define EIGEN_MKL_TRMV_CM(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_TRMV_CM(EIGTYPE, BLASTYPE, EIGPREFIX, BLASPREFIX, BLASPOSTFIX) \
 template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \
 struct triangular_matrix_vector_product_trmv<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,ColMajor> { \
   enum { \
@@ -105,17 +105,15 @@
 /* Square part handling */\
 \
    char trans, uplo, diag; \
-   MKL_INT m, n, lda, incx, incy; \
+   BlasIndex m, n, lda, incx, incy; \
    EIGTYPE const *a; \
-   MKLTYPE alpha_, beta_; \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(beta_, EIGTYPE(1)); \
+   EIGTYPE beta(1); \
 \
 /* Set m, n */ \
-   n = (MKL_INT)size; \
-   lda = lhsStride; \
+   n = convert_index<BlasIndex>(size); \
+   lda = convert_index<BlasIndex>(lhsStride); \
    incx = 1; \
-   incy = resIncr; \
+   incy = convert_index<BlasIndex>(resIncr); \
 \
 /* Set uplo, trans and diag*/ \
    trans = 'N'; \
@@ -123,40 +121,46 @@
    diag = IsUnitDiag ? 'U' : 'N'; \
 \
 /* call ?TRMV*/ \
-   MKLPREFIX##trmv(&uplo, &trans, &diag, &n, (const MKLTYPE*)_lhs, &lda, (MKLTYPE*)x, &incx); \
+   BLASPREFIX##trmv##BLASPOSTFIX(&uplo, &trans, &diag, &n, (const BLASTYPE*)_lhs, &lda, (BLASTYPE*)x, &incx); \
 \
 /* Add op(a_tr)rhs into res*/ \
-   MKLPREFIX##axpy(&n, &alpha_,(const MKLTYPE*)x, &incx, (MKLTYPE*)_res, &incy); \
-/* Non-square case - doesn't fit to MKL ?TRMV. Fall to default triangular product*/ \
+   BLASPREFIX##axpy##BLASPOSTFIX(&n, (const BLASTYPE*)&numext::real_ref(alpha),(const BLASTYPE*)x, &incx, (BLASTYPE*)_res, &incy); \
+/* Non-square case - doesn't fit to BLAS ?TRMV. Fall to default triangular product*/ \
    if (size<(std::max)(rows,cols)) { \
-     typedef Matrix<EIGTYPE, Dynamic, Dynamic> MatrixLhs; \
      if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \
      x = x_tmp.data(); \
      if (size<rows) { \
        y = _res + size*resIncr; \
        a = _lhs + size; \
-       m = rows-size; \
-       n = size; \
+       m = convert_index<BlasIndex>(rows-size); \
+       n = convert_index<BlasIndex>(size); \
      } \
      else { \
        x += size; \
        y = _res; \
        a = _lhs + size*lda; \
-       m = size; \
-       n = cols-size; \
+       m = convert_index<BlasIndex>(size); \
+       n = convert_index<BlasIndex>(cols-size); \
      } \
-     MKLPREFIX##gemv(&trans, &m, &n, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)x, &incx, &beta_, (MKLTYPE*)y, &incy); \
+     BLASPREFIX##gemv##BLASPOSTFIX(&trans, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)x, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)y, &incy); \
    } \
   } \
 };
 
-EIGEN_MKL_TRMV_CM(double, double, d, d)
-EIGEN_MKL_TRMV_CM(dcomplex, MKL_Complex16, cd, z)
-EIGEN_MKL_TRMV_CM(float, float, f, s)
-EIGEN_MKL_TRMV_CM(scomplex, MKL_Complex8, cf, c)
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMV_CM(double,   double, d,  d,)
+EIGEN_BLAS_TRMV_CM(dcomplex, MKL_Complex16, cd, z,)
+EIGEN_BLAS_TRMV_CM(float,    float,  f,  s,)
+EIGEN_BLAS_TRMV_CM(scomplex, MKL_Complex8,  cf, c,)
+#else
+EIGEN_BLAS_TRMV_CM(double,   double, d,  d, _)
+EIGEN_BLAS_TRMV_CM(dcomplex, double, cd, z, _)
+EIGEN_BLAS_TRMV_CM(float,    float,  f,  s, _)
+EIGEN_BLAS_TRMV_CM(scomplex, float,  cf, c, _)
+#endif
 
 // implements row-major: res += alpha * op(triangular) * vector
-#define EIGEN_MKL_TRMV_RM(EIGTYPE, MKLTYPE, EIGPREFIX, MKLPREFIX) \
+#define EIGEN_BLAS_TRMV_RM(EIGTYPE, BLASTYPE, EIGPREFIX, BLASPREFIX, BLASPOSTFIX) \
 template<typename Index, int Mode, bool ConjLhs, bool ConjRhs> \
 struct triangular_matrix_vector_product_trmv<Index,Mode,EIGTYPE,ConjLhs,EIGTYPE,ConjRhs,RowMajor> { \
   enum { \
@@ -190,17 +194,15 @@
 /* Square part handling */\
 \
    char trans, uplo, diag; \
-   MKL_INT m, n, lda, incx, incy; \
+   BlasIndex m, n, lda, incx, incy; \
    EIGTYPE const *a; \
-   MKLTYPE alpha_, beta_; \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(alpha_, alpha); \
-   assign_scalar_eig2mkl<MKLTYPE, EIGTYPE>(beta_, EIGTYPE(1)); \
+   EIGTYPE beta(1); \
 \
 /* Set m, n */ \
-   n = (MKL_INT)size; \
-   lda = lhsStride; \
+   n = convert_index<BlasIndex>(size); \
+   lda = convert_index<BlasIndex>(lhsStride); \
    incx = 1; \
-   incy = resIncr; \
+   incy = convert_index<BlasIndex>(resIncr); \
 \
 /* Set uplo, trans and diag*/ \
    trans = ConjLhs ? 'C' : 'T'; \
@@ -208,40 +210,46 @@
    diag = IsUnitDiag ? 'U' : 'N'; \
 \
 /* call ?TRMV*/ \
-   MKLPREFIX##trmv(&uplo, &trans, &diag, &n, (const MKLTYPE*)_lhs, &lda, (MKLTYPE*)x, &incx); \
+   BLASPREFIX##trmv##BLASPOSTFIX(&uplo, &trans, &diag, &n, (const BLASTYPE*)_lhs, &lda, (BLASTYPE*)x, &incx); \
 \
 /* Add op(a_tr)rhs into res*/ \
-   MKLPREFIX##axpy(&n, &alpha_,(const MKLTYPE*)x, &incx, (MKLTYPE*)_res, &incy); \
-/* Non-square case - doesn't fit to MKL ?TRMV. Fall to default triangular product*/ \
+   BLASPREFIX##axpy##BLASPOSTFIX(&n, (const BLASTYPE*)&numext::real_ref(alpha),(const BLASTYPE*)x, &incx, (BLASTYPE*)_res, &incy); \
+/* Non-square case - doesn't fit to BLAS ?TRMV. Fall to default triangular product*/ \
    if (size<(std::max)(rows,cols)) { \
-     typedef Matrix<EIGTYPE, Dynamic, Dynamic> MatrixLhs; \
      if (ConjRhs) x_tmp = rhs.conjugate(); else x_tmp = rhs; \
      x = x_tmp.data(); \
      if (size<rows) { \
        y = _res + size*resIncr; \
        a = _lhs + size*lda; \
-       m = rows-size; \
-       n = size; \
+       m = convert_index<BlasIndex>(rows-size); \
+       n = convert_index<BlasIndex>(size); \
      } \
      else { \
        x += size; \
        y = _res; \
        a = _lhs + size; \
-       m = size; \
-       n = cols-size; \
+       m = convert_index<BlasIndex>(size); \
+       n = convert_index<BlasIndex>(cols-size); \
      } \
-     MKLPREFIX##gemv(&trans, &n, &m, &alpha_, (const MKLTYPE*)a, &lda, (const MKLTYPE*)x, &incx, &beta_, (MKLTYPE*)y, &incy); \
+     BLASPREFIX##gemv##BLASPOSTFIX(&trans, &n, &m, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (const BLASTYPE*)x, &incx, (const BLASTYPE*)&numext::real_ref(beta), (BLASTYPE*)y, &incy); \
    } \
   } \
 };
 
-EIGEN_MKL_TRMV_RM(double, double, d, d)
-EIGEN_MKL_TRMV_RM(dcomplex, MKL_Complex16, cd, z)
-EIGEN_MKL_TRMV_RM(float, float, f, s)
-EIGEN_MKL_TRMV_RM(scomplex, MKL_Complex8, cf, c)
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRMV_RM(double,   double, d,  d,)
+EIGEN_BLAS_TRMV_RM(dcomplex, MKL_Complex16, cd, z,)
+EIGEN_BLAS_TRMV_RM(float,    float,  f,  s,)
+EIGEN_BLAS_TRMV_RM(scomplex, MKL_Complex8,  cf, c,)
+#else
+EIGEN_BLAS_TRMV_RM(double,   double, d,  d,_)
+EIGEN_BLAS_TRMV_RM(dcomplex, double, cd, z,_)
+EIGEN_BLAS_TRMV_RM(float,    float,  f,  s,_)
+EIGEN_BLAS_TRMV_RM(scomplex, float,  cf, c,_)
+#endif
 
 } // end namespase internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_TRIANGULAR_MATRIX_VECTOR_MKL_H
+#endif // EIGEN_TRIANGULAR_MATRIX_VECTOR_BLAS_H
diff --git a/Eigen/src/Core/products/TriangularSolverMatrix.h b/Eigen/src/Core/products/TriangularSolverMatrix.h
index f103eae..223c38b 100644
--- a/Eigen/src/Core/products/TriangularSolverMatrix.h
+++ b/Eigen/src/Core/products/TriangularSolverMatrix.h
@@ -52,10 +52,14 @@
     level3_blocking<Scalar,Scalar>& blocking)
   {
     Index cols = otherSize;
-    const_blas_data_mapper<Scalar, Index, TriStorageOrder> tri(_tri,triStride);
-    blas_data_mapper<Scalar, Index, ColMajor> other(_other,otherStride);
+
+    typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> TriMapper;
+    typedef blas_data_mapper<Scalar, Index, ColMajor> OtherMapper;
+    TriMapper tri(_tri, triStride);
+    OtherMapper other(_other, otherStride);
 
     typedef gebp_traits<Scalar,Scalar> Traits;
+
     enum {
       SmallPanelWidth   = EIGEN_PLAIN_ENUM_MAX(Traits::mr,Traits::nr),
       IsLower = (Mode&Lower) == Lower
@@ -66,22 +70,20 @@
 
     std::size_t sizeA = kc*mc;
     std::size_t sizeB = kc*cols;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
     conj_if<Conjugate> conj;
-    gebp_kernel<Scalar, Scalar, Index, Traits::mr, Traits::nr, Conjugate, false> gebp_kernel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, TriStorageOrder> pack_lhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr, ColMajor, false, true> pack_rhs;
+    gebp_kernel<Scalar, Scalar, Index, OtherMapper, Traits::mr, Traits::nr, Conjugate, false> gebp_kernel;
+    gemm_pack_lhs<Scalar, Index, TriMapper, Traits::mr, Traits::LhsProgress, TriStorageOrder> pack_lhs;
+    gemm_pack_rhs<Scalar, Index, OtherMapper, Traits::nr, ColMajor, false, true> pack_rhs;
 
     // the goal here is to subdivise the Rhs panels such that we keep some cache
     // coherence when accessing the rhs elements
-    std::ptrdiff_t l1, l2;
-    manage_caching_sizes(GetAction, &l1, &l2);
-    Index subcols = cols>0 ? l2/(4 * sizeof(Scalar) * otherStride) : 0;
+    std::ptrdiff_t l1, l2, l3;
+    manage_caching_sizes(GetAction, &l1, &l2, &l3);
+    Index subcols = cols>0 ? l2/(4 * sizeof(Scalar) * std::max<Index>(otherStride,size)) : 0;
     subcols = std::max<Index>((subcols/Traits::nr)*Traits::nr, Traits::nr);
 
     for(Index k2=IsLower ? 0 : size;
@@ -115,8 +117,9 @@
           {
             // TODO write a small kernel handling this (can be shared with trsv)
             Index i  = IsLower ? k2+k1+k : k2-k1-k-1;
-            Index s  = IsLower ? k2+k1 : i+1;
             Index rs = actualPanelWidth - k - 1; // remaining size
+            Index s  = TriStorageOrder==RowMajor ? (IsLower ? k2+k1 : i+1)
+                                                 :  IsLower ? i+1 : i-rs;
 
             Scalar a = (Mode & UnitDiag) ? Scalar(1) : Scalar(1)/conj(tri(i,i));
             for (Index j=j2; j<j2+actual_cols; ++j)
@@ -133,7 +136,6 @@
               }
               else
               {
-                Index s = IsLower ? i+1 : i-rs;
                 Scalar b = (other(i,j) *= a);
                 Scalar* r = &other(s,j);
                 const Scalar* l = &tri(s,i);
@@ -148,17 +150,17 @@
           Index blockBOffset = IsLower ? k1 : lengthTarget;
 
           // update the respective rows of B from other
-          pack_rhs(blockB+actual_kc*j2, &other(startBlock,j2), otherStride, actualPanelWidth, actual_cols, actual_kc, blockBOffset);
+          pack_rhs(blockB+actual_kc*j2, other.getSubMapper(startBlock,j2), actualPanelWidth, actual_cols, actual_kc, blockBOffset);
 
           // GEBP
           if (lengthTarget>0)
           {
             Index startTarget  = IsLower ? k2+k1+actualPanelWidth : k2-actual_kc;
 
-            pack_lhs(blockA, &tri(startTarget,startBlock), triStride, actualPanelWidth, lengthTarget);
+            pack_lhs(blockA, tri.getSubMapper(startTarget,startBlock), actualPanelWidth, lengthTarget);
 
-            gebp_kernel(&other(startTarget,j2), otherStride, blockA, blockB+actual_kc*j2, lengthTarget, actualPanelWidth, actual_cols, Scalar(-1),
-                        actualPanelWidth, actual_kc, 0, blockBOffset, blockW);
+            gebp_kernel(other.getSubMapper(startTarget,j2), blockA, blockB+actual_kc*j2, lengthTarget, actualPanelWidth, actual_cols, Scalar(-1),
+                        actualPanelWidth, actual_kc, 0, blockBOffset);
           }
         }
       }
@@ -172,16 +174,16 @@
           const Index actual_mc = (std::min)(mc,end-i2);
           if (actual_mc>0)
           {
-            pack_lhs(blockA, &tri(i2, IsLower ? k2 : k2-kc), triStride, actual_kc, actual_mc);
+            pack_lhs(blockA, tri.getSubMapper(i2, IsLower ? k2 : k2-kc), actual_kc, actual_mc);
 
-            gebp_kernel(_other+i2, otherStride, blockA, blockB, actual_mc, actual_kc, cols, Scalar(-1), -1, -1, 0, 0, blockW);
+            gebp_kernel(other.getSubMapper(i2, 0), blockA, blockB, actual_mc, actual_kc, cols, Scalar(-1), -1, -1, 0, 0);
           }
         }
       }
     }
   }
 
-/* Optimized triangular solver with multiple left hand sides and the trinagular matrix on the right
+/* Optimized triangular solver with multiple left hand sides and the triangular matrix on the right
  */
 template <typename Scalar, typename Index, int Mode, bool Conjugate, int TriStorageOrder>
 struct triangular_solve_matrix<Scalar,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor>
@@ -200,8 +202,12 @@
     level3_blocking<Scalar,Scalar>& blocking)
   {
     Index rows = otherSize;
-    const_blas_data_mapper<Scalar, Index, TriStorageOrder> rhs(_tri,triStride);
-    blas_data_mapper<Scalar, Index, ColMajor> lhs(_other,otherStride);
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+
+    typedef blas_data_mapper<Scalar, Index, ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<Scalar, Index, TriStorageOrder> RhsMapper;
+    LhsMapper lhs(_other, otherStride);
+    RhsMapper rhs(_tri, triStride);
 
     typedef gebp_traits<Scalar,Scalar> Traits;
     enum {
@@ -215,17 +221,15 @@
 
     std::size_t sizeA = kc*mc;
     std::size_t sizeB = kc*size;
-    std::size_t sizeW = kc*Traits::WorkSpaceFactor;
 
     ei_declare_aligned_stack_constructed_variable(Scalar, blockA, sizeA, blocking.blockA());
     ei_declare_aligned_stack_constructed_variable(Scalar, blockB, sizeB, blocking.blockB());
-    ei_declare_aligned_stack_constructed_variable(Scalar, blockW, sizeW, blocking.blockW());
 
     conj_if<Conjugate> conj;
-    gebp_kernel<Scalar,Scalar, Index, Traits::mr, Traits::nr, false, Conjugate> gebp_kernel;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder> pack_rhs;
-    gemm_pack_rhs<Scalar, Index, Traits::nr,RhsStorageOrder,false,true> pack_rhs_panel;
-    gemm_pack_lhs<Scalar, Index, Traits::mr, Traits::LhsProgress, ColMajor, false, true> pack_lhs_panel;
+    gebp_kernel<Scalar, Scalar, Index, LhsMapper, Traits::mr, Traits::nr, false, Conjugate> gebp_kernel;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs;
+    gemm_pack_rhs<Scalar, Index, RhsMapper, Traits::nr, RhsStorageOrder,false,true> pack_rhs_panel;
+    gemm_pack_lhs<Scalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, ColMajor, false, true> pack_lhs_panel;
 
     for(Index k2=IsLower ? size : 0;
         IsLower ? k2>0 : k2<size;
@@ -238,7 +242,7 @@
       Index rs = IsLower ? actual_k2 : size - actual_k2 - actual_kc;
       Scalar* geb = blockB+actual_kc*actual_kc;
 
-      if (rs>0) pack_rhs(geb, &rhs(actual_k2,startPanel), triStride, actual_kc, rs);
+      if (rs>0) pack_rhs(geb, rhs.getSubMapper(actual_k2,startPanel), actual_kc, rs);
 
       // triangular packing (we only pack the panels off the diagonal,
       // neglecting the blocks overlapping the diagonal
@@ -252,7 +256,7 @@
 
           if (panelLength>0)
           pack_rhs_panel(blockB+j2*actual_kc,
-                         &rhs(actual_k2+panelOffset, actual_j2), triStride,
+                         rhs.getSubMapper(actual_k2+panelOffset, actual_j2),
                          panelLength, actualPanelWidth,
                          actual_kc, panelOffset);
         }
@@ -280,13 +284,12 @@
             // GEBP
             if(panelLength>0)
             {
-              gebp_kernel(&lhs(i2,absolute_j2), otherStride,
+              gebp_kernel(lhs.getSubMapper(i2,absolute_j2),
                           blockA, blockB+j2*actual_kc,
                           actual_mc, panelLength, actualPanelWidth,
                           Scalar(-1),
                           actual_kc, actual_kc, // strides
-                          panelOffset, panelOffset, // offsets
-                          blockW);  // workspace
+                          panelOffset, panelOffset); // offsets
             }
 
             // unblocked triangular solve
@@ -302,22 +305,25 @@
                 for (Index i=0; i<actual_mc; ++i)
                   r[i] -= a[i] * b;
               }
-              Scalar b = (Mode & UnitDiag) ? Scalar(1) : Scalar(1)/conj(rhs(j,j));
-              for (Index i=0; i<actual_mc; ++i)
-                r[i] *= b;
+              if((Mode & UnitDiag)==0)
+              {
+                Scalar inv_rjj = RealScalar(1)/conj(rhs(j,j));
+                for (Index i=0; i<actual_mc; ++i)
+                  r[i] *= inv_rjj;
+              }
             }
 
             // pack the just computed part of lhs to A
-            pack_lhs_panel(blockA, _other+absolute_j2*otherStride+i2, otherStride,
+            pack_lhs_panel(blockA, LhsMapper(_other+absolute_j2*otherStride+i2, otherStride),
                            actualPanelWidth, actual_mc,
                            actual_kc, j2);
           }
         }
 
         if (rs>0)
-          gebp_kernel(_other+i2+startPanel*otherStride, otherStride, blockA, geb,
+          gebp_kernel(lhs.getSubMapper(i2, startPanel), blockA, geb,
                       actual_mc, actual_kc, rs, Scalar(-1),
-                      -1, -1, 0, 0, blockW);
+                      -1, -1, 0, 0);
       }
     }
   }
diff --git a/Eigen/src/Core/products/TriangularSolverMatrix_MKL.h b/Eigen/src/Core/products/TriangularSolverMatrix_BLAS.h
similarity index 69%
rename from Eigen/src/Core/products/TriangularSolverMatrix_MKL.h
rename to Eigen/src/Core/products/TriangularSolverMatrix_BLAS.h
index 6a0bb83..f077511 100644
--- a/Eigen/src/Core/products/TriangularSolverMatrix_MKL.h
+++ b/Eigen/src/Core/products/TriangularSolverMatrix_BLAS.h
@@ -25,20 +25,20 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to BLAS F77
  *   Triangular matrix * matrix product functionality based on ?TRMM.
  ********************************************************************************
 */
 
-#ifndef EIGEN_TRIANGULAR_SOLVER_MATRIX_MKL_H
-#define EIGEN_TRIANGULAR_SOLVER_MATRIX_MKL_H
+#ifndef EIGEN_TRIANGULAR_SOLVER_MATRIX_BLAS_H
+#define EIGEN_TRIANGULAR_SOLVER_MATRIX_BLAS_H
 
 namespace Eigen {
 
 namespace internal {
 
 // implements LeftSide op(triangular)^-1 * general
-#define EIGEN_MKL_TRSM_L(EIGTYPE, MKLTYPE, MKLPREFIX) \
+#define EIGEN_BLAS_TRSM_L(EIGTYPE, BLASTYPE, BLASFUNC) \
 template <typename Index, int Mode, bool Conjugate, int TriStorageOrder> \
 struct triangular_solve_matrix<EIGTYPE,Index,OnTheLeft,Mode,Conjugate,TriStorageOrder,ColMajor> \
 { \
@@ -53,13 +53,11 @@
       const EIGTYPE* _tri, Index triStride, \
       EIGTYPE* _other, Index otherStride, level3_blocking<EIGTYPE,EIGTYPE>& /*blocking*/) \
   { \
-   MKL_INT m = size, n = otherSize, lda, ldb; \
+   BlasIndex m = convert_index<BlasIndex>(size), n = convert_index<BlasIndex>(otherSize), lda, ldb; \
    char side = 'L', uplo, diag='N', transa; \
    /* Set alpha_ */ \
-   MKLTYPE alpha; \
-   EIGTYPE myone(1); \
-   assign_scalar_eig2mkl(alpha, myone); \
-   ldb = otherStride;\
+   EIGTYPE alpha(1); \
+   ldb = convert_index<BlasIndex>(otherStride);\
 \
    const EIGTYPE *a; \
 /* Set trans */ \
@@ -75,25 +73,31 @@
    if (conjA) { \
      a_tmp = tri.conjugate(); \
      a = a_tmp.data(); \
-     lda = a_tmp.outerStride(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
    } else { \
      a = _tri; \
-     lda = triStride; \
+     lda = convert_index<BlasIndex>(triStride); \
    } \
    if (IsUnitDiag) diag='U'; \
 /* call ?trsm*/ \
-   MKLPREFIX##trsm(&side, &uplo, &transa, &diag, &m, &n, &alpha, (const MKLTYPE*)a, &lda, (MKLTYPE*)_other, &ldb); \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)_other, &ldb); \
  } \
 };
 
-EIGEN_MKL_TRSM_L(double, double, d)
-EIGEN_MKL_TRSM_L(dcomplex, MKL_Complex16, z)
-EIGEN_MKL_TRSM_L(float, float, s)
-EIGEN_MKL_TRSM_L(scomplex, MKL_Complex8, c)
-
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRSM_L(double,   double, dtrsm)
+EIGEN_BLAS_TRSM_L(dcomplex, MKL_Complex16, ztrsm)
+EIGEN_BLAS_TRSM_L(float,    float,  strsm)
+EIGEN_BLAS_TRSM_L(scomplex, MKL_Complex8, ctrsm)
+#else
+EIGEN_BLAS_TRSM_L(double,   double, dtrsm_)
+EIGEN_BLAS_TRSM_L(dcomplex, double, ztrsm_)
+EIGEN_BLAS_TRSM_L(float,    float,  strsm_)
+EIGEN_BLAS_TRSM_L(scomplex, float,  ctrsm_)
+#endif
 
 // implements RightSide general * op(triangular)^-1
-#define EIGEN_MKL_TRSM_R(EIGTYPE, MKLTYPE, MKLPREFIX) \
+#define EIGEN_BLAS_TRSM_R(EIGTYPE, BLASTYPE, BLASFUNC) \
 template <typename Index, int Mode, bool Conjugate, int TriStorageOrder> \
 struct triangular_solve_matrix<EIGTYPE,Index,OnTheRight,Mode,Conjugate,TriStorageOrder,ColMajor> \
 { \
@@ -108,13 +112,11 @@
       const EIGTYPE* _tri, Index triStride, \
       EIGTYPE* _other, Index otherStride, level3_blocking<EIGTYPE,EIGTYPE>& /*blocking*/) \
   { \
-   MKL_INT m = otherSize, n = size, lda, ldb; \
+   BlasIndex m = convert_index<BlasIndex>(otherSize), n = convert_index<BlasIndex>(size), lda, ldb; \
    char side = 'R', uplo, diag='N', transa; \
    /* Set alpha_ */ \
-   MKLTYPE alpha; \
-   EIGTYPE myone(1); \
-   assign_scalar_eig2mkl(alpha, myone); \
-   ldb = otherStride;\
+   EIGTYPE alpha(1); \
+   ldb = convert_index<BlasIndex>(otherStride);\
 \
    const EIGTYPE *a; \
 /* Set trans */ \
@@ -130,26 +132,32 @@
    if (conjA) { \
      a_tmp = tri.conjugate(); \
      a = a_tmp.data(); \
-     lda = a_tmp.outerStride(); \
+     lda = convert_index<BlasIndex>(a_tmp.outerStride()); \
    } else { \
      a = _tri; \
-     lda = triStride; \
+     lda = convert_index<BlasIndex>(triStride); \
    } \
    if (IsUnitDiag) diag='U'; \
 /* call ?trsm*/ \
-   MKLPREFIX##trsm(&side, &uplo, &transa, &diag, &m, &n, &alpha, (const MKLTYPE*)a, &lda, (MKLTYPE*)_other, &ldb); \
+   BLASFUNC(&side, &uplo, &transa, &diag, &m, &n, (const BLASTYPE*)&numext::real_ref(alpha), (const BLASTYPE*)a, &lda, (BLASTYPE*)_other, &ldb); \
    /*std::cout << "TRMS_L specialization!\n";*/ \
  } \
 };
 
-EIGEN_MKL_TRSM_R(double, double, d)
-EIGEN_MKL_TRSM_R(dcomplex, MKL_Complex16, z)
-EIGEN_MKL_TRSM_R(float, float, s)
-EIGEN_MKL_TRSM_R(scomplex, MKL_Complex8, c)
-
+#ifdef EIGEN_USE_MKL
+EIGEN_BLAS_TRSM_R(double,   double, dtrsm)
+EIGEN_BLAS_TRSM_R(dcomplex, MKL_Complex16, ztrsm)
+EIGEN_BLAS_TRSM_R(float,    float,  strsm)
+EIGEN_BLAS_TRSM_R(scomplex, MKL_Complex8,  ctrsm)
+#else
+EIGEN_BLAS_TRSM_R(double,   double, dtrsm_)
+EIGEN_BLAS_TRSM_R(dcomplex, double, ztrsm_)
+EIGEN_BLAS_TRSM_R(float,    float,  strsm_)
+EIGEN_BLAS_TRSM_R(scomplex, float,  ctrsm_)
+#endif
 
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_TRIANGULAR_SOLVER_MATRIX_MKL_H
+#endif // EIGEN_TRIANGULAR_SOLVER_MATRIX_BLAS_H
diff --git a/Eigen/src/Core/products/TriangularSolverVector.h b/Eigen/src/Core/products/TriangularSolverVector.h
index ce4d100..b994759 100644
--- a/Eigen/src/Core/products/TriangularSolverVector.h
+++ b/Eigen/src/Core/products/TriangularSolverVector.h
@@ -10,7 +10,7 @@
 #ifndef EIGEN_TRIANGULAR_SOLVER_VECTOR_H
 #define EIGEN_TRIANGULAR_SOLVER_VECTOR_H
 
-namespace Eigen { 
+namespace Eigen {
 
 namespace internal {
 
@@ -25,7 +25,7 @@
       >::run(size, _lhs, lhsStride, rhs);
   }
 };
-    
+
 // forward and backward substitution, row-major, rhs is a vector
 template<typename LhsScalar, typename RhsScalar, typename Index, int Mode, bool Conjugate>
 struct triangular_solve_vector<LhsScalar, RhsScalar, Index, OnTheLeft, Mode, Conjugate, RowMajor>
@@ -37,6 +37,10 @@
   {
     typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,RowMajor>, 0, OuterStride<> > LhsMap;
     const LhsMap lhs(_lhs,size,size,OuterStride<>(lhsStride));
+
+    typedef const_blas_data_mapper<LhsScalar,Index,RowMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
+
     typename internal::conditional<
                           Conjugate,
                           const CwiseUnaryOp<typename internal::scalar_conjugate_op<LhsScalar>,LhsMap>,
@@ -58,10 +62,10 @@
         Index startRow = IsLower ? pi : pi-actualPanelWidth;
         Index startCol = IsLower ? 0 : pi;
 
-        general_matrix_vector_product<Index,LhsScalar,RowMajor,Conjugate,RhsScalar,false>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,RowMajor,Conjugate,RhsScalar,RhsMapper,false>::run(
           actualPanelWidth, r,
-          &lhs.coeffRef(startRow,startCol), lhsStride,
-          rhs + startCol, 1,
+          LhsMapper(&lhs.coeffRef(startRow,startCol), lhsStride),
+          RhsMapper(rhs + startCol, 1),
           rhs + startRow, 1,
           RhsScalar(-1));
       }
@@ -72,7 +76,7 @@
         Index s = IsLower ? pi   : i+1;
         if (k>0)
           rhs[i] -= (cjLhs.row(i).segment(s,k).transpose().cwiseProduct(Map<const Matrix<RhsScalar,Dynamic,1> >(rhs+s,k))).sum();
-        
+
         if(!(Mode & UnitDiag))
           rhs[i] /= cjLhs(i,i);
       }
@@ -91,6 +95,8 @@
   {
     typedef Map<const Matrix<LhsScalar,Dynamic,Dynamic,ColMajor>, 0, OuterStride<> > LhsMap;
     const LhsMap lhs(_lhs,size,size,OuterStride<>(lhsStride));
+    typedef const_blas_data_mapper<LhsScalar,Index,ColMajor> LhsMapper;
+    typedef const_blas_data_mapper<RhsScalar,Index,ColMajor> RhsMapper;
     typename internal::conditional<Conjugate,
                                    const CwiseUnaryOp<typename internal::scalar_conjugate_op<LhsScalar>,LhsMap>,
                                    const LhsMap&
@@ -122,10 +128,10 @@
         // let's directly call the low level product function because:
         // 1 - it is faster to compile
         // 2 - it is slighlty faster at runtime
-        general_matrix_vector_product<Index,LhsScalar,ColMajor,Conjugate,RhsScalar,false>::run(
+        general_matrix_vector_product<Index,LhsScalar,LhsMapper,ColMajor,Conjugate,RhsScalar,RhsMapper,false>::run(
             r, actualPanelWidth,
-            &lhs.coeffRef(endBlock,startBlock), lhsStride,
-            rhs+startBlock, 1,
+            LhsMapper(&lhs.coeffRef(endBlock,startBlock), lhsStride),
+            RhsMapper(rhs+startBlock, 1),
             rhs+endBlock, 1, RhsScalar(-1));
       }
     }
diff --git a/Eigen/src/Core/util/BlasUtil.h b/Eigen/src/Core/util/BlasUtil.h
old mode 100644
new mode 100755
index a28f16f..6e6ee11
--- a/Eigen/src/Core/util/BlasUtil.h
+++ b/Eigen/src/Core/util/BlasUtil.h
@@ -18,13 +18,13 @@
 namespace internal {
 
 // forward declarations
-template<typename LhsScalar, typename RhsScalar, typename Index, int mr, int nr, bool ConjugateLhs=false, bool ConjugateRhs=false>
+template<typename LhsScalar, typename RhsScalar, typename Index, typename DataMapper, int mr, int nr, bool ConjugateLhs=false, bool ConjugateRhs=false>
 struct gebp_kernel;
 
-template<typename Scalar, typename Index, int nr, int StorageOrder, bool Conjugate = false, bool PanelMode=false>
+template<typename Scalar, typename Index, typename DataMapper, int nr, int StorageOrder, bool Conjugate = false, bool PanelMode=false>
 struct gemm_pack_rhs;
 
-template<typename Scalar, typename Index, int Pack1, int Pack2, int StorageOrder, bool Conjugate = false, bool PanelMode = false>
+template<typename Scalar, typename Index, typename DataMapper, int Pack1, int Pack2, int StorageOrder, bool Conjugate = false, bool PanelMode = false>
 struct gemm_pack_lhs;
 
 template<
@@ -34,7 +34,9 @@
   int ResStorageOrder>
 struct general_matrix_matrix_product;
 
-template<typename Index, typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs, typename RhsScalar, bool ConjugateRhs, int Version=Specialized>
+template<typename Index,
+         typename LhsScalar, typename LhsMapper, int LhsStorageOrder, bool ConjugateLhs,
+         typename RhsScalar, typename RhsMapper, bool ConjugateRhs, int Version=Specialized>
 struct general_matrix_vector_product;
 
 
@@ -42,22 +44,35 @@
 
 template<> struct conj_if<true> {
   template<typename T>
-  inline T operator()(const T& x) { return numext::conj(x); }
+  inline T operator()(const T& x) const { return numext::conj(x); }
   template<typename T>
-  inline T pconj(const T& x) { return internal::pconj(x); }
+  inline T pconj(const T& x) const { return internal::pconj(x); }
 };
 
 template<> struct conj_if<false> {
   template<typename T>
-  inline const T& operator()(const T& x) { return x; }
+  inline const T& operator()(const T& x) const { return x; }
   template<typename T>
-  inline const T& pconj(const T& x) { return x; }
+  inline const T& pconj(const T& x) const { return x; }
+};
+
+// Generic implementation for custom complex types.
+template<typename LhsScalar, typename RhsScalar, bool ConjLhs, bool ConjRhs>
+struct conj_helper
+{
+  typedef typename ScalarBinaryOpTraits<LhsScalar,RhsScalar>::ReturnType Scalar;
+
+  EIGEN_STRONG_INLINE Scalar pmadd(const LhsScalar& x, const RhsScalar& y, const Scalar& c) const
+  { return padd(c, pmul(x,y)); }
+
+  EIGEN_STRONG_INLINE Scalar pmul(const LhsScalar& x, const RhsScalar& y) const
+  { return conj_if<ConjLhs>()(x) *  conj_if<ConjRhs>()(y); }
 };
 
 template<typename Scalar> struct conj_helper<Scalar,Scalar,false,false>
 {
-  EIGEN_STRONG_INLINE Scalar pmadd(const Scalar& x, const Scalar& y, const Scalar& c) const { return internal::pmadd(x,y,c); }
-  EIGEN_STRONG_INLINE Scalar pmul(const Scalar& x, const Scalar& y) const { return internal::pmul(x,y); }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar pmadd(const Scalar& x, const Scalar& y, const Scalar& c) const { return internal::pmadd(x,y,c); }
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar pmul(const Scalar& x, const Scalar& y) const { return internal::pmul(x,y); }
 };
 
 template<typename RealScalar> struct conj_helper<std::complex<RealScalar>, std::complex<RealScalar>, false,true>
@@ -109,39 +124,142 @@
 };
 
 template<typename From,typename To> struct get_factor {
-  static EIGEN_STRONG_INLINE To run(const From& x) { return x; }
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE To run(const From& x) { return To(x); }
 };
 
 template<typename Scalar> struct get_factor<Scalar,typename NumTraits<Scalar>::Real> {
+  EIGEN_DEVICE_FUNC
   static EIGEN_STRONG_INLINE typename NumTraits<Scalar>::Real run(const Scalar& x) { return numext::real(x); }
 };
 
-// Lightweight helper class to access matrix coefficients.
-// Yes, this is somehow redundant with Map<>, but this version is much much lighter,
-// and so I hope better compilation performance (time and code quality).
-template<typename Scalar, typename Index, int StorageOrder>
-class blas_data_mapper
-{
+
+template<typename Scalar, typename Index>
+class BlasVectorMapper {
   public:
-    blas_data_mapper(Scalar* data, Index stride) : m_data(data), m_stride(stride) {}
-    EIGEN_STRONG_INLINE Scalar& operator()(Index i, Index j)
-    { return m_data[StorageOrder==RowMajor ? j + i*m_stride : i + j*m_stride]; }
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE BlasVectorMapper(Scalar *data) : m_data(data) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar operator()(Index i) const {
+    return m_data[i];
+  }
+  template <typename Packet, int AlignmentType>
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet load(Index i) const {
+    return ploadt<Packet, AlignmentType>(m_data + i);
+  }
+
+  template <typename Packet>
+  EIGEN_DEVICE_FUNC bool aligned(Index i) const {
+    return (UIntPtr(m_data+i)%sizeof(Packet))==0;
+  }
+
   protected:
-    Scalar* EIGEN_RESTRICT m_data;
-    Index m_stride;
+  Scalar* m_data;
+};
+
+template<typename Scalar, typename Index, int AlignmentType>
+class BlasLinearMapper {
+  public:
+  typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename packet_traits<Scalar>::half HalfPacket;
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE BlasLinearMapper(Scalar *data) : m_data(data) {}
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void prefetch(int i) const {
+    internal::prefetch(&operator()(i));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Scalar& operator()(Index i) const {
+    return m_data[i];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i) const {
+    return ploadt<Packet, AlignmentType>(m_data + i);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE HalfPacket loadHalfPacket(Index i) const {
+    return ploadt<HalfPacket, AlignmentType>(m_data + i);
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void storePacket(Index i, const Packet &p) const {
+    pstoret<Scalar, Packet, AlignmentType>(m_data + i, p);
+  }
+
+  protected:
+  Scalar *m_data;
+};
+
+// Lightweight helper class to access matrix coefficients.
+template<typename Scalar, typename Index, int StorageOrder, int AlignmentType = Unaligned>
+class blas_data_mapper {
+  public:
+  typedef typename packet_traits<Scalar>::type Packet;
+  typedef typename packet_traits<Scalar>::half HalfPacket;
+
+  typedef BlasLinearMapper<Scalar, Index, AlignmentType> LinearMapper;
+  typedef BlasVectorMapper<Scalar, Index> VectorMapper;
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE blas_data_mapper(Scalar* data, Index stride) : m_data(data), m_stride(stride) {}
+
+  EIGEN_DEVICE_FUNC  EIGEN_ALWAYS_INLINE blas_data_mapper<Scalar, Index, StorageOrder, AlignmentType>
+  getSubMapper(Index i, Index j) const {
+    return blas_data_mapper<Scalar, Index, StorageOrder, AlignmentType>(&operator()(i, j), m_stride);
+  }
+
+  EIGEN_DEVICE_FUNC  EIGEN_ALWAYS_INLINE LinearMapper getLinearMapper(Index i, Index j) const {
+    return LinearMapper(&operator()(i, j));
+  }
+
+  EIGEN_DEVICE_FUNC  EIGEN_ALWAYS_INLINE VectorMapper getVectorMapper(Index i, Index j) const {
+    return VectorMapper(&operator()(i, j));
+  }
+
+
+  EIGEN_DEVICE_FUNC
+  EIGEN_ALWAYS_INLINE Scalar& operator()(Index i, Index j) const {
+    return m_data[StorageOrder==RowMajor ? j + i*m_stride : i + j*m_stride];
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet loadPacket(Index i, Index j) const {
+    return ploadt<Packet, AlignmentType>(&operator()(i, j));
+  }
+
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE HalfPacket loadHalfPacket(Index i, Index j) const {
+    return ploadt<HalfPacket, AlignmentType>(&operator()(i, j));
+  }
+
+  template<typename SubPacket>
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void scatterPacket(Index i, Index j, const SubPacket &p) const {
+    pscatter<Scalar, SubPacket>(&operator()(i, j), p, m_stride);
+  }
+
+  template<typename SubPacket>
+  EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE SubPacket gatherPacket(Index i, Index j) const {
+    return pgather<Scalar, SubPacket>(&operator()(i, j), m_stride);
+  }
+
+  EIGEN_DEVICE_FUNC const Index stride() const { return m_stride; }
+  EIGEN_DEVICE_FUNC const Scalar* data() const { return m_data; }
+
+  EIGEN_DEVICE_FUNC Index firstAligned(Index size) const {
+    if (UIntPtr(m_data)%sizeof(Scalar)) {
+      return -1;
+    }
+    return internal::first_default_aligned(m_data, size);
+  }
+
+  protected:
+  Scalar* EIGEN_RESTRICT m_data;
+  const Index m_stride;
 };
 
 // lightweight helper class to access matrix coefficients (const version)
 template<typename Scalar, typename Index, int StorageOrder>
-class const_blas_data_mapper
-{
+class const_blas_data_mapper : public blas_data_mapper<const Scalar, Index, StorageOrder> {
   public:
-    const_blas_data_mapper(const Scalar* data, Index stride) : m_data(data), m_stride(stride) {}
-    EIGEN_STRONG_INLINE const Scalar& operator()(Index i, Index j) const
-    { return m_data[StorageOrder==RowMajor ? j + i*m_stride : i + j*m_stride]; }
-  protected:
-    const Scalar* EIGEN_RESTRICT m_data;
-    Index m_stride;
+  EIGEN_ALWAYS_INLINE const_blas_data_mapper(const Scalar *data, Index stride) : blas_data_mapper<const Scalar, Index, StorageOrder>(data, stride) {}
+
+  EIGEN_ALWAYS_INLINE const_blas_data_mapper<Scalar, Index, StorageOrder> getSubMapper(Index i, Index j) const {
+    return const_blas_data_mapper<Scalar, Index, StorageOrder>(&(this->operator()(i, j)), this->m_stride);
+  }
 };
 
 
@@ -188,17 +306,33 @@
 };
 
 // pop scalar multiple
-template<typename Scalar, typename NestedXpr>
-struct blas_traits<CwiseUnaryOp<scalar_multiple_op<Scalar>, NestedXpr> >
+template<typename Scalar, typename NestedXpr, typename Plain>
+struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> >
  : blas_traits<NestedXpr>
 {
   typedef blas_traits<NestedXpr> Base;
-  typedef CwiseUnaryOp<scalar_multiple_op<Scalar>, NestedXpr> XprType;
+  typedef CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain>, NestedXpr> XprType;
   typedef typename Base::ExtractType ExtractType;
-  static inline ExtractType extract(const XprType& x) { return Base::extract(x.nestedExpression()); }
+  static inline ExtractType extract(const XprType& x) { return Base::extract(x.rhs()); }
   static inline Scalar extractScalarFactor(const XprType& x)
-  { return x.functor().m_other * Base::extractScalarFactor(x.nestedExpression()); }
+  { return x.lhs().functor().m_other * Base::extractScalarFactor(x.rhs()); }
 };
+template<typename Scalar, typename NestedXpr, typename Plain>
+struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > >
+ : blas_traits<NestedXpr>
+{
+  typedef blas_traits<NestedXpr> Base;
+  typedef CwiseBinaryOp<scalar_product_op<Scalar>, NestedXpr, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain> > XprType;
+  typedef typename Base::ExtractType ExtractType;
+  static inline ExtractType extract(const XprType& x) { return Base::extract(x.lhs()); }
+  static inline Scalar extractScalarFactor(const XprType& x)
+  { return Base::extractScalarFactor(x.lhs()) * x.rhs().functor().m_other; }
+};
+template<typename Scalar, typename Plain1, typename Plain2>
+struct blas_traits<CwiseBinaryOp<scalar_product_op<Scalar>, const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain1>,
+                                                            const CwiseNullaryOp<scalar_constant_op<Scalar>,Plain2> > >
+ : blas_traits<CwiseNullaryOp<scalar_constant_op<Scalar>,Plain1> >
+{};
 
 // pop opposite
 template<typename Scalar, typename NestedXpr>
@@ -230,7 +364,7 @@
   enum {
     IsTransposed = Base::IsTransposed ? 0 : 1
   };
-  static inline ExtractType extract(const XprType& x) { return Base::extract(x.nestedExpression()); }
+  static inline ExtractType extract(const XprType& x) { return ExtractType(Base::extract(x.nestedExpression())); }
   static inline Scalar extractScalarFactor(const XprType& x) { return Base::extractScalarFactor(x.nestedExpression()); }
 };
 
diff --git a/Eigen/src/Core/util/CMakeLists.txt b/Eigen/src/Core/util/CMakeLists.txt
deleted file mode 100644
index a1e2e52..0000000
--- a/Eigen/src/Core/util/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Core_util_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_Core_util_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core/util COMPONENT Devel
-  )
diff --git a/Eigen/src/Core/util/Constants.h b/Eigen/src/Core/util/Constants.h
index 1e6277c..7587d68 100644
--- a/Eigen/src/Core/util/Constants.h
+++ b/Eigen/src/Core/util/Constants.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2007-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -30,6 +30,14 @@
   */
 const int Infinity = -1;
 
+/** This value means that the cost to evaluate an expression coefficient is either very expensive or
+  * cannot be known at compile time.
+  *
+  * This value has to be positive to (1) simplify cost computation, and (2) allow to distinguish between a very expensive and very very expensive expressions.
+  * It thus must also be large enough to make sure unrolling won't happen and that sub expressions will be evaluated, but not too large to avoid overflow.
+  */
+const int HugeCost = 10000;
+
 /** \defgroup flags Flags
   * \ingroup Core_Module
   *
@@ -48,19 +56,19 @@
   * for a matrix, this means that the storage order is row-major.
   * If this bit is not set, the storage order is column-major.
   * For an expression, this determines the storage order of
-  * the matrix created by evaluation of that expression. 
-  * \sa \ref TopicStorageOrders */
+  * the matrix created by evaluation of that expression.
+  * \sa \blank  \ref TopicStorageOrders */
 const unsigned int RowMajorBit = 0x1;
 
 /** \ingroup flags
-  *
   * means the expression should be evaluated by the calling expression */
 const unsigned int EvalBeforeNestingBit = 0x2;
 
 /** \ingroup flags
-  *
+  * \deprecated
   * means the expression should be evaluated before any assignment */
-const unsigned int EvalBeforeAssigningBit = 0x4;
+EIGEN_DEPRECATED
+const unsigned int EvalBeforeAssigningBit = 0x4; // FIXME deprecated
 
 /** \ingroup flags
   *
@@ -141,17 +149,46 @@
   */
 const unsigned int DirectAccessBit = 0x40;
 
-/** \ingroup flags
+/** \deprecated \ingroup flags
   *
-  * means the first coefficient packet is guaranteed to be aligned */
-const unsigned int AlignedBit = 0x80;
+  * means the first coefficient packet is guaranteed to be aligned.
+  * An expression cannot has the AlignedBit without the PacketAccessBit flag.
+  * In other words, this means we are allow to perform an aligned packet access to the first element regardless
+  * of the expression kind:
+  * \code
+  * expression.packet<Aligned>(0);
+  * \endcode
+  */
+EIGEN_DEPRECATED const unsigned int AlignedBit = 0x80;
 
 const unsigned int NestByRefBit = 0x100;
 
+/** \ingroup flags
+  *
+  * for an expression, this means that the storage order
+  * can be either row-major or column-major.
+  * The precise choice will be decided at evaluation time or when
+  * combined with other expressions.
+  * \sa \blank  \ref RowMajorBit, \ref TopicStorageOrders */
+const unsigned int NoPreferredStorageOrderBit = 0x200;
+
+/** \ingroup flags
+  *
+  * Means that the underlying coefficients can be accessed through pointers to the sparse (un)compressed storage format,
+  * that is, the expression provides:
+  * \code
+    inline const Scalar* valuePtr() const;
+    inline const Index* innerIndexPtr() const;
+    inline const Index* outerIndexPtr() const;
+    inline const Index* innerNonZeroPtr() const;
+    \endcode
+  */
+const unsigned int CompressedAccessBit = 0x400;
+
+
 // list of flags that are inherited by default
 const unsigned int HereditaryBits = RowMajorBit
-                                  | EvalBeforeNestingBit
-                                  | EvalBeforeAssigningBit;
+                                  | EvalBeforeNestingBit;
 
 /** \defgroup enums Enumerations
   * \ingroup Core_Module
@@ -160,9 +197,9 @@
   */
 
 /** \ingroup enums
-  * Enum containing possible values for the \p Mode parameter of 
-  * MatrixBase::selfadjointView() and MatrixBase::triangularView(). */
-enum {
+  * Enum containing possible values for the \c Mode or \c UpLo parameter of
+  * MatrixBase::selfadjointView() and MatrixBase::triangularView(), and selfadjoint solvers. */
+enum UpLoType {
   /** View matrix as a lower triangular matrix. */
   Lower=0x1,                      
   /** View matrix as an upper triangular matrix. */
@@ -186,12 +223,31 @@
 };
 
 /** \ingroup enums
-  * Enum for indicating whether an object is aligned or not. */
-enum { 
-  /** Object is not correctly aligned for vectorization. */
-  Unaligned=0, 
-  /** Object is aligned for vectorization. */
-  Aligned=1 
+  * Enum for indicating whether a buffer is aligned or not. */
+enum AlignmentType {
+  Unaligned=0,        /**< Data pointer has no specific alignment. */
+  Aligned8=8,         /**< Data pointer is aligned on a 8 bytes boundary. */
+  Aligned16=16,       /**< Data pointer is aligned on a 16 bytes boundary. */
+  Aligned32=32,       /**< Data pointer is aligned on a 32 bytes boundary. */
+  Aligned64=64,       /**< Data pointer is aligned on a 64 bytes boundary. */
+  Aligned128=128,     /**< Data pointer is aligned on a 128 bytes boundary. */
+  AlignedMask=255,
+  Aligned=16,         /**< \deprecated Synonym for Aligned16. */
+#if EIGEN_MAX_ALIGN_BYTES==128
+  AlignedMax = Aligned128
+#elif EIGEN_MAX_ALIGN_BYTES==64
+  AlignedMax = Aligned64
+#elif EIGEN_MAX_ALIGN_BYTES==32
+  AlignedMax = Aligned32
+#elif EIGEN_MAX_ALIGN_BYTES==16
+  AlignedMax = Aligned16
+#elif EIGEN_MAX_ALIGN_BYTES==8
+  AlignedMax = Aligned8
+#elif EIGEN_MAX_ALIGN_BYTES==0
+  AlignedMax = Unaligned
+#else
+#error Invalid value for EIGEN_MAX_ALIGN_BYTES
+#endif
 };
 
 /** \ingroup enums
@@ -217,7 +273,7 @@
 
 /** \internal \ingroup enums
   * Enum to specify how to traverse the entries of a matrix. */
-enum {
+enum TraversalType {
   /** \internal Default traversal, no vectorization, no index-based access */
   DefaultTraversal,
   /** \internal No vectorization, use index-based access to have only one for loop instead of 2 nested loops */
@@ -239,7 +295,7 @@
 
 /** \internal \ingroup enums
   * Enum to specify whether to unroll loops when traversing over the entries of a matrix. */
-enum {
+enum UnrollingType {
   /** \internal Do not unroll loops. */
   NoUnrolling,
   /** \internal Unroll only the inner loop, but not the outer loop. */
@@ -251,7 +307,7 @@
 
 /** \internal \ingroup enums
   * Enum to specify whether to use the default (built-in) implementation or the specialization. */
-enum {
+enum SpecializedType {
   Specialized,
   BuiltIn
 };
@@ -259,7 +315,7 @@
 /** \ingroup enums
   * Enum containing possible values for the \p _Options template parameter of
   * Matrix, Array and BandMatrix. */
-enum {
+enum StorageOptions {
   /** Storage order is column major (see \ref TopicStorageOrders). */
   ColMajor = 0,
   /** Storage order is row major (see \ref TopicStorageOrders). */
@@ -272,7 +328,7 @@
 
 /** \ingroup enums
   * Enum for specifying whether to apply or solve on the left or right. */
-enum {
+enum SideType {
   /** Apply transformation on the left. */
   OnTheLeft = 1,  
   /** Apply transformation on the right. */
@@ -297,7 +353,7 @@
 
 /** \internal \ingroup enums
   * Used in AmbiVector. */
-enum {
+enum AmbiVectorMode {
   IsDense         = 0,
   IsSparse
 };
@@ -406,10 +462,16 @@
     Generic = 0x0,
     SSE = 0x1,
     AltiVec = 0x2,
+    VSX = 0x3,
+    NEON = 0x4,
 #if defined EIGEN_VECTORIZE_SSE
     Target = SSE
 #elif defined EIGEN_VECTORIZE_ALTIVEC
     Target = AltiVec
+#elif defined EIGEN_VECTORIZE_VSX
+    Target = VSX
+#elif defined EIGEN_VECTORIZE_NEON
+    Target = NEON
 #else
     Target = Generic
 #endif
@@ -417,8 +479,9 @@
 }
 
 /** \internal \ingroup enums
-  * Enum used as template parameter in GeneralProduct. */
-enum { CoeffBasedProductMode, LazyCoeffBasedProductMode, OuterProduct, InnerProduct, GemvProduct, GemmProduct };
+  * Enum used as template parameter in Product and product evaluators. */
+enum ProductImplType
+{ DefaultProduct=0, LazyProduct, AliasFreeProduct, CoeffBasedProductMode, LazyCoeffBasedProductMode, OuterProduct, InnerProduct, GemvProduct, GemmProduct };
 
 /** \internal \ingroup enums
   * Enum used in experimental parallel implementation. */
@@ -427,24 +490,57 @@
 /** The type used to identify a dense storage. */
 struct Dense {};
 
+/** The type used to identify a general sparse storage. */
+struct Sparse {};
+
+/** The type used to identify a general solver (factored) storage. */
+struct SolverStorage {};
+
+/** The type used to identify a permutation storage. */
+struct PermutationStorage {};
+
+/** The type used to identify a permutation storage. */
+struct TranspositionsStorage {};
+
 /** The type used to identify a matrix expression */
 struct MatrixXpr {};
 
 /** The type used to identify an array expression */
 struct ArrayXpr {};
 
+// An evaluator must define its shape. By default, it can be one of the following:
+struct DenseShape             { static std::string debugName() { return "DenseShape"; } };
+struct SolverShape            { static std::string debugName() { return "SolverShape"; } };
+struct HomogeneousShape       { static std::string debugName() { return "HomogeneousShape"; } };
+struct DiagonalShape          { static std::string debugName() { return "DiagonalShape"; } };
+struct BandShape              { static std::string debugName() { return "BandShape"; } };
+struct TriangularShape        { static std::string debugName() { return "TriangularShape"; } };
+struct SelfAdjointShape       { static std::string debugName() { return "SelfAdjointShape"; } };
+struct PermutationShape       { static std::string debugName() { return "PermutationShape"; } };
+struct TranspositionsShape    { static std::string debugName() { return "TranspositionsShape"; } };
+struct SparseShape            { static std::string debugName() { return "SparseShape"; } };
+
 namespace internal {
-  /** \internal
-  * Constants for comparison functors
-  */
-  enum ComparisonName {
-    cmp_EQ = 0,
-    cmp_LT = 1,
-    cmp_LE = 2,
-    cmp_UNORD = 3,
-    cmp_NEQ = 4
-  };
-}
+
+  // random access iterators based on coeff*() accessors.
+struct IndexBased {};
+
+// evaluator based on iterators to access coefficients. 
+struct IteratorBased {};
+
+/** \internal
+ * Constants for comparison functors
+ */
+enum ComparisonName {
+  cmp_EQ = 0,
+  cmp_LT = 1,
+  cmp_LE = 2,
+  cmp_UNORD = 3,
+  cmp_NEQ = 4,
+  cmp_GT = 5,
+  cmp_GE = 6
+};
+} // end namespace internal
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/Core/util/DisableStupidWarnings.h b/Eigen/src/Core/util/DisableStupidWarnings.h
old mode 100644
new mode 100755
index 6a0bf06..351bd6c
--- a/Eigen/src/Core/util/DisableStupidWarnings.h
+++ b/Eigen/src/Core/util/DisableStupidWarnings.h
@@ -10,24 +10,31 @@
   // 4244 - 'argument' : conversion from 'type1' to 'type2', possible loss of data
   // 4273 - QtAlignedMalloc, inconsistent DLL linkage
   // 4324 - structure was padded due to declspec(align())
+  // 4503 - decorated name length exceeded, name was truncated
   // 4512 - assignment operator could not be generated
   // 4522 - 'class' : multiple assignment operators specified
   // 4700 - uninitialized local variable 'xyz' used
+  // 4714 - function marked as __forceinline not inlined
   // 4717 - 'function' : recursive on all control paths, function will cause runtime stack overflow
+  // 4800 - 'type' : forcing value to bool 'true' or 'false' (performance warning)
   #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS
     #pragma warning( push )
   #endif
-  #pragma warning( disable : 4100 4101 4127 4181 4211 4244 4273 4324 4512 4522 4700 4717 )
+  #pragma warning( disable : 4100 4101 4127 4181 4211 4244 4273 4324 4503 4512 4522 4700 4714 4717 4800)
+
 #elif defined __INTEL_COMPILER
   // 2196 - routine is both "inline" and "noinline" ("noinline" assumed)
   //        ICC 12 generates this warning even without any inline keyword, when defining class methods 'inline' i.e. inside of class body
   //        typedef that may be a reference type.
   // 279  - controlling expression is constant
   //        ICC 12 generates this warning on assert(constant_expression_depending_on_template_params) and frankly this is a legitimate use case.
+  // 1684 - conversion from pointer to same-sized integral type (potential portability problem)
+  // 2259 - non-pointer conversion from "Eigen::Index={ptrdiff_t={long}}" to "int" may lose significant bits
   #ifndef EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS
     #pragma warning push
   #endif
-  #pragma warning disable 2196 279
+  #pragma warning disable 2196 279 1684 2259
+
 #elif defined __clang__
   // -Wconstant-logical-operand - warning: use of logical && with constant operand; switch to bitwise & or remove constant
   //     this is really a stupid warning as it warns on compile-time expressions involving enums
@@ -35,6 +42,42 @@
     #pragma clang diagnostic push
   #endif
   #pragma clang diagnostic ignored "-Wconstant-logical-operand"
+
+#elif defined __GNUC__
+
+  #if (!defined(EIGEN_PERMANENTLY_DISABLE_STUPID_WARNINGS)) &&  (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))
+    #pragma GCC diagnostic push
+  #endif
+  // g++ warns about local variables shadowing member functions, which is too strict
+  #pragma GCC diagnostic ignored "-Wshadow"
+  #if __GNUC__ == 4 && __GNUC_MINOR__ < 8
+    // Until g++-4.7 there are warnings when comparing unsigned int vs 0, even in templated functions:
+    #pragma GCC diagnostic ignored "-Wtype-limits"
+  #endif
+  #if __GNUC__>=6
+    #pragma GCC diagnostic ignored "-Wignored-attributes"
+  #endif
+
+#endif
+
+#if defined __NVCC__
+  // Disable the "statement is unreachable" message
+  #pragma diag_suppress code_is_unreachable
+  // Disable the "dynamic initialization in unreachable code" message
+  #pragma diag_suppress initialization_not_reachable
+  // Disable the "invalid error number" message that we get with older versions of nvcc
+  #pragma diag_suppress 1222
+  // Disable the "calling a __host__ function from a __host__ __device__ function is not allowed" messages (yes, there are many of them and they seem to change with every version of the compiler)
+  #pragma diag_suppress 2527
+  #pragma diag_suppress 2529
+  #pragma diag_suppress 2651
+  #pragma diag_suppress 2653
+  #pragma diag_suppress 2668
+  #pragma diag_suppress 2669
+  #pragma diag_suppress 2670
+  #pragma diag_suppress 2671
+  #pragma diag_suppress 2735
+  #pragma diag_suppress 2737
 #endif
 
 #endif // not EIGEN_WARNINGS_DISABLED
diff --git a/Eigen/src/Core/util/ForwardDeclarations.h b/Eigen/src/Core/util/ForwardDeclarations.h
index d6a8145..ea10739 100644
--- a/Eigen/src/Core/util/ForwardDeclarations.h
+++ b/Eigen/src/Core/util/ForwardDeclarations.h
@@ -36,6 +36,10 @@
   };
 };
 
+template<typename T> struct evaluator_traits;
+
+template< typename T> struct evaluator;
+
 } // end namespace internal
 
 template<typename T> struct NumTraits;
@@ -51,18 +55,18 @@
 
 template<typename _Scalar, int _Rows, int _Cols,
          int _Options = AutoAlign |
-#if defined(__GNUC__) && __GNUC__==3 && __GNUC_MINOR__==4
+#if EIGEN_GNUC_AT(3,4)
     // workaround a bug in at least gcc 3.4.6
     // the innermost ?: ternary operator is misparsed. We write it slightly
     // differently and this makes gcc 3.4.6 happy, but it's ugly.
     // The error would only show up with EIGEN_DEFAULT_TO_ROW_MAJOR is defined
     // (when EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION is RowMajor)
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
                           : !(_Cols==1 && _Rows!=1) ?  EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION
-                          : ColMajor ),
+                          : Eigen::ColMajor ),
 #else
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
-                          : (_Cols==1 && _Rows!=1) ? ColMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
+                          : (_Cols==1 && _Rows!=1) ? Eigen::ColMajor
                           : EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ),
 #endif
          int _MaxRows = _Rows,
@@ -87,10 +91,11 @@
 template<typename UnaryOp,   typename MatrixType>         class CwiseUnaryOp;
 template<typename ViewOp,    typename MatrixType>         class CwiseUnaryView;
 template<typename BinaryOp,  typename Lhs, typename Rhs>  class CwiseBinaryOp;
-template<typename BinOp,     typename Lhs, typename Rhs>  class SelfCwiseBinaryOp;
-template<typename Derived,   typename Lhs, typename Rhs>  class ProductBase;
-template<typename Lhs, typename Rhs, int Mode>            class GeneralProduct;
-template<typename Lhs, typename Rhs, int NestingFlags>    class CoeffBasedProduct;
+template<typename TernaryOp, typename Arg1, typename Arg2, typename Arg3>  class CwiseTernaryOp;
+template<typename Decomposition, typename Rhstype>        class Solve;
+template<typename XprType>                                class Inverse;
+
+template<typename Lhs, typename Rhs, int Option = DefaultProduct> class Product;
 
 template<typename Derived> class DiagonalBase;
 template<typename _DiagonalVectorType> class DiagonalWrapper;
@@ -108,7 +113,12 @@
          int Level = internal::accessors_level<Derived>::has_write_access ? WriteAccessors : ReadOnlyAccessors
 > class MapBase;
 template<int InnerStrideAtCompileTime, int OuterStrideAtCompileTime> class Stride;
+template<int Value = Dynamic> class InnerStride;
+template<int Value = Dynamic> class OuterStride;
 template<typename MatrixType, int MapOptions=Unaligned, typename StrideType = Stride<0,0> > class Map;
+template<typename Derived> class RefBase;
+template<typename PlainObjectType, int Options = 0,
+         typename StrideType = typename internal::conditional<PlainObjectType::IsVectorAtCompileTime,InnerStride<1>,OuterStride<> >::type > class Ref;
 
 template<typename Derived> class TriangularBase;
 template<typename MatrixType, unsigned int Mode> class TriangularView;
@@ -119,10 +129,10 @@
 template<typename Derived> class ReturnByValue;
 template<typename ExpressionType> class ArrayWrapper;
 template<typename ExpressionType> class MatrixWrapper;
+template<typename Derived> class SolverBase;
+template<typename XprType> class InnerIterator;
 
 namespace internal {
-template<typename DecompositionType, typename Rhs> struct solve_retval_base;
-template<typename DecompositionType, typename Rhs> struct solve_retval;
 template<typename DecompositionType> struct kernel_retval_base;
 template<typename DecompositionType> struct kernel_retval;
 template<typename DecompositionType> struct image_retval_base;
@@ -135,6 +145,21 @@
 
 namespace internal {
 template<typename Lhs, typename Rhs> struct product_type;
+
+template<bool> struct EnableIf;
+
+/** \internal
+  * \class product_evaluator
+  * Products need their own evaluator with more template arguments allowing for
+  * easier partial template specializations.
+  */
+template< typename T,
+          int ProductTag = internal::product_type<typename T::Lhs,typename T::Rhs>::ret,
+          typename LhsShape = typename evaluator_traits<typename T::Lhs>::Shape,
+          typename RhsShape = typename evaluator_traits<typename T::Rhs>::Shape,
+          typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
+          typename RhsScalar = typename traits<typename T::Rhs>::Scalar
+        > struct product_evaluator;
 }
 
 template<typename Lhs, typename Rhs,
@@ -150,9 +175,11 @@
 // with optional conjugation of the arguments.
 template<typename LhsScalar, typename RhsScalar, bool ConjLhs=false, bool ConjRhs=false> struct conj_helper;
 
-template<typename Scalar> struct scalar_sum_op;
-template<typename Scalar> struct scalar_difference_op;
-template<typename LhsScalar,typename RhsScalar> struct scalar_conj_product_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_sum_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_difference_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_conj_product_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_min_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_max_op;
 template<typename Scalar> struct scalar_opposite_op;
 template<typename Scalar> struct scalar_conjugate_op;
 template<typename Scalar> struct scalar_real_op;
@@ -160,6 +187,7 @@
 template<typename Scalar> struct scalar_abs_op;
 template<typename Scalar> struct scalar_abs2_op;
 template<typename Scalar> struct scalar_sqrt_op;
+template<typename Scalar> struct scalar_rsqrt_op;
 template<typename Scalar> struct scalar_exp_op;
 template<typename Scalar> struct scalar_log_op;
 template<typename Scalar> struct scalar_cos_op;
@@ -167,24 +195,29 @@
 template<typename Scalar> struct scalar_acos_op;
 template<typename Scalar> struct scalar_asin_op;
 template<typename Scalar> struct scalar_tan_op;
-template<typename Scalar> struct scalar_pow_op;
 template<typename Scalar> struct scalar_inverse_op;
 template<typename Scalar> struct scalar_square_op;
 template<typename Scalar> struct scalar_cube_op;
 template<typename Scalar, typename NewType> struct scalar_cast_op;
-template<typename Scalar> struct scalar_multiple_op;
-template<typename Scalar> struct scalar_quotient1_op;
-template<typename Scalar> struct scalar_min_op;
-template<typename Scalar> struct scalar_max_op;
 template<typename Scalar> struct scalar_random_op;
-template<typename Scalar> struct scalar_add_op;
 template<typename Scalar> struct scalar_constant_op;
 template<typename Scalar> struct scalar_identity_op;
-
+template<typename Scalar,bool iscpx> struct scalar_sign_op;
+template<typename Scalar,typename ScalarExponent> struct scalar_pow_op;
+template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_hypot_op;
 template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_product_op;
-template<typename LhsScalar,typename RhsScalar> struct scalar_multiple2_op;
 template<typename LhsScalar,typename RhsScalar=LhsScalar> struct scalar_quotient_op;
 
+// SpecialFunctions module
+template<typename Scalar> struct scalar_lgamma_op;
+template<typename Scalar> struct scalar_digamma_op;
+template<typename Scalar> struct scalar_erf_op;
+template<typename Scalar> struct scalar_erfc_op;
+template<typename Scalar> struct scalar_igamma_op;
+template<typename Scalar> struct scalar_igammac_op;
+template<typename Scalar> struct scalar_zeta_op;
+template<typename Scalar> struct scalar_betainc_op;
+
 } // end namespace internal
 
 struct IOFormat;
@@ -192,18 +225,18 @@
 // Array module
 template<typename _Scalar, int _Rows, int _Cols,
          int _Options = AutoAlign |
-#if defined(__GNUC__) && __GNUC__==3 && __GNUC_MINOR__==4
+#if EIGEN_GNUC_AT(3,4)
     // workaround a bug in at least gcc 3.4.6
     // the innermost ?: ternary operator is misparsed. We write it slightly
     // differently and this makes gcc 3.4.6 happy, but it's ugly.
     // The error would only show up with EIGEN_DEFAULT_TO_ROW_MAJOR is defined
     // (when EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION is RowMajor)
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
                           : !(_Cols==1 && _Rows!=1) ?  EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION
-                          : ColMajor ),
+                          : Eigen::ColMajor ),
 #else
-                          ( (_Rows==1 && _Cols!=1) ? RowMajor
-                          : (_Cols==1 && _Rows!=1) ? ColMajor
+                          ( (_Rows==1 && _Cols!=1) ? Eigen::RowMajor
+                          : (_Cols==1 && _Rows!=1) ? Eigen::ColMajor
                           : EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ),
 #endif
          int _MaxRows = _Rows, int _MaxCols = _Cols> class Array;
@@ -221,7 +254,9 @@
 template<typename MatrixType> class HouseholderQR;
 template<typename MatrixType> class ColPivHouseholderQR;
 template<typename MatrixType> class FullPivHouseholderQR;
+template<typename MatrixType> class CompleteOrthogonalDecomposition;
 template<typename MatrixType, int QRPreconditioner = ColPivHouseholderQRPreconditioner> class JacobiSVD;
+template<typename MatrixType> class BDCSVD;
 template<typename MatrixType, int UpLo = Lower> class LLT;
 template<typename MatrixType, int UpLo = Lower> class LDLT;
 template<typename VectorsType, typename CoeffsType, int Side=OnTheLeft> class HouseholderSequence;
@@ -234,36 +269,16 @@
 template<typename Scalar> class Rotation2D;
 template<typename Scalar> class AngleAxis;
 template<typename Scalar,int Dim> class Translation;
-
-#ifdef EIGEN2_SUPPORT
-template<typename Derived, int _Dim> class eigen2_RotationBase;
-template<typename Lhs, typename Rhs> class eigen2_Cross;
-template<typename Scalar> class eigen2_Quaternion;
-template<typename Scalar> class eigen2_Rotation2D;
-template<typename Scalar> class eigen2_AngleAxis;
-template<typename Scalar,int Dim> class eigen2_Transform;
-template <typename _Scalar, int _AmbientDim> class eigen2_ParametrizedLine;
-template <typename _Scalar, int _AmbientDim> class eigen2_Hyperplane;
-template<typename Scalar,int Dim> class eigen2_Translation;
-template<typename Scalar,int Dim> class eigen2_Scaling;
-#endif
-
-#if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS
-template<typename Scalar> class Quaternion;
-template<typename Scalar,int Dim> class Transform;
-template <typename _Scalar, int _AmbientDim> class ParametrizedLine;
-template <typename _Scalar, int _AmbientDim> class Hyperplane;
-template<typename Scalar,int Dim> class Scaling;
-#endif
-
-#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
+template<typename Scalar,int Dim> class AlignedBox;
 template<typename Scalar, int Options = AutoAlign> class Quaternion;
 template<typename Scalar,int Dim,int Mode,int _Options=AutoAlign> class Transform;
 template <typename _Scalar, int _AmbientDim, int Options=AutoAlign> class ParametrizedLine;
 template <typename _Scalar, int _AmbientDim, int Options=AutoAlign> class Hyperplane;
 template<typename Scalar> class UniformScaling;
 template<typename MatrixType,int Direction> class Homogeneous;
-#endif
+
+// Sparse module:
+template<typename Derived> class SparseMatrixBase;
 
 // MatrixFunctions module
 template<typename Derived> struct MatrixExponentialReturnValue;
@@ -271,7 +286,7 @@
 template<typename Derived> class MatrixSquareRootReturnValue;
 template<typename Derived> class MatrixLogarithmReturnValue;
 template<typename Derived> class MatrixPowerReturnValue;
-template<typename Derived, typename Lhs, typename Rhs> class MatrixPowerProduct;
+template<typename Derived> class MatrixComplexPowerReturnValue;
 
 namespace internal {
 template <typename Scalar>
@@ -282,18 +297,6 @@
 };
 }
 
-
-#ifdef EIGEN2_SUPPORT
-template<typename ExpressionType> class Cwise;
-template<typename MatrixType> class Minor;
-template<typename MatrixType> class LU;
-template<typename MatrixType> class QR;
-template<typename MatrixType> class SVD;
-namespace internal {
-template<typename MatrixType, unsigned int Mode> struct eigen2_part_return_type;
-}
-#endif
-
 } // end namespace Eigen
 
 #endif // EIGEN_FORWARDDECLARATIONS_H
diff --git a/Eigen/src/Core/util/MKL_support.h b/Eigen/src/Core/util/MKL_support.h
old mode 100644
new mode 100755
index 8acca9c..b7d6ecc
--- a/Eigen/src/Core/util/MKL_support.h
+++ b/Eigen/src/Core/util/MKL_support.h
@@ -49,10 +49,11 @@
   #define EIGEN_USE_LAPACKE
 #endif
 
-#if defined(EIGEN_USE_BLAS) || defined(EIGEN_USE_LAPACKE) || defined(EIGEN_USE_MKL_VML)
+#if defined(EIGEN_USE_MKL_VML) && !defined(EIGEN_USE_MKL)
   #define EIGEN_USE_MKL
 #endif
 
+
 #if defined EIGEN_USE_MKL
 #   include <mkl.h> 
 /*Check IMKL version for compatibility: < 10.3 is not usable with Eigen*/
@@ -64,7 +65,6 @@
 #   ifndef EIGEN_USE_MKL
     /*If the MKL version is too old, undef everything*/
 #       undef   EIGEN_USE_MKL_ALL
-#       undef   EIGEN_USE_BLAS
 #       undef   EIGEN_USE_LAPACKE
 #       undef   EIGEN_USE_MKL_VML
 #       undef   EIGEN_USE_LAPACKE_STRICT
@@ -73,54 +73,58 @@
 #endif
 
 #if defined EIGEN_USE_MKL
-#include <mkl_lapacke.h>
+
 #define EIGEN_MKL_VML_THRESHOLD 128
 
+/* MKL_DOMAIN_BLAS, etc are defined only in 10.3 update 7 */
+/* MKL_BLAS, etc are not defined in 11.2 */
+#ifdef MKL_DOMAIN_ALL
+#define EIGEN_MKL_DOMAIN_ALL MKL_DOMAIN_ALL
+#else
+#define EIGEN_MKL_DOMAIN_ALL MKL_ALL
+#endif
+
+#ifdef MKL_DOMAIN_BLAS
+#define EIGEN_MKL_DOMAIN_BLAS MKL_DOMAIN_BLAS
+#else
+#define EIGEN_MKL_DOMAIN_BLAS MKL_BLAS
+#endif
+
+#ifdef MKL_DOMAIN_FFT
+#define EIGEN_MKL_DOMAIN_FFT MKL_DOMAIN_FFT
+#else
+#define EIGEN_MKL_DOMAIN_FFT MKL_FFT
+#endif
+
+#ifdef MKL_DOMAIN_VML
+#define EIGEN_MKL_DOMAIN_VML MKL_DOMAIN_VML
+#else
+#define EIGEN_MKL_DOMAIN_VML MKL_VML
+#endif
+
+#ifdef MKL_DOMAIN_PARDISO
+#define EIGEN_MKL_DOMAIN_PARDISO MKL_DOMAIN_PARDISO
+#else
+#define EIGEN_MKL_DOMAIN_PARDISO MKL_PARDISO
+#endif
+#endif
+
+#if defined(EIGEN_USE_BLAS) && !defined(EIGEN_USE_MKL)
+#include "../../misc/blas.h"
+#endif
+
 namespace Eigen {
 
 typedef std::complex<double> dcomplex;
 typedef std::complex<float>  scomplex;
 
-namespace internal {
-
-template<typename MKLType, typename EigenType>
-static inline void assign_scalar_eig2mkl(MKLType& mklScalar, const EigenType& eigenScalar) {
-  mklScalar=eigenScalar;
-}
-
-template<typename MKLType, typename EigenType>
-static inline void assign_conj_scalar_eig2mkl(MKLType& mklScalar, const EigenType& eigenScalar) {
-  mklScalar=eigenScalar;
-}
-
-template <>
-inline void assign_scalar_eig2mkl<MKL_Complex16,dcomplex>(MKL_Complex16& mklScalar, const dcomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=eigenScalar.imag();
-}
-
-template <>
-inline void assign_scalar_eig2mkl<MKL_Complex8,scomplex>(MKL_Complex8& mklScalar, const scomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=eigenScalar.imag();
-}
-
-template <>
-inline void assign_conj_scalar_eig2mkl<MKL_Complex16,dcomplex>(MKL_Complex16& mklScalar, const dcomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=-eigenScalar.imag();
-}
-
-template <>
-inline void assign_conj_scalar_eig2mkl<MKL_Complex8,scomplex>(MKL_Complex8& mklScalar, const scomplex& eigenScalar) {
-  mklScalar.real=eigenScalar.real();
-  mklScalar.imag=-eigenScalar.imag();
-}
-
-} // end namespace internal
+#if defined(EIGEN_USE_MKL)
+typedef MKL_INT BlasIndex;
+#else
+typedef int BlasIndex;
+#endif
 
 } // end namespace Eigen
 
-#endif
 
 #endif // EIGEN_MKL_SUPPORT_H
diff --git a/Eigen/src/Core/util/Macros.h b/Eigen/src/Core/util/Macros.h
index f69970f..aa054a0 100644
--- a/Eigen/src/Core/util/Macros.h
+++ b/Eigen/src/Core/util/Macros.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -12,84 +12,324 @@
 #define EIGEN_MACROS_H
 
 #define EIGEN_WORLD_VERSION 3
-#define EIGEN_MAJOR_VERSION 2
-#define EIGEN_MINOR_VERSION 5
+#define EIGEN_MAJOR_VERSION 3
+#define EIGEN_MINOR_VERSION 7
 
 #define EIGEN_VERSION_AT_LEAST(x,y,z) (EIGEN_WORLD_VERSION>x || (EIGEN_WORLD_VERSION>=x && \
                                       (EIGEN_MAJOR_VERSION>y || (EIGEN_MAJOR_VERSION>=y && \
                                                                  EIGEN_MINOR_VERSION>=z))))
+
+// Compiler identification, EIGEN_COMP_*
+
+/// \internal EIGEN_COMP_GNUC set to 1 for all compilers compatible with GCC
 #ifdef __GNUC__
-  #define EIGEN_GNUC_AT_LEAST(x,y) ((__GNUC__==x && __GNUC_MINOR__>=y) || __GNUC__>x)
+  #define EIGEN_COMP_GNUC 1
 #else
-  #define EIGEN_GNUC_AT_LEAST(x,y) 0
-#endif
- 
-#ifdef __GNUC__
-  #define EIGEN_GNUC_AT_MOST(x,y) ((__GNUC__==x && __GNUC_MINOR__<=y) || __GNUC__<x)
-#else
-  #define EIGEN_GNUC_AT_MOST(x,y) 0
+  #define EIGEN_COMP_GNUC 0
 #endif
 
-#if EIGEN_GNUC_AT_MOST(4,3) && !defined(__clang__)
+/// \internal EIGEN_COMP_CLANG set to major+minor version (e.g., 307 for clang 3.7) if the compiler is clang
+#if defined(__clang__)
+  #define EIGEN_COMP_CLANG (__clang_major__*100+__clang_minor__)
+#else
+  #define EIGEN_COMP_CLANG 0
+#endif
+
+
+/// \internal EIGEN_COMP_LLVM set to 1 if the compiler backend is llvm
+#if defined(__llvm__)
+  #define EIGEN_COMP_LLVM 1
+#else
+  #define EIGEN_COMP_LLVM 0
+#endif
+
+/// \internal EIGEN_COMP_ICC set to __INTEL_COMPILER if the compiler is Intel compiler, 0 otherwise
+#if defined(__INTEL_COMPILER)
+  #define EIGEN_COMP_ICC __INTEL_COMPILER
+#else
+  #define EIGEN_COMP_ICC 0
+#endif
+
+/// \internal EIGEN_COMP_MINGW set to 1 if the compiler is mingw
+#if defined(__MINGW32__)
+  #define EIGEN_COMP_MINGW 1
+#else
+  #define EIGEN_COMP_MINGW 0
+#endif
+
+/// \internal EIGEN_COMP_SUNCC set to 1 if the compiler is Solaris Studio
+#if defined(__SUNPRO_CC)
+  #define EIGEN_COMP_SUNCC 1
+#else
+  #define EIGEN_COMP_SUNCC 0
+#endif
+
+/// \internal EIGEN_COMP_MSVC set to _MSC_VER if the compiler is Microsoft Visual C++, 0 otherwise.
+#if defined(_MSC_VER)
+  #define EIGEN_COMP_MSVC _MSC_VER
+#else
+  #define EIGEN_COMP_MSVC 0
+#endif
+
+// For the record, here is a table summarizing the possible values for EIGEN_COMP_MSVC:
+//  name  ver   MSC_VER
+//  2008    9      1500
+//  2010   10      1600
+//  2012   11      1700
+//  2013   12      1800
+//  2015   14      1900
+//  "15"   15      1900
+
+/// \internal EIGEN_COMP_MSVC_STRICT set to 1 if the compiler is really Microsoft Visual C++ and not ,e.g., ICC or clang-cl
+#if EIGEN_COMP_MSVC && !(EIGEN_COMP_ICC || EIGEN_COMP_LLVM || EIGEN_COMP_CLANG)
+  #define EIGEN_COMP_MSVC_STRICT _MSC_VER
+#else
+  #define EIGEN_COMP_MSVC_STRICT 0
+#endif
+
+/// \internal EIGEN_COMP_IBM set to 1 if the compiler is IBM XL C++
+#if defined(__IBMCPP__) || defined(__xlc__)
+  #define EIGEN_COMP_IBM 1
+#else
+  #define EIGEN_COMP_IBM 0
+#endif
+
+/// \internal EIGEN_COMP_PGI set to 1 if the compiler is Portland Group Compiler
+#if defined(__PGI)
+  #define EIGEN_COMP_PGI 1
+#else
+  #define EIGEN_COMP_PGI 0
+#endif
+
+/// \internal EIGEN_COMP_ARM set to 1 if the compiler is ARM Compiler
+#if defined(__CC_ARM) || defined(__ARMCC_VERSION)
+  #define EIGEN_COMP_ARM 1
+#else
+  #define EIGEN_COMP_ARM 0
+#endif
+
+/// \internal EIGEN_COMP_ARM set to 1 if the compiler is ARM Compiler
+#if defined(__EMSCRIPTEN__)
+  #define EIGEN_COMP_EMSCRIPTEN 1
+#else
+  #define EIGEN_COMP_EMSCRIPTEN 0
+#endif
+
+
+/// \internal EIGEN_GNUC_STRICT set to 1 if the compiler is really GCC and not a compatible compiler (e.g., ICC, clang, mingw, etc.)
+#if EIGEN_COMP_GNUC && !(EIGEN_COMP_CLANG || EIGEN_COMP_ICC || EIGEN_COMP_MINGW || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM || EIGEN_COMP_EMSCRIPTEN)
+  #define EIGEN_COMP_GNUC_STRICT 1
+#else
+  #define EIGEN_COMP_GNUC_STRICT 0
+#endif
+
+
+#if EIGEN_COMP_GNUC
+  #define EIGEN_GNUC_AT_LEAST(x,y) ((__GNUC__==x && __GNUC_MINOR__>=y) || __GNUC__>x)
+  #define EIGEN_GNUC_AT_MOST(x,y)  ((__GNUC__==x && __GNUC_MINOR__<=y) || __GNUC__<x)
+  #define EIGEN_GNUC_AT(x,y)       ( __GNUC__==x && __GNUC_MINOR__==y )
+#else
+  #define EIGEN_GNUC_AT_LEAST(x,y) 0
+  #define EIGEN_GNUC_AT_MOST(x,y)  0
+  #define EIGEN_GNUC_AT(x,y)       0
+#endif
+
+// FIXME: could probably be removed as we do not support gcc 3.x anymore
+#if EIGEN_COMP_GNUC && (__GNUC__ <= 3)
+#define EIGEN_GCC3_OR_OLDER 1
+#else
+#define EIGEN_GCC3_OR_OLDER 0
+#endif
+
+
+// Architecture identification, EIGEN_ARCH_*
+
+#if defined(__x86_64__) || defined(_M_X64) || defined(__amd64)
+  #define EIGEN_ARCH_x86_64 1
+#else
+  #define EIGEN_ARCH_x86_64 0
+#endif
+
+#if defined(__i386__) || defined(_M_IX86) || defined(_X86_) || defined(__i386)
+  #define EIGEN_ARCH_i386 1
+#else
+  #define EIGEN_ARCH_i386 0
+#endif
+
+#if EIGEN_ARCH_x86_64 || EIGEN_ARCH_i386
+  #define EIGEN_ARCH_i386_OR_x86_64 1
+#else
+  #define EIGEN_ARCH_i386_OR_x86_64 0
+#endif
+
+/// \internal EIGEN_ARCH_ARM set to 1 if the architecture is ARM
+#if defined(__arm__)
+  #define EIGEN_ARCH_ARM 1
+#else
+  #define EIGEN_ARCH_ARM 0
+#endif
+
+/// \internal EIGEN_ARCH_ARM64 set to 1 if the architecture is ARM64
+#if defined(__aarch64__)
+  #define EIGEN_ARCH_ARM64 1
+#else
+  #define EIGEN_ARCH_ARM64 0
+#endif
+
+#if EIGEN_ARCH_ARM || EIGEN_ARCH_ARM64
+  #define EIGEN_ARCH_ARM_OR_ARM64 1
+#else
+  #define EIGEN_ARCH_ARM_OR_ARM64 0
+#endif
+
+/// \internal EIGEN_ARCH_MIPS set to 1 if the architecture is MIPS
+#if defined(__mips__) || defined(__mips)
+  #define EIGEN_ARCH_MIPS 1
+#else
+  #define EIGEN_ARCH_MIPS 0
+#endif
+
+/// \internal EIGEN_ARCH_SPARC set to 1 if the architecture is SPARC
+#if defined(__sparc__) || defined(__sparc)
+  #define EIGEN_ARCH_SPARC 1
+#else
+  #define EIGEN_ARCH_SPARC 0
+#endif
+
+/// \internal EIGEN_ARCH_IA64 set to 1 if the architecture is Intel Itanium
+#if defined(__ia64__)
+  #define EIGEN_ARCH_IA64 1
+#else
+  #define EIGEN_ARCH_IA64 0
+#endif
+
+/// \internal EIGEN_ARCH_PPC set to 1 if the architecture is PowerPC
+#if defined(__powerpc__) || defined(__ppc__) || defined(_M_PPC)
+  #define EIGEN_ARCH_PPC 1
+#else
+  #define EIGEN_ARCH_PPC 0
+#endif
+
+
+
+// Operating system identification, EIGEN_OS_*
+
+/// \internal EIGEN_OS_UNIX set to 1 if the OS is a unix variant
+#if defined(__unix__) || defined(__unix)
+  #define EIGEN_OS_UNIX 1
+#else
+  #define EIGEN_OS_UNIX 0
+#endif
+
+/// \internal EIGEN_OS_LINUX set to 1 if the OS is based on Linux kernel
+#if defined(__linux__)
+  #define EIGEN_OS_LINUX 1
+#else
+  #define EIGEN_OS_LINUX 0
+#endif
+
+/// \internal EIGEN_OS_ANDROID set to 1 if the OS is Android
+// note: ANDROID is defined when using ndk_build, __ANDROID__ is defined when using a standalone toolchain.
+#if defined(__ANDROID__) || defined(ANDROID)
+  #define EIGEN_OS_ANDROID 1
+#else
+  #define EIGEN_OS_ANDROID 0
+#endif
+
+/// \internal EIGEN_OS_GNULINUX set to 1 if the OS is GNU Linux and not Linux-based OS (e.g., not android)
+#if defined(__gnu_linux__) && !(EIGEN_OS_ANDROID)
+  #define EIGEN_OS_GNULINUX 1
+#else
+  #define EIGEN_OS_GNULINUX 0
+#endif
+
+/// \internal EIGEN_OS_BSD set to 1 if the OS is a BSD variant
+#if defined(__FreeBSD__) || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__bsdi__) || defined(__DragonFly__)
+  #define EIGEN_OS_BSD 1
+#else
+  #define EIGEN_OS_BSD 0
+#endif
+
+/// \internal EIGEN_OS_MAC set to 1 if the OS is MacOS
+#if defined(__APPLE__)
+  #define EIGEN_OS_MAC 1
+#else
+  #define EIGEN_OS_MAC 0
+#endif
+
+/// \internal EIGEN_OS_QNX set to 1 if the OS is QNX
+#if defined(__QNX__)
+  #define EIGEN_OS_QNX 1
+#else
+  #define EIGEN_OS_QNX 0
+#endif
+
+/// \internal EIGEN_OS_WIN set to 1 if the OS is Windows based
+#if defined(_WIN32)
+  #define EIGEN_OS_WIN 1
+#else
+  #define EIGEN_OS_WIN 0
+#endif
+
+/// \internal EIGEN_OS_WIN64 set to 1 if the OS is Windows 64bits
+#if defined(_WIN64)
+  #define EIGEN_OS_WIN64 1
+#else
+  #define EIGEN_OS_WIN64 0
+#endif
+
+/// \internal EIGEN_OS_WINCE set to 1 if the OS is Windows CE
+#if defined(_WIN32_WCE)
+  #define EIGEN_OS_WINCE 1
+#else
+  #define EIGEN_OS_WINCE 0
+#endif
+
+/// \internal EIGEN_OS_CYGWIN set to 1 if the OS is Windows/Cygwin
+#if defined(__CYGWIN__)
+  #define EIGEN_OS_CYGWIN 1
+#else
+  #define EIGEN_OS_CYGWIN 0
+#endif
+
+/// \internal EIGEN_OS_WIN_STRICT set to 1 if the OS is really Windows and not some variants
+#if EIGEN_OS_WIN && !( EIGEN_OS_WINCE || EIGEN_OS_CYGWIN )
+  #define EIGEN_OS_WIN_STRICT 1
+#else
+  #define EIGEN_OS_WIN_STRICT 0
+#endif
+
+/// \internal EIGEN_OS_SUN set to 1 if the OS is SUN
+#if (defined(sun) || defined(__sun)) && !(defined(__SVR4) || defined(__svr4__))
+  #define EIGEN_OS_SUN 1
+#else
+  #define EIGEN_OS_SUN 0
+#endif
+
+/// \internal EIGEN_OS_SOLARIS set to 1 if the OS is Solaris
+#if (defined(sun) || defined(__sun)) && (defined(__SVR4) || defined(__svr4__))
+  #define EIGEN_OS_SOLARIS 1
+#else
+  #define EIGEN_OS_SOLARIS 0
+#endif
+
+
+
+#if EIGEN_GNUC_AT_MOST(4,3) && !EIGEN_COMP_CLANG
   // see bug 89
   #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 0
 #else
   #define EIGEN_SAFE_TO_USE_STANDARD_ASSERT_MACRO 1
 #endif
 
-#if defined(__GNUC__) && (__GNUC__ <= 3)
-#define EIGEN_GCC3_OR_OLDER 1
-#else
-#define EIGEN_GCC3_OR_OLDER 0
-#endif
-
-// 16 byte alignment is only useful for vectorization. Since it affects the ABI, we need to enable
-// 16 byte alignment on all platforms where vectorization might be enabled. In theory we could always
-// enable alignment, but it can be a cause of problems on some platforms, so we just disable it in
-// certain common platform (compiler+architecture combinations) to avoid these problems.
-// Only static alignment is really problematic (relies on nonstandard compiler extensions that don't
-// work everywhere, for example don't work on GCC/ARM), try to keep heap alignment even
-// when we have to disable static alignment.
-#if defined(__GNUC__) && !(defined(__i386__) || defined(__x86_64__) || defined(__powerpc__) || defined(__ppc__) || defined(__ia64__))
-#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1
-#else
-#define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 0
-#endif
-
-// static alignment is completely disabled with GCC 3, Sun Studio, and QCC/QNX
-#if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT \
- && !EIGEN_GCC3_OR_OLDER \
- && !defined(__SUNPRO_CC) \
- && !defined(__QNXNTO__)
-  #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 1
-#else
-  #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 0
-#endif
-
-#ifdef EIGEN_DONT_ALIGN
-  #ifndef EIGEN_DONT_ALIGN_STATICALLY
-    #define EIGEN_DONT_ALIGN_STATICALLY
-  #endif
-  #define EIGEN_ALIGN 0
-#else
-  #define EIGEN_ALIGN 1
-#endif
-
-// EIGEN_ALIGN_STATICALLY is the true test whether we want to align arrays on the stack or not. It takes into account both the user choice to explicitly disable
-// alignment (EIGEN_DONT_ALIGN_STATICALLY) and the architecture config (EIGEN_ARCH_WANTS_STACK_ALIGNMENT). Henceforth, only EIGEN_ALIGN_STATICALLY should be used.
-#if EIGEN_ARCH_WANTS_STACK_ALIGNMENT && !defined(EIGEN_DONT_ALIGN_STATICALLY)
-  #define EIGEN_ALIGN_STATICALLY 1
-#else
-  #define EIGEN_ALIGN_STATICALLY 0
-  #ifndef EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
-    #define EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
-  #endif
-#endif
+// This macro can be used to prevent from macro expansion, e.g.:
+//   std::max EIGEN_NOT_A_MACRO(a,b)
+#define EIGEN_NOT_A_MACRO
 
 #ifdef EIGEN_DEFAULT_TO_ROW_MAJOR
-#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION RowMajor
+#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Eigen::RowMajor
 #else
-#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION ColMajor
+#define EIGEN_DEFAULT_MATRIX_STORAGE_ORDER_OPTION Eigen::ColMajor
 #endif
 
 #ifndef EIGEN_DEFAULT_DENSE_INDEX_TYPE
@@ -103,10 +343,130 @@
 #  define EIGEN_HAS_BUILTIN(x) 0
 #endif
 
+// A Clang feature extension to determine compiler features.
+// We use it to determine 'cxx_rvalue_references'
+#ifndef __has_feature
+# define __has_feature(x) 0
+#endif
+
+// Upperbound on the C++ version to use.
+// Expected values are 03, 11, 14, 17, etc.
+// By default, let's use an arbitrarily large C++ version.
+#ifndef EIGEN_MAX_CPP_VER
+#define EIGEN_MAX_CPP_VER 99
+#endif
+
+#if EIGEN_MAX_CPP_VER>=11 && (defined(__cplusplus) && (__cplusplus >= 201103L) || EIGEN_COMP_MSVC >= 1900)
+#define EIGEN_HAS_CXX11 1
+#else
+#define EIGEN_HAS_CXX11 0
+#endif
+
+
+// Do we support r-value references?
+#ifndef EIGEN_HAS_RVALUE_REFERENCES
+#if EIGEN_MAX_CPP_VER>=11 && \
+    (__has_feature(cxx_rvalue_references) || \
+    (defined(__cplusplus) && __cplusplus >= 201103L) || \
+    (EIGEN_COMP_MSVC >= 1600))
+  #define EIGEN_HAS_RVALUE_REFERENCES 1
+#else
+  #define EIGEN_HAS_RVALUE_REFERENCES 0
+#endif
+#endif
+
+// Does the compiler support C99?
+#ifndef EIGEN_HAS_C99_MATH
+#if EIGEN_MAX_CPP_VER>=11 && \
+    ((defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 199901))       \
+  || (defined(__GNUC__) && defined(_GLIBCXX_USE_C99)) \
+  || (defined(_LIBCPP_VERSION) && !defined(_MSC_VER)))
+  #define EIGEN_HAS_C99_MATH 1
+#else
+  #define EIGEN_HAS_C99_MATH 0
+#endif
+#endif
+
+// Does the compiler support result_of?
+#ifndef EIGEN_HAS_STD_RESULT_OF
+#if EIGEN_MAX_CPP_VER>=11 && ((__has_feature(cxx_lambdas) || (defined(__cplusplus) && __cplusplus >= 201103L)))
+#define EIGEN_HAS_STD_RESULT_OF 1
+#else
+#define EIGEN_HAS_STD_RESULT_OF 0
+#endif
+#endif
+
+// Does the compiler support variadic templates?
+#ifndef EIGEN_HAS_VARIADIC_TEMPLATES
+#if EIGEN_MAX_CPP_VER>=11 && (__cplusplus > 199711L || EIGEN_COMP_MSVC >= 1900) \
+  && (!defined(__NVCC__) || !EIGEN_ARCH_ARM_OR_ARM64 || (EIGEN_CUDACC_VER >= 80000) )
+    // ^^ Disable the use of variadic templates when compiling with versions of nvcc older than 8.0 on ARM devices:
+    //    this prevents nvcc from crashing when compiling Eigen on Tegra X1
+#define EIGEN_HAS_VARIADIC_TEMPLATES 1
+#else
+#define EIGEN_HAS_VARIADIC_TEMPLATES 0
+#endif
+#endif
+
+// Does the compiler fully support const expressions? (as in c++14)
+#ifndef EIGEN_HAS_CONSTEXPR
+
+#ifdef __CUDACC__
+// Const expressions are supported provided that c++11 is enabled and we're using either clang or nvcc 7.5 or above
+#if EIGEN_MAX_CPP_VER>=14 && (__cplusplus > 199711L && (EIGEN_COMP_CLANG || EIGEN_CUDACC_VER >= 70500))
+  #define EIGEN_HAS_CONSTEXPR 1
+#endif
+#elif EIGEN_MAX_CPP_VER>=14 && (__has_feature(cxx_relaxed_constexpr) || (defined(__cplusplus) && __cplusplus >= 201402L) || \
+  (EIGEN_GNUC_AT_LEAST(4,8) && (__cplusplus > 199711L)))
+#define EIGEN_HAS_CONSTEXPR 1
+#endif
+
+#ifndef EIGEN_HAS_CONSTEXPR
+#define EIGEN_HAS_CONSTEXPR 0
+#endif
+
+#endif
+
+// Does the compiler support C++11 math?
+// Let's be conservative and enable the default C++11 implementation only if we are sure it exists
+#ifndef EIGEN_HAS_CXX11_MATH
+  #if EIGEN_MAX_CPP_VER>=11 && ((__cplusplus > 201103L) || (__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_MSVC || EIGEN_COMP_ICC)  \
+      && (EIGEN_ARCH_i386_OR_x86_64) && (EIGEN_OS_GNULINUX || EIGEN_OS_WIN_STRICT || EIGEN_OS_MAC))
+    #define EIGEN_HAS_CXX11_MATH 1
+  #else
+    #define EIGEN_HAS_CXX11_MATH 0
+  #endif
+#endif
+
+// Does the compiler support proper C++11 containers?
+#ifndef EIGEN_HAS_CXX11_CONTAINERS
+  #if    EIGEN_MAX_CPP_VER>=11 && \
+         ((__cplusplus > 201103L) \
+      || ((__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_ICC>=1400)) \
+      || EIGEN_COMP_MSVC >= 1900)
+    #define EIGEN_HAS_CXX11_CONTAINERS 1
+  #else
+    #define EIGEN_HAS_CXX11_CONTAINERS 0
+  #endif
+#endif
+
+// Does the compiler support C++11 noexcept?
+#ifndef EIGEN_HAS_CXX11_NOEXCEPT
+  #if    EIGEN_MAX_CPP_VER>=11 && \
+         (__has_feature(cxx_noexcept) \
+      || (__cplusplus > 201103L) \
+      || ((__cplusplus >= 201103L) && (EIGEN_COMP_GNUC_STRICT || EIGEN_COMP_CLANG || EIGEN_COMP_ICC>=1400)) \
+      || EIGEN_COMP_MSVC >= 1900)
+    #define EIGEN_HAS_CXX11_NOEXCEPT 1
+  #else
+    #define EIGEN_HAS_CXX11_NOEXCEPT 0
+  #endif
+#endif
+
 /** Allows to disable some optimizations which might affect the accuracy of the result.
   * Such optimization are enabled by default, and set EIGEN_FAST_MATH to 0 to disable them.
   * They currently include:
-  *   - single precision Cwise::sin() and Cwise::cos() when SSE vectorization is enabled.
+  *   - single precision ArrayBase::sin() and ArrayBase::cos() for SSE and AVX vectorization.
   */
 #ifndef EIGEN_FAST_MATH
 #define EIGEN_FAST_MATH 1
@@ -118,6 +478,8 @@
 #define EIGEN_CAT2(a,b) a ## b
 #define EIGEN_CAT(a,b) EIGEN_CAT2(a,b)
 
+#define EIGEN_COMMA ,
+
 // convert a token to a string
 #define EIGEN_MAKESTRING2(a) #a
 #define EIGEN_MAKESTRING(a) EIGEN_MAKESTRING2(a)
@@ -125,34 +487,37 @@
 // EIGEN_STRONG_INLINE is a stronger version of the inline, using __forceinline on MSVC,
 // but it still doesn't use GCC's always_inline. This is useful in (common) situations where MSVC needs forceinline
 // but GCC is still doing fine with just inline.
-#if (defined _MSC_VER) || (defined __INTEL_COMPILER)
+#ifndef EIGEN_STRONG_INLINE
+#if EIGEN_COMP_MSVC || EIGEN_COMP_ICC
 #define EIGEN_STRONG_INLINE __forceinline
 #else
 #define EIGEN_STRONG_INLINE inline
 #endif
+#endif
 
 // EIGEN_ALWAYS_INLINE is the stronget, it has the effect of making the function inline and adding every possible
 // attribute to maximize inlining. This should only be used when really necessary: in particular,
 // it uses __attribute__((always_inline)) on GCC, which most of the time is useless and can severely harm compile times.
 // FIXME with the always_inline attribute,
-// gcc 3.4.x reports the following compilation error:
+// gcc 3.4.x and 4.1 reports the following compilation error:
 //   Eval.h:91: sorry, unimplemented: inlining failed in call to 'const Eigen::Eval<Derived> Eigen::MatrixBase<Scalar, Derived>::eval() const'
 //    : function body not available
-#if EIGEN_GNUC_AT_LEAST(4,0)
+//   See also bug 1367
+#if EIGEN_GNUC_AT_LEAST(4,2)
 #define EIGEN_ALWAYS_INLINE __attribute__((always_inline)) inline
 #else
 #define EIGEN_ALWAYS_INLINE EIGEN_STRONG_INLINE
 #endif
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 #define EIGEN_DONT_INLINE __attribute__((noinline))
-#elif (defined _MSC_VER)
+#elif EIGEN_COMP_MSVC
 #define EIGEN_DONT_INLINE __declspec(noinline)
 #else
 #define EIGEN_DONT_INLINE
 #endif
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 #define EIGEN_PERMISSIVE_EXPR __extension__
 #else
 #define EIGEN_PERMISSIVE_EXPR
@@ -221,15 +586,15 @@
 #endif
 
 #ifdef EIGEN_NO_DEBUG
-#define EIGEN_ONLY_USED_FOR_DEBUG(x) (void)x
+#define EIGEN_ONLY_USED_FOR_DEBUG(x) EIGEN_UNUSED_VARIABLE(x)
 #else
 #define EIGEN_ONLY_USED_FOR_DEBUG(x)
 #endif
 
 #ifndef EIGEN_NO_DEPRECATED_WARNING
-  #if (defined __GNUC__)
+  #if EIGEN_COMP_GNUC
     #define EIGEN_DEPRECATED __attribute__((deprecated))
-  #elif (defined _MSC_VER)
+  #elif EIGEN_COMP_MSVC
     #define EIGEN_DEPRECATED __declspec(deprecated)
   #else
     #define EIGEN_DEPRECATED
@@ -238,7 +603,7 @@
   #define EIGEN_DEPRECATED
 #endif
 
-#if (defined __GNUC__)
+#if EIGEN_COMP_GNUC
 #define EIGEN_UNUSED __attribute__((unused))
 #else
 #define EIGEN_UNUSED
@@ -247,19 +612,33 @@
 // Suppresses 'unused variable' warnings.
 namespace Eigen {
   namespace internal {
-    template<typename T> void ignore_unused_variable(const T&) {}
+    template<typename T> EIGEN_DEVICE_FUNC void ignore_unused_variable(const T&) {}
   }
 }
 #define EIGEN_UNUSED_VARIABLE(var) Eigen::internal::ignore_unused_variable(var);
 
 #if !defined(EIGEN_ASM_COMMENT)
-  #if (defined __GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
+  #if EIGEN_COMP_GNUC && (EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM_OR_ARM64)
     #define EIGEN_ASM_COMMENT(X)  __asm__("#" X)
   #else
     #define EIGEN_ASM_COMMENT(X)
   #endif
 #endif
 
+
+//------------------------------------------------------------------------------------------
+// Static and dynamic alignment control
+//
+// The main purpose of this section is to define EIGEN_MAX_ALIGN_BYTES and EIGEN_MAX_STATIC_ALIGN_BYTES
+// as the maximal boundary in bytes on which dynamically and statically allocated data may be alignment respectively.
+// The values of EIGEN_MAX_ALIGN_BYTES and EIGEN_MAX_STATIC_ALIGN_BYTES can be specified by the user. If not,
+// a default value is automatically computed based on architecture, compiler, and OS.
+//
+// This section also defines macros EIGEN_ALIGN_TO_BOUNDARY(N) and the shortcuts EIGEN_ALIGN{8,16,32,_MAX}
+// to be used to declare statically aligned buffers.
+//------------------------------------------------------------------------------------------
+
+
 /* EIGEN_ALIGN_TO_BOUNDARY(n) forces data to be n-byte aligned. This is used to satisfy SIMD requirements.
  * However, we do that EVEN if vectorization (EIGEN_VECTORIZE) is disabled,
  * so that vectorization doesn't affect binary compatibility.
@@ -267,28 +646,149 @@
  * If we made alignment depend on whether or not EIGEN_VECTORIZE is defined, it would be impossible to link
  * vectorized and non-vectorized code.
  */
-#if (defined __GNUC__) || (defined __PGI) || (defined __IBMCPP__) || (defined __ARMCC_VERSION)
+#if (defined __CUDACC__)
+  #define EIGEN_ALIGN_TO_BOUNDARY(n) __align__(n)
+#elif EIGEN_COMP_GNUC || EIGEN_COMP_PGI || EIGEN_COMP_IBM || EIGEN_COMP_ARM
   #define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
-#elif (defined _MSC_VER)
+#elif EIGEN_COMP_MSVC
   #define EIGEN_ALIGN_TO_BOUNDARY(n) __declspec(align(n))
-#elif (defined __SUNPRO_CC)
+#elif EIGEN_COMP_SUNCC
   // FIXME not sure about this one:
   #define EIGEN_ALIGN_TO_BOUNDARY(n) __attribute__((aligned(n)))
 #else
   #error Please tell me what is the equivalent of __attribute__((aligned(n))) for your compiler
 #endif
 
+// If the user explicitly disable vectorization, then we also disable alignment
+#if defined(EIGEN_DONT_VECTORIZE)
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 0
+#elif defined(EIGEN_VECTORIZE_AVX512)
+  // 64 bytes static alignmeent is preferred only if really required
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 64
+#elif defined(__AVX__)
+  // 32 bytes static alignmeent is preferred only if really required
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 32
+#else
+  #define EIGEN_IDEAL_MAX_ALIGN_BYTES 16
+#endif
+
+
+// EIGEN_MIN_ALIGN_BYTES defines the minimal value for which the notion of explicit alignment makes sense
+#define EIGEN_MIN_ALIGN_BYTES 16
+
+// Defined the boundary (in bytes) on which the data needs to be aligned. Note
+// that unless EIGEN_ALIGN is defined and not equal to 0, the data may not be
+// aligned at all regardless of the value of this #define.
+
+#if (defined(EIGEN_DONT_ALIGN_STATICALLY) || defined(EIGEN_DONT_ALIGN))  && defined(EIGEN_MAX_STATIC_ALIGN_BYTES) && EIGEN_MAX_STATIC_ALIGN_BYTES>0
+#error EIGEN_MAX_STATIC_ALIGN_BYTES and EIGEN_DONT_ALIGN[_STATICALLY] are both defined with EIGEN_MAX_STATIC_ALIGN_BYTES!=0. Use EIGEN_MAX_STATIC_ALIGN_BYTES=0 as a synonym of EIGEN_DONT_ALIGN_STATICALLY.
+#endif
+
+// EIGEN_DONT_ALIGN_STATICALLY and EIGEN_DONT_ALIGN are deprectated
+// They imply EIGEN_MAX_STATIC_ALIGN_BYTES=0
+#if defined(EIGEN_DONT_ALIGN_STATICALLY) || defined(EIGEN_DONT_ALIGN)
+  #ifdef EIGEN_MAX_STATIC_ALIGN_BYTES
+    #undef EIGEN_MAX_STATIC_ALIGN_BYTES
+  #endif
+  #define EIGEN_MAX_STATIC_ALIGN_BYTES 0
+#endif
+
+#ifndef EIGEN_MAX_STATIC_ALIGN_BYTES
+
+  // Try to automatically guess what is the best default value for EIGEN_MAX_STATIC_ALIGN_BYTES
+
+  // 16 byte alignment is only useful for vectorization. Since it affects the ABI, we need to enable
+  // 16 byte alignment on all platforms where vectorization might be enabled. In theory we could always
+  // enable alignment, but it can be a cause of problems on some platforms, so we just disable it in
+  // certain common platform (compiler+architecture combinations) to avoid these problems.
+  // Only static alignment is really problematic (relies on nonstandard compiler extensions),
+  // try to keep heap alignment even when we have to disable static alignment.
+  #if EIGEN_COMP_GNUC && !(EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM_OR_ARM64 || EIGEN_ARCH_PPC || EIGEN_ARCH_IA64)
+  #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1
+  #elif EIGEN_ARCH_ARM_OR_ARM64 && EIGEN_COMP_GNUC_STRICT && EIGEN_GNUC_AT_MOST(4, 6)
+  // Old versions of GCC on ARM, at least 4.4, were once seen to have buggy static alignment support.
+  // Not sure which version fixed it, hopefully it doesn't affect 4.7, which is still somewhat in use.
+  // 4.8 and newer seem definitely unaffected.
+  #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 1
+  #else
+  #define EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT 0
+  #endif
+
+  // static alignment is completely disabled with GCC 3, Sun Studio, and QCC/QNX
+  #if !EIGEN_GCC_AND_ARCH_DOESNT_WANT_STACK_ALIGNMENT \
+  && !EIGEN_GCC3_OR_OLDER \
+  && !EIGEN_COMP_SUNCC \
+  && !EIGEN_OS_QNX
+    #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 1
+  #else
+    #define EIGEN_ARCH_WANTS_STACK_ALIGNMENT 0
+  #endif
+
+  #if EIGEN_ARCH_WANTS_STACK_ALIGNMENT
+    #define EIGEN_MAX_STATIC_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES
+  #else
+    #define EIGEN_MAX_STATIC_ALIGN_BYTES 0
+  #endif
+
+#endif
+
+// If EIGEN_MAX_ALIGN_BYTES is defined, then it is considered as an upper bound for EIGEN_MAX_ALIGN_BYTES
+#if defined(EIGEN_MAX_ALIGN_BYTES) && EIGEN_MAX_ALIGN_BYTES<EIGEN_MAX_STATIC_ALIGN_BYTES
+#undef EIGEN_MAX_STATIC_ALIGN_BYTES
+#define EIGEN_MAX_STATIC_ALIGN_BYTES EIGEN_MAX_ALIGN_BYTES
+#endif
+
+#if EIGEN_MAX_STATIC_ALIGN_BYTES==0 && !defined(EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT)
+  #define EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT
+#endif
+
+// At this stage, EIGEN_MAX_STATIC_ALIGN_BYTES>0 is the true test whether we want to align arrays on the stack or not.
+// It takes into account both the user choice to explicitly enable/disable alignment (by settting EIGEN_MAX_STATIC_ALIGN_BYTES)
+// and the architecture config (EIGEN_ARCH_WANTS_STACK_ALIGNMENT).
+// Henceforth, only EIGEN_MAX_STATIC_ALIGN_BYTES should be used.
+
+
+// Shortcuts to EIGEN_ALIGN_TO_BOUNDARY
 #define EIGEN_ALIGN8  EIGEN_ALIGN_TO_BOUNDARY(8)
 #define EIGEN_ALIGN16 EIGEN_ALIGN_TO_BOUNDARY(16)
-
-#if EIGEN_ALIGN_STATICALLY
-#define EIGEN_USER_ALIGN_TO_BOUNDARY(n) EIGEN_ALIGN_TO_BOUNDARY(n)
-#define EIGEN_USER_ALIGN16 EIGEN_ALIGN16
+#define EIGEN_ALIGN32 EIGEN_ALIGN_TO_BOUNDARY(32)
+#define EIGEN_ALIGN64 EIGEN_ALIGN_TO_BOUNDARY(64)
+#if EIGEN_MAX_STATIC_ALIGN_BYTES>0
+#define EIGEN_ALIGN_MAX EIGEN_ALIGN_TO_BOUNDARY(EIGEN_MAX_STATIC_ALIGN_BYTES)
 #else
-#define EIGEN_USER_ALIGN_TO_BOUNDARY(n)
-#define EIGEN_USER_ALIGN16
+#define EIGEN_ALIGN_MAX
 #endif
 
+
+// Dynamic alignment control
+
+#if defined(EIGEN_DONT_ALIGN) && defined(EIGEN_MAX_ALIGN_BYTES) && EIGEN_MAX_ALIGN_BYTES>0
+#error EIGEN_MAX_ALIGN_BYTES and EIGEN_DONT_ALIGN are both defined with EIGEN_MAX_ALIGN_BYTES!=0. Use EIGEN_MAX_ALIGN_BYTES=0 as a synonym of EIGEN_DONT_ALIGN.
+#endif
+
+#ifdef EIGEN_DONT_ALIGN
+  #ifdef EIGEN_MAX_ALIGN_BYTES
+    #undef EIGEN_MAX_ALIGN_BYTES
+  #endif
+  #define EIGEN_MAX_ALIGN_BYTES 0
+#elif !defined(EIGEN_MAX_ALIGN_BYTES)
+  #define EIGEN_MAX_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES
+#endif
+
+#if EIGEN_IDEAL_MAX_ALIGN_BYTES > EIGEN_MAX_ALIGN_BYTES
+#define EIGEN_DEFAULT_ALIGN_BYTES EIGEN_IDEAL_MAX_ALIGN_BYTES
+#else
+#define EIGEN_DEFAULT_ALIGN_BYTES EIGEN_MAX_ALIGN_BYTES
+#endif
+
+
+#ifndef EIGEN_UNALIGNED_VECTORIZE
+#define EIGEN_UNALIGNED_VECTORIZE 1
+#endif
+
+//----------------------------------------------------------------------
+
+
 #ifdef EIGEN_DONT_USE_RESTRICT_KEYWORD
   #define EIGEN_RESTRICT
 #endif
@@ -314,25 +814,27 @@
 // just an empty macro !
 #define EIGEN_EMPTY
 
-#if defined(_MSC_VER) && (_MSC_VER < 1800) && (!defined(__INTEL_COMPILER))
-#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
-  using Base::operator =;
-#elif defined(__clang__) // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653)
-#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
-  using Base::operator =; \
-  EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) { Base::operator=(other); return *this; } \
-  template <typename OtherDerived> \
-  EIGEN_STRONG_INLINE Derived& operator=(const DenseBase<OtherDerived>& other) { Base::operator=(other.derived()); return *this; }
+#if EIGEN_COMP_MSVC_STRICT && (EIGEN_COMP_MSVC < 1900 || EIGEN_CUDACC_VER>0)
+  // for older MSVC versions, as well as 1900 && CUDA 8, using the base operator is sufficient (cf Bugs 1000, 1324)
+  #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
+    using Base::operator =;
+#elif EIGEN_COMP_CLANG // workaround clang bug (see http://forum.kde.org/viewtopic.php?f=74&t=102653)
+  #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
+    using Base::operator =; \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) { Base::operator=(other); return *this; } \
+    template <typename OtherDerived> \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const DenseBase<OtherDerived>& other) { Base::operator=(other.derived()); return *this; }
 #else
-#define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
-  using Base::operator =; \
-  EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) \
-  { \
-    Base::operator=(other); \
-    return *this; \
-  }
+  #define EIGEN_INHERIT_ASSIGNMENT_EQUAL_OPERATOR(Derived) \
+    using Base::operator =; \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const Derived& other) \
+    { \
+      Base::operator=(other); \
+      return *this; \
+    }
 #endif
 
+
 /** \internal
  * \brief Macro to manually inherit assignment operators.
  * This is necessary, because the implicitly defined assignment operator gets deleted when a custom operator= is defined.
@@ -351,32 +853,12 @@
   typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \
   typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \
   typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \
-  typedef typename Eigen::internal::nested<Derived>::type Nested; \
+  typedef typename Eigen::internal::ref_selector<Derived>::type Nested; \
   typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \
-  typedef typename Eigen::internal::traits<Derived>::Index Index; \
+  typedef typename Eigen::internal::traits<Derived>::StorageIndex StorageIndex; \
   enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \
         ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \
         Flags = Eigen::internal::traits<Derived>::Flags, \
-        CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
-        SizeAtCompileTime = Base::SizeAtCompileTime, \
-        MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
-        IsVectorAtCompileTime = Base::IsVectorAtCompileTime };
-
-
-#define EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
-  typedef typename Eigen::internal::traits<Derived>::Scalar Scalar; /*!< \brief Numeric type, e.g. float, double, int or std::complex<float>. */ \
-  typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; /*!< \brief The underlying numeric type for composed scalar types. \details In cases where Scalar is e.g. std::complex<T>, T were corresponding to RealScalar. */ \
-  typedef typename Base::PacketScalar PacketScalar; \
-  typedef typename Base::CoeffReturnType CoeffReturnType; /*!< \brief The return type for coefficient access. \details Depending on whether the object allows direct coefficient access (e.g. for a MatrixXd), this type is either 'const Scalar&' or simply 'Scalar' for objects that do not allow direct coefficient access. */ \
-  typedef typename Eigen::internal::nested<Derived>::type Nested; \
-  typedef typename Eigen::internal::traits<Derived>::StorageKind StorageKind; \
-  typedef typename Eigen::internal::traits<Derived>::Index Index; \
-  enum { RowsAtCompileTime = Eigen::internal::traits<Derived>::RowsAtCompileTime, \
-        ColsAtCompileTime = Eigen::internal::traits<Derived>::ColsAtCompileTime, \
-        MaxRowsAtCompileTime = Eigen::internal::traits<Derived>::MaxRowsAtCompileTime, \
-        MaxColsAtCompileTime = Eigen::internal::traits<Derived>::MaxColsAtCompileTime, \
-        Flags = Eigen::internal::traits<Derived>::Flags, \
-        CoeffReadCost = Eigen::internal::traits<Derived>::CoeffReadCost, \
         SizeAtCompileTime = Base::SizeAtCompileTime, \
         MaxSizeAtCompileTime = Base::MaxSizeAtCompileTime, \
         IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; \
@@ -384,6 +866,12 @@
   using Base::const_cast_derived;
 
 
+// FIXME Maybe the EIGEN_DENSE_PUBLIC_INTERFACE could be removed as importing PacketScalar is rarely needed
+#define EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \
+  EIGEN_GENERIC_PUBLIC_INTERFACE(Derived) \
+  typedef typename Base::PacketScalar PacketScalar;
+
+
 #define EIGEN_PLAIN_ENUM_MIN(a,b) (((int)a <= (int)b) ? (int)a : (int)b)
 #define EIGEN_PLAIN_ENUM_MAX(a,b) (((int)a >= (int)b) ? (int)a : (int)b)
 
@@ -413,18 +901,10 @@
 
 #define EIGEN_IMPLIES(a,b) (!(a) || (b))
 
-#define EIGEN_MAKE_CWISE_BINARY_OP(METHOD,FUNCTOR) \
-  template<typename OtherDerived> \
-  EIGEN_STRONG_INLINE const CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived> \
-  (METHOD)(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
-  { \
-    return CwiseBinaryOp<FUNCTOR<Scalar>, const Derived, const OtherDerived>(derived(), other.derived()); \
-  }
-
-// the expression type of a cwise product
-#define EIGEN_CWISE_PRODUCT_RETURN_TYPE(LHS,RHS) \
+// the expression type of a standard coefficient wise binary operation
+#define EIGEN_CWISE_BINARY_RETURN_TYPE(LHS,RHS,OPNAME) \
     CwiseBinaryOp< \
-      internal::scalar_product_op< \
+      EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)< \
           typename internal::traits<LHS>::Scalar, \
           typename internal::traits<RHS>::Scalar \
       >, \
@@ -432,4 +912,90 @@
       const RHS \
     >
 
+#define EIGEN_MAKE_CWISE_BINARY_OP(METHOD,OPNAME) \
+  template<typename OtherDerived> \
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,OPNAME) \
+  (METHOD)(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
+  { \
+    return EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,OPNAME)(derived(), other.derived()); \
+  }
+
+#define EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,TYPEA,TYPEB) \
+  (Eigen::internal::has_ReturnType<Eigen::ScalarBinaryOpTraits<TYPEA,TYPEB,EIGEN_CAT(EIGEN_CAT(Eigen::internal::scalar_,OPNAME),_op)<TYPEA,TYPEB>  > >::value)
+
+#define EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(EXPR,SCALAR,OPNAME) \
+  CwiseBinaryOp<EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)<typename internal::traits<EXPR>::Scalar,SCALAR>, const EXPR, \
+                const typename internal::plain_constant_type<EXPR,SCALAR>::type>
+
+#define EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(SCALAR,EXPR,OPNAME) \
+  CwiseBinaryOp<EIGEN_CAT(EIGEN_CAT(internal::scalar_,OPNAME),_op)<SCALAR,typename internal::traits<EXPR>::Scalar>, \
+                const typename internal::plain_constant_type<EXPR,SCALAR>::type, const EXPR>
+
+// Workaround for MSVC 2010 (see ML thread "patch with compile for for MSVC 2010")
+#if EIGEN_COMP_MSVC_STRICT<=1600
+#define EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(X) typename internal::enable_if<true,X>::type
+#else
+#define EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(X) X
+#endif
+
+#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(METHOD,OPNAME) \
+  template <typename T> EIGEN_DEVICE_FUNC inline \
+  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename internal::promote_scalar_arg<Scalar EIGEN_COMMA T EIGEN_COMMA EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,Scalar,T)>::type,OPNAME))\
+  (METHOD)(const T& scalar) const { \
+    typedef typename internal::promote_scalar_arg<Scalar,T,EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,Scalar,T)>::type PromotedT; \
+    return EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,PromotedT,OPNAME)(derived(), \
+           typename internal::plain_constant_type<Derived,PromotedT>::type(derived().rows(), derived().cols(), internal::scalar_constant_op<PromotedT>(scalar))); \
+  }
+
+#define EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(METHOD,OPNAME) \
+  template <typename T> EIGEN_DEVICE_FUNC inline friend \
+  EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE(const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename internal::promote_scalar_arg<Scalar EIGEN_COMMA T EIGEN_COMMA EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,T,Scalar)>::type,Derived,OPNAME)) \
+  (METHOD)(const T& scalar, const StorageBaseType& matrix) { \
+    typedef typename internal::promote_scalar_arg<Scalar,T,EIGEN_SCALAR_BINARY_SUPPORTED(OPNAME,T,Scalar)>::type PromotedT; \
+    return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(PromotedT,Derived,OPNAME)( \
+           typename internal::plain_constant_type<Derived,PromotedT>::type(matrix.derived().rows(), matrix.derived().cols(), internal::scalar_constant_op<PromotedT>(scalar)), matrix.derived()); \
+  }
+
+#define EIGEN_MAKE_SCALAR_BINARY_OP(METHOD,OPNAME) \
+  EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(METHOD,OPNAME) \
+  EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(METHOD,OPNAME)
+
+
+#ifdef EIGEN_EXCEPTIONS
+#  define EIGEN_THROW_X(X) throw X
+#  define EIGEN_THROW throw
+#  define EIGEN_TRY try
+#  define EIGEN_CATCH(X) catch (X)
+#else
+#  ifdef __CUDA_ARCH__
+#    define EIGEN_THROW_X(X) asm("trap;")
+#    define EIGEN_THROW asm("trap;")
+#  else
+#    define EIGEN_THROW_X(X) std::abort()
+#    define EIGEN_THROW std::abort()
+#  endif
+#  define EIGEN_TRY if (true)
+#  define EIGEN_CATCH(X) else
+#endif
+
+
+#if EIGEN_HAS_CXX11_NOEXCEPT
+#   define EIGEN_INCLUDE_TYPE_TRAITS
+#   define EIGEN_NOEXCEPT noexcept
+#   define EIGEN_NOEXCEPT_IF(x) noexcept(x)
+#   define EIGEN_NO_THROW noexcept(true)
+#   define EIGEN_EXCEPTION_SPEC(X) noexcept(false)
+#else
+#   define EIGEN_NOEXCEPT
+#   define EIGEN_NOEXCEPT_IF(x)
+#   define EIGEN_NO_THROW throw()
+#   if EIGEN_COMP_MSVC
+      // MSVC does not support exception specifications (warning C4290),
+      // and they are deprecated in c++11 anyway.
+#     define EIGEN_EXCEPTION_SPEC(X) throw()
+#   else
+#     define EIGEN_EXCEPTION_SPEC(X) throw(X)
+#   endif
+#endif
+
 #endif // EIGEN_MACROS_H
diff --git a/Eigen/src/Core/util/Memory.h b/Eigen/src/Core/util/Memory.h
index b9af5cf..291383c 100644
--- a/Eigen/src/Core/util/Memory.h
+++ b/Eigen/src/Core/util/Memory.h
@@ -1,11 +1,12 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2008-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
 // Copyright (C) 2009 Kenneth Riddile <kfriddile@yahoo.com>
 // Copyright (C) 2010 Hauke Heibel <hauke.heibel@gmail.com>
 // Copyright (C) 2010 Thomas Capricelli <orzel@freehackers.org>
+// Copyright (C) 2013 Pavel Holoborodko <pavel@holoborodko.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -31,7 +32,7 @@
 // page 114, "[The] LP64 model [...] is used by all 64-bit UNIX ports" so it's indeed
 // quite safe, at least within the context of glibc, to equate 64-bit with LP64.
 #if defined(__GLIBC__) && ((__GLIBC__>=2 && __GLIBC_MINOR__ >= 8) || __GLIBC__>2) \
- && defined(__LP64__) && ! defined( __SANITIZE_ADDRESS__ )
+ && defined(__LP64__) && ! defined( __SANITIZE_ADDRESS__ ) && (EIGEN_DEFAULT_ALIGN_BYTES == 16)
   #define EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED 1
 #else
   #define EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED 0
@@ -41,15 +42,15 @@
 //   See http://svn.freebsd.org/viewvc/base/stable/6/lib/libc/stdlib/malloc.c?view=markup
 // FreeBSD 7 seems to have 16-byte aligned malloc except on ARM and MIPS architectures
 //   See http://svn.freebsd.org/viewvc/base/stable/7/lib/libc/stdlib/malloc.c?view=markup
-#if defined(__FreeBSD__) && !defined(__arm__) && !defined(__mips__)
+#if defined(__FreeBSD__) && !(EIGEN_ARCH_ARM || EIGEN_ARCH_MIPS) && (EIGEN_DEFAULT_ALIGN_BYTES == 16)
   #define EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED 1
 #else
   #define EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED 0
 #endif
 
-#if defined(__APPLE__) \
- || defined(_WIN64) \
- || EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED \
+#if (EIGEN_OS_MAC && (EIGEN_DEFAULT_ALIGN_BYTES == 16))     \
+ || (EIGEN_OS_WIN64 && (EIGEN_DEFAULT_ALIGN_BYTES == 16))   \
+ || EIGEN_GLIBC_MALLOC_ALREADY_ALIGNED              \
  || EIGEN_FREEBSD_MALLOC_ALREADY_ALIGNED
   #define EIGEN_MALLOC_ALREADY_ALIGNED 1
 #else
@@ -58,37 +59,18 @@
 
 #endif
 
-// See bug 554 (http://eigen.tuxfamily.org/bz/show_bug.cgi?id=554)
-// It seems to be unsafe to check _POSIX_ADVISORY_INFO without including unistd.h first.
-// Currently, let's include it only on unix systems:
-#if defined(__unix__) || defined(__unix)
-  #include <unistd.h>
-  #if ((defined __QNXNTO__) || (defined _GNU_SOURCE) || (defined __PGI) || ((defined _XOPEN_SOURCE) && (_XOPEN_SOURCE >= 600))) && (defined _POSIX_ADVISORY_INFO) && (_POSIX_ADVISORY_INFO > 0)
-    #define EIGEN_HAS_POSIX_MEMALIGN 1
-  #endif
-#endif
-
-#ifndef EIGEN_HAS_POSIX_MEMALIGN
-  #define EIGEN_HAS_POSIX_MEMALIGN 0
-#endif
-
-#ifdef EIGEN_VECTORIZE_SSE
-  #define EIGEN_HAS_MM_MALLOC 1
-#else
-  #define EIGEN_HAS_MM_MALLOC 0
-#endif
-
 namespace Eigen {
 
 namespace internal {
 
+EIGEN_DEVICE_FUNC 
 inline void throw_std_bad_alloc()
 {
   #ifdef EIGEN_EXCEPTIONS
     throw std::bad_alloc();
   #else
-    std::size_t huge = -1;
-    new int[huge];
+    std::size_t huge = static_cast<std::size_t>(-1);
+    ::operator new(huge);
   #endif
 }
 
@@ -103,9 +85,9 @@
   */
 inline void* handmade_aligned_malloc(std::size_t size)
 {
-  void *original = std::malloc(size+16);
+  void *original = std::malloc(size+EIGEN_DEFAULT_ALIGN_BYTES);
   if (original == 0) return 0;
-  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(15))) + 16);
+  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1))) + EIGEN_DEFAULT_ALIGN_BYTES);
   *(reinterpret_cast<void**>(aligned) - 1) = original;
   return aligned;
 }
@@ -118,7 +100,7 @@
 
 /** \internal
   * \brief Reallocates aligned memory.
-  * Since we know that our handmade version is based on std::realloc
+  * Since we know that our handmade version is based on std::malloc
   * we can use std::realloc to implement efficient reallocation.
   */
 inline void* handmade_aligned_realloc(void* ptr, std::size_t size, std::size_t = 0)
@@ -126,9 +108,9 @@
   if (ptr == 0) return handmade_aligned_malloc(size);
   void *original = *(reinterpret_cast<void**>(ptr) - 1);
   std::ptrdiff_t previous_offset = static_cast<char *>(ptr)-static_cast<char *>(original);
-  original = std::realloc(original,size+16);
+  original = std::realloc(original,size+EIGEN_DEFAULT_ALIGN_BYTES);
   if (original == 0) return 0;
-  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(15))) + 16);
+  void *aligned = reinterpret_cast<void*>((reinterpret_cast<std::size_t>(original) & ~(std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1))) + EIGEN_DEFAULT_ALIGN_BYTES);
   void *previous_aligned = static_cast<char *>(original)+previous_offset;
   if(aligned!=previous_aligned)
     std::memmove(aligned, previous_aligned, size);
@@ -138,92 +120,46 @@
 }
 
 /*****************************************************************************
-*** Implementation of generic aligned realloc (when no realloc can be used)***
-*****************************************************************************/
-
-void* aligned_malloc(std::size_t size);
-void  aligned_free(void *ptr);
-
-/** \internal
-  * \brief Reallocates aligned memory.
-  * Allows reallocation with aligned ptr types. This implementation will
-  * always create a new memory chunk and copy the old data.
-  */
-inline void* generic_aligned_realloc(void* ptr, size_t size, size_t old_size)
-{
-  if (ptr==0)
-    return aligned_malloc(size);
-
-  if (size==0)
-  {
-    aligned_free(ptr);
-    return 0;
-  }
-
-  void* newptr = aligned_malloc(size);
-  if (newptr == 0)
-  {
-    #ifdef EIGEN_HAS_ERRNO
-    errno = ENOMEM; // according to the standard
-    #endif
-    return 0;
-  }
-
-  if (ptr != 0)
-  {
-    std::memcpy(newptr, ptr, (std::min)(size,old_size));
-    aligned_free(ptr);
-  }
-
-  return newptr;
-}
-
-/*****************************************************************************
 *** Implementation of portable aligned versions of malloc/free/realloc     ***
 *****************************************************************************/
 
 #ifdef EIGEN_NO_MALLOC
-inline void check_that_malloc_is_allowed()
+EIGEN_DEVICE_FUNC inline void check_that_malloc_is_allowed()
 {
   eigen_assert(false && "heap allocation is forbidden (EIGEN_NO_MALLOC is defined)");
 }
 #elif defined EIGEN_RUNTIME_NO_MALLOC
-inline bool is_malloc_allowed_impl(bool update, bool new_value = false)
+EIGEN_DEVICE_FUNC inline bool is_malloc_allowed_impl(bool update, bool new_value = false)
 {
   static bool value = true;
   if (update == 1)
     value = new_value;
   return value;
 }
-inline bool is_malloc_allowed() { return is_malloc_allowed_impl(false); }
-inline bool set_is_malloc_allowed(bool new_value) { return is_malloc_allowed_impl(true, new_value); }
-inline void check_that_malloc_is_allowed()
+EIGEN_DEVICE_FUNC inline bool is_malloc_allowed() { return is_malloc_allowed_impl(false); }
+EIGEN_DEVICE_FUNC inline bool set_is_malloc_allowed(bool new_value) { return is_malloc_allowed_impl(true, new_value); }
+EIGEN_DEVICE_FUNC inline void check_that_malloc_is_allowed()
 {
   eigen_assert(is_malloc_allowed() && "heap allocation is forbidden (EIGEN_RUNTIME_NO_MALLOC is defined and g_is_malloc_allowed is false)");
 }
 #else 
-inline void check_that_malloc_is_allowed()
+EIGEN_DEVICE_FUNC inline void check_that_malloc_is_allowed()
 {}
 #endif
 
-/** \internal Allocates \a size bytes. The returned pointer is guaranteed to have 16 bytes alignment.
+/** \internal Allocates \a size bytes. The returned pointer is guaranteed to have 16 or 32 bytes alignment depending on the requirements.
   * On allocation error, the returned pointer is null, and std::bad_alloc is thrown.
   */
-inline void* aligned_malloc(size_t size)
+EIGEN_DEVICE_FUNC inline void* aligned_malloc(std::size_t size)
 {
   check_that_malloc_is_allowed();
 
   void *result;
-  #if !EIGEN_ALIGN
+  #if (EIGEN_DEFAULT_ALIGN_BYTES==0) || EIGEN_MALLOC_ALREADY_ALIGNED
     result = std::malloc(size);
-  #elif EIGEN_MALLOC_ALREADY_ALIGNED
-    result = std::malloc(size);
-  #elif EIGEN_HAS_POSIX_MEMALIGN
-    if(posix_memalign(&result, 16, size)) result = 0;
-  #elif EIGEN_HAS_MM_MALLOC
-    result = _mm_malloc(size, 16);
-  #elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
-    result = _aligned_malloc(size, 16);
+    #if EIGEN_DEFAULT_ALIGN_BYTES==16
+    eigen_assert((size<16 || (std::size_t(result)%16)==0) && "System's malloc returned an unaligned pointer. Compile with EIGEN_MALLOC_ALREADY_ALIGNED=0 to fallback to handmade alignd memory allocator.");
+    #endif
   #else
     result = handmade_aligned_malloc(size);
   #endif
@@ -235,50 +171,27 @@
 }
 
 /** \internal Frees memory allocated with aligned_malloc. */
-inline void aligned_free(void *ptr)
+EIGEN_DEVICE_FUNC inline void aligned_free(void *ptr)
 {
-  #if !EIGEN_ALIGN
+  #if (EIGEN_DEFAULT_ALIGN_BYTES==0) || EIGEN_MALLOC_ALREADY_ALIGNED
     std::free(ptr);
-  #elif EIGEN_MALLOC_ALREADY_ALIGNED
-    std::free(ptr);
-  #elif EIGEN_HAS_POSIX_MEMALIGN
-    std::free(ptr);
-  #elif EIGEN_HAS_MM_MALLOC
-    _mm_free(ptr);
-  #elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
-    _aligned_free(ptr);
   #else
     handmade_aligned_free(ptr);
   #endif
 }
 
 /**
-* \internal
-* \brief Reallocates an aligned block of memory.
-* \throws std::bad_alloc on allocation failure
-**/
-inline void* aligned_realloc(void *ptr, size_t new_size, size_t old_size)
+  * \internal
+  * \brief Reallocates an aligned block of memory.
+  * \throws std::bad_alloc on allocation failure
+  */
+inline void* aligned_realloc(void *ptr, std::size_t new_size, std::size_t old_size)
 {
   EIGEN_UNUSED_VARIABLE(old_size);
 
   void *result;
-#if !EIGEN_ALIGN
+#if (EIGEN_DEFAULT_ALIGN_BYTES==0) || EIGEN_MALLOC_ALREADY_ALIGNED
   result = std::realloc(ptr,new_size);
-#elif EIGEN_MALLOC_ALREADY_ALIGNED
-  result = std::realloc(ptr,new_size);
-#elif EIGEN_HAS_POSIX_MEMALIGN
-  result = generic_aligned_realloc(ptr,new_size,old_size);
-#elif EIGEN_HAS_MM_MALLOC
-  // The defined(_mm_free) is just here to verify that this MSVC version
-  // implements _mm_malloc/_mm_free based on the corresponding _aligned_
-  // functions. This may not always be the case and we just try to be safe.
-  #if defined(_MSC_VER) && (!defined(_WIN32_WCE)) && defined(_mm_free)
-    result = _aligned_realloc(ptr,new_size,16);
-  #else
-    result = generic_aligned_realloc(ptr,new_size,old_size);
-  #endif
-#elif defined(_MSC_VER) && (!defined(_WIN32_WCE))
-  result = _aligned_realloc(ptr,new_size,16);
 #else
   result = handmade_aligned_realloc(ptr,new_size,old_size);
 #endif
@@ -296,12 +209,12 @@
 /** \internal Allocates \a size bytes. If Align is true, then the returned ptr is 16-byte-aligned.
   * On allocation error, the returned pointer is null, and a std::bad_alloc is thrown.
   */
-template<bool Align> inline void* conditional_aligned_malloc(size_t size)
+template<bool Align> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc(std::size_t size)
 {
   return aligned_malloc(size);
 }
 
-template<> inline void* conditional_aligned_malloc<false>(size_t size)
+template<> EIGEN_DEVICE_FUNC inline void* conditional_aligned_malloc<false>(std::size_t size)
 {
   check_that_malloc_is_allowed();
 
@@ -312,22 +225,22 @@
 }
 
 /** \internal Frees memory allocated with conditional_aligned_malloc */
-template<bool Align> inline void conditional_aligned_free(void *ptr)
+template<bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_free(void *ptr)
 {
   aligned_free(ptr);
 }
 
-template<> inline void conditional_aligned_free<false>(void *ptr)
+template<> EIGEN_DEVICE_FUNC inline void conditional_aligned_free<false>(void *ptr)
 {
   std::free(ptr);
 }
 
-template<bool Align> inline void* conditional_aligned_realloc(void* ptr, size_t new_size, size_t old_size)
+template<bool Align> inline void* conditional_aligned_realloc(void* ptr, std::size_t new_size, std::size_t old_size)
 {
   return aligned_realloc(ptr, new_size, old_size);
 }
 
-template<> inline void* conditional_aligned_realloc<false>(void* ptr, size_t new_size, size_t)
+template<> inline void* conditional_aligned_realloc<false>(void* ptr, std::size_t new_size, std::size_t)
 {
   return std::realloc(ptr, new_size);
 }
@@ -336,33 +249,43 @@
 *** Construction/destruction of array elements                             ***
 *****************************************************************************/
 
-/** \internal Constructs the elements of an array.
-  * The \a size parameter tells on how many objects to call the constructor of T.
-  */
-template<typename T> inline T* construct_elements_of_array(T *ptr, size_t size)
-{
-  for (size_t i=0; i < size; ++i) ::new (ptr + i) T;
-  return ptr;
-}
-
 /** \internal Destructs the elements of an array.
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T> inline void destruct_elements_of_array(T *ptr, size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline void destruct_elements_of_array(T *ptr, std::size_t size)
 {
   // always destruct an array starting from the end.
   if(ptr)
     while(size) ptr[--size].~T();
 }
 
+/** \internal Constructs the elements of an array.
+  * The \a size parameter tells on how many objects to call the constructor of T.
+  */
+template<typename T> EIGEN_DEVICE_FUNC inline T* construct_elements_of_array(T *ptr, std::size_t size)
+{
+  std::size_t i;
+  EIGEN_TRY
+  {
+      for (i = 0; i < size; ++i) ::new (ptr + i) T;
+      return ptr;
+  }
+  EIGEN_CATCH(...)
+  {
+    destruct_elements_of_array(ptr, i);
+    EIGEN_THROW;
+  }
+  return NULL;
+}
+
 /*****************************************************************************
 *** Implementation of aligned new/delete-like functions                    ***
 *****************************************************************************/
 
 template<typename T>
-EIGEN_ALWAYS_INLINE void check_size_for_overflow(size_t size)
+EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void check_size_for_overflow(std::size_t size)
 {
-  if(size > size_t(-1) / sizeof(T))
+  if(size > std::size_t(-1) / sizeof(T))
     throw_std_bad_alloc();
 }
 
@@ -370,24 +293,42 @@
   * On allocation error, the returned pointer is undefined, but a std::bad_alloc is thrown.
   * The default constructor of T is called.
   */
-template<typename T> inline T* aligned_new(size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline T* aligned_new(std::size_t size)
 {
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(aligned_malloc(sizeof(T)*size));
-  return construct_elements_of_array(result, size);
+  EIGEN_TRY
+  {
+    return construct_elements_of_array(result, size);
+  }
+  EIGEN_CATCH(...)
+  {
+    aligned_free(result);
+    EIGEN_THROW;
+  }
+  return result;
 }
 
-template<typename T, bool Align> inline T* conditional_aligned_new(size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new(std::size_t size)
 {
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
-  return construct_elements_of_array(result, size);
+  EIGEN_TRY
+  {
+    return construct_elements_of_array(result, size);
+  }
+  EIGEN_CATCH(...)
+  {
+    conditional_aligned_free<Align>(result);
+    EIGEN_THROW;
+  }
+  return result;
 }
 
 /** \internal Deletes objects constructed with aligned_new
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T> inline void aligned_delete(T *ptr, size_t size)
+template<typename T> EIGEN_DEVICE_FUNC inline void aligned_delete(T *ptr, std::size_t size)
 {
   destruct_elements_of_array<T>(ptr, size);
   aligned_free(ptr);
@@ -396,13 +337,13 @@
 /** \internal Deletes objects constructed with conditional_aligned_new
   * The \a size parameters tells on how many objects to call the destructor of T.
   */
-template<typename T, bool Align> inline void conditional_aligned_delete(T *ptr, size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete(T *ptr, std::size_t size)
 {
   destruct_elements_of_array<T>(ptr, size);
   conditional_aligned_free<Align>(ptr);
 }
 
-template<typename T, bool Align> inline T* conditional_aligned_realloc_new(T* pts, size_t new_size, size_t old_size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_realloc_new(T* pts, std::size_t new_size, std::size_t old_size)
 {
   check_size_for_overflow<T>(new_size);
   check_size_for_overflow<T>(old_size);
@@ -410,23 +351,43 @@
     destruct_elements_of_array(pts+new_size, old_size-new_size);
   T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
   if(new_size > old_size)
-    construct_elements_of_array(result+old_size, new_size-old_size);
+  {
+    EIGEN_TRY
+    {
+      construct_elements_of_array(result+old_size, new_size-old_size);
+    }
+    EIGEN_CATCH(...)
+    {
+      conditional_aligned_free<Align>(result);
+      EIGEN_THROW;
+    }
+  }
   return result;
 }
 
 
-template<typename T, bool Align> inline T* conditional_aligned_new_auto(size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline T* conditional_aligned_new_auto(std::size_t size)
 {
   if(size==0)
     return 0; // short-cut. Also fixes Bug 884
   check_size_for_overflow<T>(size);
   T *result = reinterpret_cast<T*>(conditional_aligned_malloc<Align>(sizeof(T)*size));
   if(NumTraits<T>::RequireInitialization)
-    construct_elements_of_array(result, size);
+  {
+    EIGEN_TRY
+    {
+      construct_elements_of_array(result, size);
+    }
+    EIGEN_CATCH(...)
+    {
+      conditional_aligned_free<Align>(result);
+      EIGEN_THROW;
+    }
+  }
   return result;
 }
 
-template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, size_t new_size, size_t old_size)
+template<typename T, bool Align> inline T* conditional_aligned_realloc_new_auto(T* pts, std::size_t new_size, std::size_t old_size)
 {
   check_size_for_overflow<T>(new_size);
   check_size_for_overflow<T>(old_size);
@@ -434,11 +395,21 @@
     destruct_elements_of_array(pts+new_size, old_size-new_size);
   T *result = reinterpret_cast<T*>(conditional_aligned_realloc<Align>(reinterpret_cast<void*>(pts), sizeof(T)*new_size, sizeof(T)*old_size));
   if(NumTraits<T>::RequireInitialization && (new_size > old_size))
-    construct_elements_of_array(result+old_size, new_size-old_size);
+  {
+    EIGEN_TRY
+    {
+      construct_elements_of_array(result+old_size, new_size-old_size);
+    }
+    EIGEN_CATCH(...)
+    {
+      conditional_aligned_free<Align>(result);
+      EIGEN_THROW;
+    }
+  }
   return result;
 }
 
-template<typename T, bool Align> inline void conditional_aligned_delete_auto(T *ptr, size_t size)
+template<typename T, bool Align> EIGEN_DEVICE_FUNC inline void conditional_aligned_delete_auto(T *ptr, std::size_t size)
 {
   if(NumTraits<T>::RequireInitialization)
     destruct_elements_of_array<T>(ptr, size);
@@ -447,51 +418,62 @@
 
 /****************************************************************************/
 
-/** \internal Returns the index of the first element of the array that is well aligned for vectorization.
+/** \internal Returns the index of the first element of the array that is well aligned with respect to the requested \a Alignment.
   *
+  * \tparam Alignment requested alignment in Bytes.
   * \param array the address of the start of the array
   * \param size the size of the array
   *
-  * \note If no element of the array is well aligned, the size of the array is returned. Typically,
-  * for example with SSE, "well aligned" means 16-byte-aligned. If vectorization is disabled or if the
+  * \note If no element of the array is well aligned or the requested alignment is not a multiple of a scalar,
+  * the size of the array is returned. For example with SSE, the requested alignment is typically 16-bytes. If
   * packet size for the given scalar type is 1, then everything is considered well-aligned.
   *
-  * \note If the scalar type is vectorizable, we rely on the following assumptions: sizeof(Scalar) is a
-  * power of 2, the packet size in bytes is also a power of 2, and is a multiple of sizeof(Scalar). On the
-  * other hand, we do not assume that the array address is a multiple of sizeof(Scalar), as that fails for
+  * \note Otherwise, if the Alignment is larger that the scalar size, we rely on the assumptions that sizeof(Scalar) is a
+  * power of 2. On the other hand, we do not assume that the array address is a multiple of sizeof(Scalar), as that fails for
   * example with Scalar=double on certain 32-bit platforms, see bug #79.
   *
   * There is also the variant first_aligned(const MatrixBase&) defined in DenseCoeffsBase.h.
+  * \sa first_default_aligned()
   */
-template<typename Scalar, typename Index>
-static inline Index first_aligned(const Scalar* array, Index size)
+template<int Alignment, typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC inline Index first_aligned(const Scalar* array, Index size)
 {
-  static const Index PacketSize = packet_traits<Scalar>::size;
-  static const Index PacketAlignedMask = PacketSize-1;
+  const Index ScalarSize = sizeof(Scalar);
+  const Index AlignmentSize = Alignment / ScalarSize;
+  const Index AlignmentMask = AlignmentSize-1;
 
-  if(PacketSize==1)
+  if(AlignmentSize<=1)
   {
-    // Either there is no vectorization, or a packet consists of exactly 1 scalar so that all elements
-    // of the array have the same alignment.
+    // Either the requested alignment if smaller than a scalar, or it exactly match a 1 scalar
+    // so that all elements of the array have the same alignment.
     return 0;
   }
-  else if(size_t(array) & (sizeof(Scalar)-1))
+  else if( (UIntPtr(array) & (sizeof(Scalar)-1)) || (Alignment%ScalarSize)!=0)
   {
-    // There is vectorization for this scalar type, but the array is not aligned to the size of a single scalar.
+    // The array is not aligned to the size of a single scalar, or the requested alignment is not a multiple of the scalar size.
     // Consequently, no element of the array is well aligned.
     return size;
   }
   else
   {
-    return std::min<Index>( (PacketSize - (Index((size_t(array)/sizeof(Scalar))) & PacketAlignedMask))
-                           & PacketAlignedMask, size);
+    Index first = (AlignmentSize - (Index((UIntPtr(array)/sizeof(Scalar))) & AlignmentMask)) & AlignmentMask;
+    return (first < size) ? first : size;
   }
 }
 
+/** \internal Returns the index of the first element of the array that is well aligned with respect the largest packet requirement.
+   * \sa first_aligned(Scalar*,Index) and first_default_aligned(DenseBase<Derived>) */
+template<typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC inline Index first_default_aligned(const Scalar* array, Index size)
+{
+  typedef typename packet_traits<Scalar>::type DefaultPacketType;
+  return first_aligned<unpacket_traits<DefaultPacketType>::alignment>(array, size);
+}
+
 /** \internal Returns the smallest integer multiple of \a base and greater or equal to \a size
   */ 
 template<typename Index> 
-inline static Index first_multiple(Index size, Index base)
+inline Index first_multiple(Index size, Index base)
 {
   return ((size+base-1)/base)*base;
 }
@@ -500,21 +482,59 @@
 // use memcpy on trivial types, i.e., on types that does not require an initialization ctor.
 template<typename T, bool UseMemcpy> struct smart_copy_helper;
 
-template<typename T> void smart_copy(const T* start, const T* end, T* target)
+template<typename T> EIGEN_DEVICE_FUNC void smart_copy(const T* start, const T* end, T* target)
 {
   smart_copy_helper<T,!NumTraits<T>::RequireInitialization>::run(start, end, target);
 }
 
 template<typename T> struct smart_copy_helper<T,true> {
-  static inline void run(const T* start, const T* end, T* target)
-  { memcpy(target, start, std::ptrdiff_t(end)-std::ptrdiff_t(start)); }
+  EIGEN_DEVICE_FUNC static inline void run(const T* start, const T* end, T* target)
+  {
+    IntPtr size = IntPtr(end)-IntPtr(start);
+    if(size==0) return;
+    eigen_internal_assert(start!=0 && end!=0 && target!=0);
+    std::memcpy(target, start, size);
+  }
 };
 
 template<typename T> struct smart_copy_helper<T,false> {
-  static inline void run(const T* start, const T* end, T* target)
+  EIGEN_DEVICE_FUNC static inline void run(const T* start, const T* end, T* target)
   { std::copy(start, end, target); }
 };
 
+// intelligent memmove. falls back to std::memmove for POD types, uses std::copy otherwise. 
+template<typename T, bool UseMemmove> struct smart_memmove_helper;
+
+template<typename T> void smart_memmove(const T* start, const T* end, T* target)
+{
+  smart_memmove_helper<T,!NumTraits<T>::RequireInitialization>::run(start, end, target);
+}
+
+template<typename T> struct smart_memmove_helper<T,true> {
+  static inline void run(const T* start, const T* end, T* target)
+  {
+    IntPtr size = IntPtr(end)-IntPtr(start);
+    if(size==0) return;
+    eigen_internal_assert(start!=0 && end!=0 && target!=0);
+    std::memmove(target, start, size);
+  }
+};
+
+template<typename T> struct smart_memmove_helper<T,false> {
+  static inline void run(const T* start, const T* end, T* target)
+  { 
+    if (UIntPtr(target) < UIntPtr(start))
+    {
+      std::copy(start, end, target);
+    }
+    else                                 
+    {
+      std::ptrdiff_t count = (std::ptrdiff_t(end)-std::ptrdiff_t(start)) / sizeof(T);
+      std::copy_backward(start, end, target + count); 
+    }
+  }
+};
+
 
 /*****************************************************************************
 *** Implementation of runtime stack allocation (falling back to malloc)    ***
@@ -523,16 +543,16 @@
 // you can overwrite Eigen's default behavior regarding alloca by defining EIGEN_ALLOCA
 // to the appropriate stack allocation function
 #ifndef EIGEN_ALLOCA
-  #if (defined __linux__) || (defined __APPLE__) || (defined alloca)
+  #if EIGEN_OS_LINUX || EIGEN_OS_MAC || (defined alloca)
     #define EIGEN_ALLOCA alloca
-  #elif defined(_MSC_VER)
+  #elif EIGEN_COMP_MSVC
     #define EIGEN_ALLOCA _alloca
   #endif
 #endif
 
 // This helper class construct the allocated memory, and takes care of destructing and freeing the handled data
 // at destruction time. In practice this helper class is mainly useful to avoid memory leak in case of exceptions.
-template<typename T> class aligned_stack_memory_handler
+template<typename T> class aligned_stack_memory_handler : noncopyable
 {
   public:
     /* Creates a stack_memory_handler responsible for the buffer \a ptr of size \a size.
@@ -541,7 +561,7 @@
      * In this case, the buffer elements will also be destructed when this handler will be destructed.
      * Finally, if \a dealloc is true, then the pointer \a ptr is freed.
      **/
-    aligned_stack_memory_handler(T* ptr, size_t size, bool dealloc)
+    aligned_stack_memory_handler(T* ptr, std::size_t size, bool dealloc)
       : m_ptr(ptr), m_size(size), m_deallocate(dealloc)
     {
       if(NumTraits<T>::RequireInitialization && m_ptr)
@@ -556,10 +576,34 @@
     }
   protected:
     T* m_ptr;
-    size_t m_size;
+    std::size_t m_size;
     bool m_deallocate;
 };
 
+template<typename T> class scoped_array : noncopyable
+{
+  T* m_ptr;
+public:
+  explicit scoped_array(std::ptrdiff_t size)
+  {
+    m_ptr = new T[size];
+  }
+  ~scoped_array()
+  {
+    delete[] m_ptr;
+  }
+  T& operator[](std::ptrdiff_t i) { return m_ptr[i]; }
+  const T& operator[](std::ptrdiff_t i) const { return m_ptr[i]; }
+  T* &ptr() { return m_ptr; }
+  const T* ptr() const { return m_ptr; }
+  operator const T*() const { return m_ptr; }
+};
+
+template<typename T> void swap(scoped_array<T> &a,scoped_array<T> &b)
+{
+  std::swap(a.ptr(),b.ptr());
+}
+    
 } // end namespace internal
 
 /** \internal
@@ -578,11 +622,13 @@
   * The underlying stack allocation function can controlled with the EIGEN_ALLOCA preprocessor token.
   */
 #ifdef EIGEN_ALLOCA
-
-  #if defined(__arm__) || defined(_WIN32)
-    #define EIGEN_ALIGNED_ALLOCA(SIZE) reinterpret_cast<void*>((reinterpret_cast<size_t>(EIGEN_ALLOCA(SIZE+16)) & ~(size_t(15))) + 16)
+  
+  #if EIGEN_DEFAULT_ALIGN_BYTES>0
+    // We always manually re-align the result of EIGEN_ALLOCA.
+    // If alloca is already aligned, the compiler should be smart enough to optimize away the re-alignment.
+    #define EIGEN_ALIGNED_ALLOCA(SIZE) reinterpret_cast<void*>((internal::UIntPtr(EIGEN_ALLOCA(SIZE+EIGEN_DEFAULT_ALIGN_BYTES-1)) + EIGEN_DEFAULT_ALIGN_BYTES-1) & ~(std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1)))
   #else
-    #define EIGEN_ALIGNED_ALLOCA EIGEN_ALLOCA
+    #define EIGEN_ALIGNED_ALLOCA(SIZE) EIGEN_ALLOCA(SIZE)
   #endif
 
   #define ei_declare_aligned_stack_constructed_variable(TYPE,NAME,SIZE,BUFFER) \
@@ -607,39 +653,33 @@
 *** Implementation of EIGEN_MAKE_ALIGNED_OPERATOR_NEW [_IF]                ***
 *****************************************************************************/
 
-#if EIGEN_ALIGN
-  #ifdef EIGEN_EXCEPTIONS
-    #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
-      void* operator new(size_t size, const std::nothrow_t&) throw() { \
-        try { return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); } \
-        catch (...) { return 0; } \
+#if EIGEN_MAX_ALIGN_BYTES!=0
+  #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
+      void* operator new(std::size_t size, const std::nothrow_t&) EIGEN_NO_THROW { \
+        EIGEN_TRY { return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); } \
+        EIGEN_CATCH (...) { return 0; } \
       }
-  #else
-    #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
-      void* operator new(size_t size, const std::nothrow_t&) throw() { \
-        return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
-      }
-  #endif
-
   #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign) \
-      void *operator new(size_t size) { \
+      void *operator new(std::size_t size) { \
         return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
       } \
-      void *operator new[](size_t size) { \
+      void *operator new[](std::size_t size) { \
         return Eigen::internal::conditional_aligned_malloc<NeedsToAlign>(size); \
       } \
-      void operator delete(void * ptr) throw() { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
-      void operator delete[](void * ptr) throw() { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete(void * ptr) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete[](void * ptr) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete(void * ptr, std::size_t /* sz */) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
+      void operator delete[](void * ptr, std::size_t /* sz */) EIGEN_NO_THROW { Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); } \
       /* in-place new and delete. since (at least afaik) there is no actual   */ \
       /* memory allocated we can safely let the default implementation handle */ \
       /* this particular case. */ \
-      static void *operator new(size_t size, void *ptr) { return ::operator new(size,ptr); } \
-      static void *operator new[](size_t size, void* ptr) { return ::operator new[](size,ptr); } \
-      void operator delete(void * memory, void *ptr) throw() { return ::operator delete(memory,ptr); } \
-      void operator delete[](void * memory, void *ptr) throw() { return ::operator delete[](memory,ptr); } \
+      static void *operator new(std::size_t size, void *ptr) { return ::operator new(size,ptr); } \
+      static void *operator new[](std::size_t size, void* ptr) { return ::operator new[](size,ptr); } \
+      void operator delete(void * memory, void *ptr) EIGEN_NO_THROW { return ::operator delete(memory,ptr); } \
+      void operator delete[](void * memory, void *ptr) EIGEN_NO_THROW { return ::operator delete[](memory,ptr); } \
       /* nothrow-new (returns zero instead of std::bad_alloc) */ \
       EIGEN_MAKE_ALIGNED_OPERATOR_NEW_NOTHROW(NeedsToAlign) \
-      void operator delete(void *ptr, const std::nothrow_t&) throw() { \
+      void operator delete(void *ptr, const std::nothrow_t&) EIGEN_NO_THROW { \
         Eigen::internal::conditional_aligned_free<NeedsToAlign>(ptr); \
       } \
       typedef void eigen_aligned_operator_new_marker_type;
@@ -649,14 +689,22 @@
 
 #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(true)
 #define EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(Scalar,Size) \
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(bool(((Size)!=Eigen::Dynamic) && ((sizeof(Scalar)*(Size))%16==0)))
+  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(bool(((Size)!=Eigen::Dynamic) && ((sizeof(Scalar)*(Size))%EIGEN_MAX_ALIGN_BYTES==0)))
 
 /****************************************************************************/
 
 /** \class aligned_allocator
 * \ingroup Core_Module
 *
-* \brief STL compatible allocator to use with with 16 byte aligned types
+* \brief STL compatible allocator to use with types requiring a non standrad alignment.
+*
+* The memory is aligned as for dynamically aligned matrix/array types such as MatrixXd.
+* By default, it will thus provide at least 16 bytes alignment and more in following cases:
+*  - 32 bytes alignment if AVX is enabled.
+*  - 64 bytes alignment if AVX512 is enabled.
+*
+* This can be controled using the \c EIGEN_MAX_ALIGN_BYTES macro as documented
+* \link TopicPreprocessorDirectivesPerformance there \endlink.
 *
 * Example:
 * \code
@@ -667,96 +715,64 @@
 * std::map< int, Vector3f > my_map_vec3;
 * \endcode
 *
-* \sa \ref TopicStlContainers.
+* \sa \blank \ref TopicStlContainers.
 */
 template<class T>
-class aligned_allocator
+class aligned_allocator : public std::allocator<T>
 {
 public:
-    typedef size_t    size_type;
-    typedef std::ptrdiff_t difference_type;
-    typedef T*        pointer;
-    typedef const T*  const_pointer;
-    typedef T&        reference;
-    typedef const T&  const_reference;
-    typedef T         value_type;
+  typedef std::size_t     size_type;
+  typedef std::ptrdiff_t  difference_type;
+  typedef T*              pointer;
+  typedef const T*        const_pointer;
+  typedef T&              reference;
+  typedef const T&        const_reference;
+  typedef T               value_type;
 
-    template<class U>
-    struct rebind
-    {
-        typedef aligned_allocator<U> other;
-    };
+  template<class U>
+  struct rebind
+  {
+    typedef aligned_allocator<U> other;
+  };
 
-    pointer address( reference value ) const
-    {
-        return &value;
-    }
+  aligned_allocator() : std::allocator<T>() {}
 
-    const_pointer address( const_reference value ) const
-    {
-        return &value;
-    }
+  aligned_allocator(const aligned_allocator& other) : std::allocator<T>(other) {}
 
-    aligned_allocator()
-    {
-    }
+  template<class U>
+  aligned_allocator(const aligned_allocator<U>& other) : std::allocator<T>(other) {}
 
-    aligned_allocator( const aligned_allocator& )
-    {
-    }
+  ~aligned_allocator() {}
 
-    template<class U>
-    aligned_allocator( const aligned_allocator<U>& )
-    {
-    }
+  pointer allocate(size_type num, const void* /*hint*/ = 0)
+  {
+    internal::check_size_for_overflow<T>(num);
+    size_type size = num * sizeof(T);
+#if EIGEN_COMP_GNUC_STRICT && EIGEN_GNUC_AT_LEAST(7,0)
+    // workaround gcc bug https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87544
+    // It triggered eigen/Eigen/src/Core/util/Memory.h:189:12: warning: argument 1 value '18446744073709551612' exceeds maximum object size 9223372036854775807
+    if(size>=std::size_t((std::numeric_limits<std::ptrdiff_t>::max)()))
+      return 0;
+    else
+#endif
+      return static_cast<pointer>( internal::aligned_malloc(size) );
+  }
 
-    ~aligned_allocator()
-    {
-    }
-
-    size_type max_size() const
-    {
-        return (std::numeric_limits<size_type>::max)();
-    }
-
-    pointer allocate( size_type num, const void* hint = 0 )
-    {
-        EIGEN_UNUSED_VARIABLE(hint);
-        internal::check_size_for_overflow<T>(num);
-        return static_cast<pointer>( internal::aligned_malloc( num * sizeof(T) ) );
-    }
-
-    void construct( pointer p, const T& value )
-    {
-        ::new( p ) T( value );
-    }
-
-    void destroy( pointer p )
-    {
-        p->~T();
-    }
-
-    void deallocate( pointer p, size_type /*num*/ )
-    {
-        internal::aligned_free( p );
-    }
-
-    bool operator!=(const aligned_allocator<T>& ) const
-    { return false; }
-
-    bool operator==(const aligned_allocator<T>& ) const
-    { return true; }
+  void deallocate(pointer p, size_type /*num*/)
+  {
+    internal::aligned_free(p);
+  }
 };
 
 //---------- Cache sizes ----------
 
 #if !defined(EIGEN_NO_CPUID)
-#  if defined(__GNUC__) && ( defined(__i386__) || defined(__x86_64__) )
-#    if defined(__PIC__) && defined(__i386__)
+#  if EIGEN_COMP_GNUC && EIGEN_ARCH_i386_OR_x86_64
+#    if defined(__PIC__) && EIGEN_ARCH_i386
        // Case for x86 with PIC
 #      define EIGEN_CPUID(abcd,func,id) \
          __asm__ __volatile__ ("xchgl %%ebx, %k1;cpuid; xchgl %%ebx,%k1": "=a" (abcd[0]), "=&r" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "a" (func), "c" (id));
-#    elif defined(__PIC__) && defined(__x86_64__)
+#    elif defined(__PIC__) && EIGEN_ARCH_x86_64
        // Case for x64 with PIC. In theory this is only a problem with recent gcc and with medium or large code model, not with the default small code model.
        // However, we cannot detect which code model is used, and the xchg overhead is negligible anyway.
 #      define EIGEN_CPUID(abcd,func,id) \
@@ -766,8 +782,8 @@
 #      define EIGEN_CPUID(abcd,func,id) \
          __asm__ __volatile__ ("cpuid": "=a" (abcd[0]), "=b" (abcd[1]), "=c" (abcd[2]), "=d" (abcd[3]) : "0" (func), "2" (id) );
 #    endif
-#  elif defined(_MSC_VER)
-#    if (_MSC_VER > 1500) && ( defined(_M_IX86) || defined(_M_X64) )
+#  elif EIGEN_COMP_MSVC
+#    if (EIGEN_COMP_MSVC > 1500) && EIGEN_ARCH_i386_OR_x86_64
 #      define EIGEN_CPUID(abcd,func,id) __cpuidex((int*)abcd,func,id)
 #    endif
 #  endif
diff --git a/Eigen/src/Core/util/Meta.h b/Eigen/src/Core/util/Meta.h
old mode 100644
new mode 100755
index 71d5871..d31e954
--- a/Eigen/src/Core/util/Meta.h
+++ b/Eigen/src/Core/util/Meta.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -11,8 +11,27 @@
 #ifndef EIGEN_META_H
 #define EIGEN_META_H
 
+#if defined(__CUDA_ARCH__)
+#include <cfloat>
+#include <math_constants.h>
+#endif
+
+#if EIGEN_COMP_ICC>=1600 &&  __cplusplus >= 201103L
+#include <cstdint>
+#endif
+
 namespace Eigen {
 
+typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE DenseIndex;
+
+/**
+ * \brief The Index type as used for the API.
+ * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE.
+ * \sa \blank \ref TopicPreprocessorDirectives, StorageIndex.
+ */
+
+typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE Index;
+
 namespace internal {
 
 /** \internal
@@ -22,6 +41,16 @@
   * we however don't want to add a dependency to Boost.
   */
 
+// Only recent versions of ICC complain about using ptrdiff_t to hold pointers,
+// and older versions do not provide *intptr_t types.
+#if EIGEN_COMP_ICC>=1600 &&  __cplusplus >= 201103L
+typedef std::intptr_t  IntPtr;
+typedef std::uintptr_t UIntPtr;
+#else
+typedef std::ptrdiff_t IntPtr;
+typedef std::size_t UIntPtr;
+#endif
+
 struct true_type {  enum { value = 1 }; };
 struct false_type { enum { value = 0 }; };
 
@@ -68,6 +97,40 @@
 template<> struct is_arithmetic<signed long>   { enum { value = true }; };
 template<> struct is_arithmetic<unsigned long> { enum { value = true }; };
 
+template<typename T> struct is_integral        { enum { value = false }; };
+template<> struct is_integral<bool>            { enum { value = true }; };
+template<> struct is_integral<char>            { enum { value = true }; };
+template<> struct is_integral<signed char>     { enum { value = true }; };
+template<> struct is_integral<unsigned char>   { enum { value = true }; };
+template<> struct is_integral<signed short>    { enum { value = true }; };
+template<> struct is_integral<unsigned short>  { enum { value = true }; };
+template<> struct is_integral<signed int>      { enum { value = true }; };
+template<> struct is_integral<unsigned int>    { enum { value = true }; };
+template<> struct is_integral<signed long>     { enum { value = true }; };
+template<> struct is_integral<unsigned long>   { enum { value = true }; };
+
+#if EIGEN_HAS_CXX11
+using std::make_unsigned;
+#else
+// TODO: Possibly improve this implementation of make_unsigned.
+// It is currently used only by
+// template<typename Scalar> struct random_default_impl<Scalar, false, true>.
+template<typename> struct make_unsigned;
+template<> struct make_unsigned<char>             { typedef unsigned char type; };
+template<> struct make_unsigned<signed char>      { typedef unsigned char type; };
+template<> struct make_unsigned<unsigned char>    { typedef unsigned char type; };
+template<> struct make_unsigned<signed short>     { typedef unsigned short type; };
+template<> struct make_unsigned<unsigned short>   { typedef unsigned short type; };
+template<> struct make_unsigned<signed int>       { typedef unsigned int type; };
+template<> struct make_unsigned<unsigned int>     { typedef unsigned int type; };
+template<> struct make_unsigned<signed long>      { typedef unsigned long type; };
+template<> struct make_unsigned<unsigned long>    { typedef unsigned long type; };
+#if EIGEN_COMP_MSVC
+template<> struct make_unsigned<signed __int64>   { typedef unsigned __int64 type; };
+template<> struct make_unsigned<unsigned __int64> { typedef unsigned __int64 type; };
+#endif
+#endif
+
 template <typename T> struct add_const { typedef const T type; };
 template <typename T> struct add_const<T&> { typedef T& type; };
 
@@ -80,29 +143,163 @@
 template<typename T> struct add_const_on_value_type<T* const>  { typedef T const* const type; };
 template<typename T> struct add_const_on_value_type<T const* const>  { typedef T const* const type; };
 
+
+template<typename From, typename To>
+struct is_convertible_impl
+{
+private:
+  struct any_conversion
+  {
+    template <typename T> any_conversion(const volatile T&);
+    template <typename T> any_conversion(T&);
+  };
+  struct yes {int a[1];};
+  struct no  {int a[2];};
+
+  static yes test(const To&, int);
+  static no  test(any_conversion, ...);
+
+public:
+  static From ms_from;
+#ifdef __INTEL_COMPILER
+  #pragma warning push
+  #pragma warning ( disable : 2259 )
+#endif
+  enum { value = sizeof(test(ms_from, 0))==sizeof(yes) };
+#ifdef __INTEL_COMPILER
+  #pragma warning pop
+#endif
+};
+
+template<typename From, typename To>
+struct is_convertible
+{
+  enum { value = is_convertible_impl<typename remove_all<From>::type,
+                                     typename remove_all<To  >::type>::value };
+};
+
 /** \internal Allows to enable/disable an overload
   * according to a compile time condition.
   */
-template<bool Condition, typename T> struct enable_if;
+template<bool Condition, typename T=void> struct enable_if;
 
 template<typename T> struct enable_if<true,T>
 { typedef T type; };
 
+#if defined(__CUDA_ARCH__)
+#if !defined(__FLT_EPSILON__)
+#define __FLT_EPSILON__ FLT_EPSILON
+#define __DBL_EPSILON__ DBL_EPSILON
+#endif
 
+namespace device {
+
+template<typename T> struct numeric_limits
+{
+  EIGEN_DEVICE_FUNC
+  static T epsilon() { return 0; }
+  static T (max)() { assert(false && "Highest not supported for this type"); }
+  static T (min)() { assert(false && "Lowest not supported for this type"); }
+  static T infinity() { assert(false && "Infinity not supported for this type"); }
+  static T quiet_NaN() { assert(false && "quiet_NaN not supported for this type"); }
+};
+template<> struct numeric_limits<float>
+{
+  EIGEN_DEVICE_FUNC
+  static float epsilon() { return __FLT_EPSILON__; }
+  EIGEN_DEVICE_FUNC
+  static float (max)() { return CUDART_MAX_NORMAL_F; }
+  EIGEN_DEVICE_FUNC
+  static float (min)() { return FLT_MIN; }
+  EIGEN_DEVICE_FUNC
+  static float infinity() { return CUDART_INF_F; }
+  EIGEN_DEVICE_FUNC
+  static float quiet_NaN() { return CUDART_NAN_F; }
+};
+template<> struct numeric_limits<double>
+{
+  EIGEN_DEVICE_FUNC
+  static double epsilon() { return __DBL_EPSILON__; }
+  EIGEN_DEVICE_FUNC
+  static double (max)() { return DBL_MAX; }
+  EIGEN_DEVICE_FUNC
+  static double (min)() { return DBL_MIN; }
+  EIGEN_DEVICE_FUNC
+  static double infinity() { return CUDART_INF; }
+  EIGEN_DEVICE_FUNC
+  static double quiet_NaN() { return CUDART_NAN; }
+};
+template<> struct numeric_limits<int>
+{
+  EIGEN_DEVICE_FUNC
+  static int epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static int (max)() { return INT_MAX; }
+  EIGEN_DEVICE_FUNC
+  static int (min)() { return INT_MIN; }
+};
+template<> struct numeric_limits<unsigned int>
+{
+  EIGEN_DEVICE_FUNC
+  static unsigned int epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static unsigned int (max)() { return UINT_MAX; }
+  EIGEN_DEVICE_FUNC
+  static unsigned int (min)() { return 0; }
+};
+template<> struct numeric_limits<long>
+{
+  EIGEN_DEVICE_FUNC
+  static long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static long (max)() { return LONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static long (min)() { return LONG_MIN; }
+};
+template<> struct numeric_limits<unsigned long>
+{
+  EIGEN_DEVICE_FUNC
+  static unsigned long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long (max)() { return ULONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long (min)() { return 0; }
+};
+template<> struct numeric_limits<long long>
+{
+  EIGEN_DEVICE_FUNC
+  static long long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static long long (max)() { return LLONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static long long (min)() { return LLONG_MIN; }
+};
+template<> struct numeric_limits<unsigned long long>
+{
+  EIGEN_DEVICE_FUNC
+  static unsigned long long epsilon() { return 0; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long long (max)() { return ULLONG_MAX; }
+  EIGEN_DEVICE_FUNC
+  static unsigned long long (min)() { return 0; }
+};
+
+}
+
+#endif
 
 /** \internal
   * A base class do disable default copy ctor and copy assignement operator.
   */
 class noncopyable
 {
-  noncopyable(const noncopyable&);
-  const noncopyable& operator=(const noncopyable&);
+  EIGEN_DEVICE_FUNC noncopyable(const noncopyable&);
+  EIGEN_DEVICE_FUNC const noncopyable& operator=(const noncopyable&);
 protected:
-  noncopyable() {}
-  ~noncopyable() {}
+  EIGEN_DEVICE_FUNC noncopyable() {}
+  EIGEN_DEVICE_FUNC ~noncopyable() {}
 };
 
-
 /** \internal
   * Convenient struct to get the result type of a unary or binary functor.
   *
@@ -110,14 +307,20 @@
   * upcoming next STL generation (using a templated result member).
   * If none of these members is provided, then the type of the first argument is returned. FIXME, that behavior is a pretty bad hack.
   */
-template<typename T> struct result_of {};
+#if EIGEN_HAS_STD_RESULT_OF
+template<typename T> struct result_of {
+  typedef typename std::result_of<T>::type type1;
+  typedef typename remove_all<type1>::type type;
+};
+#else
+template<typename T> struct result_of { };
 
 struct has_none {int a[1];};
 struct has_std_result_type {int a[2];};
 struct has_tr1_result {int a[3];};
 
 template<typename Func, typename ArgType, int SizeOf=sizeof(has_none)>
-struct unary_result_of_select {typedef ArgType type;};
+struct unary_result_of_select {typedef typename internal::remove_all<ArgType>::type type;};
 
 template<typename Func, typename ArgType>
 struct unary_result_of_select<Func, ArgType, sizeof(has_std_result_type)> {typedef typename Func::result_type type;};
@@ -128,10 +331,10 @@
 template<typename Func, typename ArgType>
 struct result_of<Func(ArgType)> {
     template<typename T>
-    static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
+    static has_std_result_type    testFunctor(T const *, typename T::result_type const * = 0);
     template<typename T>
-    static has_tr1_result      testFunctor(T const *, typename T::template result<T(ArgType)>::type const * = 0);
-    static has_none            testFunctor(...);
+    static has_tr1_result         testFunctor(T const *, typename T::template result<T(ArgType)>::type const * = 0);
+    static has_none               testFunctor(...);
 
     // note that the following indirection is needed for gcc-3.3
     enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
@@ -139,7 +342,7 @@
 };
 
 template<typename Func, typename ArgType0, typename ArgType1, int SizeOf=sizeof(has_none)>
-struct binary_result_of_select {typedef ArgType0 type;};
+struct binary_result_of_select {typedef typename internal::remove_all<ArgType0>::type type;};
 
 template<typename Func, typename ArgType0, typename ArgType1>
 struct binary_result_of_select<Func, ArgType0, ArgType1, sizeof(has_std_result_type)>
@@ -152,16 +355,83 @@
 template<typename Func, typename ArgType0, typename ArgType1>
 struct result_of<Func(ArgType0,ArgType1)> {
     template<typename T>
-    static has_std_result_type testFunctor(T const *, typename T::result_type const * = 0);
+    static has_std_result_type    testFunctor(T const *, typename T::result_type const * = 0);
     template<typename T>
-    static has_tr1_result      testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1)>::type const * = 0);
-    static has_none            testFunctor(...);
+    static has_tr1_result         testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1)>::type const * = 0);
+    static has_none               testFunctor(...);
 
     // note that the following indirection is needed for gcc-3.3
     enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
     typedef typename binary_result_of_select<Func, ArgType0, ArgType1, FunctorType>::type type;
 };
 
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2, int SizeOf=sizeof(has_none)>
+struct ternary_result_of_select {typedef typename internal::remove_all<ArgType0>::type type;};
+
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
+struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_std_result_type)>
+{typedef typename Func::result_type type;};
+
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
+struct ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, sizeof(has_tr1_result)>
+{typedef typename Func::template result<Func(ArgType0,ArgType1,ArgType2)>::type type;};
+
+template<typename Func, typename ArgType0, typename ArgType1, typename ArgType2>
+struct result_of<Func(ArgType0,ArgType1,ArgType2)> {
+    template<typename T>
+    static has_std_result_type    testFunctor(T const *, typename T::result_type const * = 0);
+    template<typename T>
+    static has_tr1_result         testFunctor(T const *, typename T::template result<T(ArgType0,ArgType1,ArgType2)>::type const * = 0);
+    static has_none               testFunctor(...);
+
+    // note that the following indirection is needed for gcc-3.3
+    enum {FunctorType = sizeof(testFunctor(static_cast<Func*>(0)))};
+    typedef typename ternary_result_of_select<Func, ArgType0, ArgType1, ArgType2, FunctorType>::type type;
+};
+#endif
+
+struct meta_yes { char a[1]; };
+struct meta_no  { char a[2]; };
+
+// Check whether T::ReturnType does exist
+template <typename T>
+struct has_ReturnType
+{
+  template <typename C> static meta_yes testFunctor(typename C::ReturnType const *);
+  template <typename C> static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor<T>(0)) == sizeof(meta_yes) };
+};
+
+template<typename T> const T* return_ptr();
+
+template <typename T, typename IndexType=Index>
+struct has_nullary_operator
+{
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()())>0)>::type * = 0);
+  static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
+};
+
+template <typename T, typename IndexType=Index>
+struct has_unary_operator
+{
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0)))>0)>::type * = 0);
+  static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
+};
+
+template <typename T, typename IndexType=Index>
+struct has_binary_operator
+{
+  template <typename C> static meta_yes testFunctor(C const *,typename enable_if<(sizeof(return_ptr<C>()->operator()(IndexType(0),IndexType(0)))>0)>::type * = 0);
+  static meta_no testFunctor(...);
+
+  enum { value = sizeof(testFunctor(static_cast<T*>(0))) == sizeof(meta_yes) };
+};
+
 /** \internal In short, it computes int(sqrt(\a Y)) with \a Y an integer.
   * Usage example: \code meta_sqrt<1023>::ret \endcode
   */
@@ -185,59 +455,80 @@
 template<int Y, int InfX, int SupX>
 class meta_sqrt<Y, InfX, SupX, true> { public:  enum { ret = (SupX*SupX <= Y) ? SupX : InfX }; };
 
+
+/** \internal Computes the least common multiple of two positive integer A and B
+  * at compile-time. It implements a naive algorithm testing all multiples of A.
+  * It thus works better if A>=B.
+  */
+template<int A, int B, int K=1, bool Done = ((A*K)%B)==0>
+struct meta_least_common_multiple
+{
+  enum { ret = meta_least_common_multiple<A,B,K+1>::ret };
+};
+template<int A, int B, int K>
+struct meta_least_common_multiple<A,B,K,true>
+{
+  enum { ret = A*K };
+};
+
 /** \internal determines whether the product of two numeric types is allowed and what the return type is */
 template<typename T, typename U> struct scalar_product_traits
 {
   enum { Defined = 0 };
 };
 
-template<typename T> struct scalar_product_traits<T,T>
-{
-  enum {
-    // Cost = NumTraits<T>::MulCost,
-    Defined = 1
-  };
-  typedef T ReturnType;
-};
-
-template<typename T> struct scalar_product_traits<T,std::complex<T> >
-{
-  enum {
-    // Cost = 2*NumTraits<T>::MulCost,
-    Defined = 1
-  };
-  typedef std::complex<T> ReturnType;
-};
-
-template<typename T> struct scalar_product_traits<std::complex<T>, T>
-{
-  enum {
-    // Cost = 2*NumTraits<T>::MulCost,
-    Defined = 1
-  };
-  typedef std::complex<T> ReturnType;
-};
-
 // FIXME quick workaround around current limitation of result_of
 // template<typename Scalar, typename ArgType0, typename ArgType1>
 // struct result_of<scalar_product_op<Scalar>(ArgType0,ArgType1)> {
 // typedef typename scalar_product_traits<typename remove_all<ArgType0>::type, typename remove_all<ArgType1>::type>::ReturnType type;
 // };
 
-template<typename T> struct is_diagonal
-{ enum { ret = false }; };
-
-template<typename T> struct is_diagonal<DiagonalBase<T> >
-{ enum { ret = true }; };
-
-template<typename T> struct is_diagonal<DiagonalWrapper<T> >
-{ enum { ret = true }; };
-
-template<typename T, int S> struct is_diagonal<DiagonalMatrix<T,S> >
-{ enum { ret = true }; };
-
 } // end namespace internal
 
+namespace numext {
+  
+#if defined(__CUDA_ARCH__)
+template<typename T> EIGEN_DEVICE_FUNC   void swap(T &a, T &b) { T tmp = b; b = a; a = tmp; }
+#else
+template<typename T> EIGEN_STRONG_INLINE void swap(T &a, T &b) { std::swap(a,b); }
+#endif
+
+#if defined(__CUDA_ARCH__)
+using internal::device::numeric_limits;
+#else
+using std::numeric_limits;
+#endif
+
+// Integer division with rounding up.
+// T is assumed to be an integer type with a>=0, and b>0
+template<typename T>
+T div_ceil(const T &a, const T &b)
+{
+  return (a+b-1) / b;
+}
+
+// The aim of the following functions is to bypass -Wfloat-equal warnings
+// when we really want a strict equality comparison on floating points.
+template<typename X, typename Y> EIGEN_STRONG_INLINE
+bool equal_strict(const X& x,const Y& y) { return x == y; }
+
+template<> EIGEN_STRONG_INLINE
+bool equal_strict(const float& x,const float& y) { return std::equal_to<float>()(x,y); }
+
+template<> EIGEN_STRONG_INLINE
+bool equal_strict(const double& x,const double& y) { return std::equal_to<double>()(x,y); }
+
+template<typename X, typename Y> EIGEN_STRONG_INLINE
+bool not_equal_strict(const X& x,const Y& y) { return x != y; }
+
+template<> EIGEN_STRONG_INLINE
+bool not_equal_strict(const float& x,const float& y) { return std::not_equal_to<float>()(x,y); }
+
+template<> EIGEN_STRONG_INLINE
+bool not_equal_strict(const double& x,const double& y) { return std::not_equal_to<double>()(x,y); }
+
+} // end namespace numext
+
 } // end namespace Eigen
 
 #endif // EIGEN_META_H
diff --git a/Eigen/src/Core/util/ReenableStupidWarnings.h b/Eigen/src/Core/util/ReenableStupidWarnings.h
index 5ddfbd4..ecc82b7 100644
--- a/Eigen/src/Core/util/ReenableStupidWarnings.h
+++ b/Eigen/src/Core/util/ReenableStupidWarnings.h
@@ -8,7 +8,20 @@
     #pragma warning pop
   #elif defined __clang__
     #pragma clang diagnostic pop
+  #elif defined __GNUC__  &&  (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 6))
+    #pragma GCC diagnostic pop
   #endif
+
+  #if defined __NVCC__
+//    Don't reenable the diagnostic messages, as it turns out these messages need
+//    to be disabled at the point of the template instantiation (i.e the user code)
+//    otherwise they'll be triggered by nvcc.
+//    #pragma diag_default code_is_unreachable
+//    #pragma diag_default initialization_not_reachable
+//    #pragma diag_default 2651
+//    #pragma diag_default 2653
+  #endif
+
 #endif
 
 #endif // EIGEN_WARNINGS_DISABLED
diff --git a/Eigen/src/Core/util/StaticAssert.h b/Eigen/src/Core/util/StaticAssert.h
index bac5d9f..500e477 100644
--- a/Eigen/src/Core/util/StaticAssert.h
+++ b/Eigen/src/Core/util/StaticAssert.h
@@ -24,9 +24,10 @@
  *
  */
 
+#ifndef EIGEN_STATIC_ASSERT
 #ifndef EIGEN_NO_STATIC_ASSERT
 
-  #if defined(__GXX_EXPERIMENTAL_CXX0X__) || (defined(_MSC_VER) && (_MSC_VER >= 1600))
+  #if EIGEN_MAX_CPP_VER>=11 && (__has_feature(cxx_static_assert) || (defined(__cplusplus) && __cplusplus >= 201103L) || (EIGEN_COMP_MSVC >= 1600))
 
     // if native static_assert is enabled, let's use it
     #define EIGEN_STATIC_ASSERT(X,MSG) static_assert(X,#MSG);
@@ -44,55 +45,65 @@
     struct static_assertion<true>
     {
       enum {
-        YOU_TRIED_CALLING_A_VECTOR_METHOD_ON_A_MATRIX,
-        YOU_MIXED_VECTORS_OF_DIFFERENT_SIZES,
-        YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES,
-        THIS_METHOD_IS_ONLY_FOR_VECTORS_OF_A_SPECIFIC_SIZE,
-        THIS_METHOD_IS_ONLY_FOR_MATRICES_OF_A_SPECIFIC_SIZE,
-        THIS_METHOD_IS_ONLY_FOR_OBJECTS_OF_A_SPECIFIC_SIZE,
-        YOU_MADE_A_PROGRAMMING_MISTAKE,
-        EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT,
-        EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE,
-        YOU_CALLED_A_FIXED_SIZE_METHOD_ON_A_DYNAMIC_SIZE_MATRIX_OR_VECTOR,
-        YOU_CALLED_A_DYNAMIC_SIZE_METHOD_ON_A_FIXED_SIZE_MATRIX_OR_VECTOR,
-        UNALIGNED_LOAD_AND_STORE_OPERATIONS_UNIMPLEMENTED_ON_ALTIVEC,
-        THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES,
-        FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED,
-        NUMERIC_TYPE_MUST_BE_REAL,
-        COEFFICIENT_WRITE_ACCESS_TO_SELFADJOINT_NOT_SUPPORTED,
-        WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED,
-        THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE,
-        INVALID_MATRIX_PRODUCT,
-        INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS,
-        INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION,
-        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY,
-        THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES,
-        THIS_METHOD_IS_ONLY_FOR_ROW_MAJOR_MATRICES,
-        INVALID_MATRIX_TEMPLATE_PARAMETERS,
-        INVALID_MATRIXBASE_TEMPLATE_PARAMETERS,
-        BOTH_MATRICES_MUST_HAVE_THE_SAME_STORAGE_ORDER,
-        THIS_METHOD_IS_ONLY_FOR_DIAGONAL_MATRIX,
-        THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE,
-        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_WITH_DIRECT_MEMORY_ACCESS_SUCH_AS_MAP_OR_PLAIN_MATRICES,
-        YOU_ALREADY_SPECIFIED_THIS_STRIDE,
-        INVALID_STORAGE_ORDER_FOR_THIS_VECTOR_EXPRESSION,
-        THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD,
-        PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1,
-        THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS,
-        YOU_CANNOT_MIX_ARRAYS_AND_MATRICES,
-        YOU_PERFORMED_AN_INVALID_TRANSFORMATION_CONVERSION,
-        THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY,
-        YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT,
-        THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS,
-        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL,
-        THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES,
-        YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED,
-        YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED,
-        THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE,
-        THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH,
-        OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG,
-        IMPLICIT_CONVERSION_TO_SCALAR_IS_FOR_INNER_PRODUCT_ONLY,
-        STORAGE_LAYOUT_DOES_NOT_MATCH
+        YOU_TRIED_CALLING_A_VECTOR_METHOD_ON_A_MATRIX=1,
+        YOU_MIXED_VECTORS_OF_DIFFERENT_SIZES=1,
+        YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES=1,
+        THIS_METHOD_IS_ONLY_FOR_VECTORS_OF_A_SPECIFIC_SIZE=1,
+        THIS_METHOD_IS_ONLY_FOR_MATRICES_OF_A_SPECIFIC_SIZE=1,
+        THIS_METHOD_IS_ONLY_FOR_OBJECTS_OF_A_SPECIFIC_SIZE=1,
+        OUT_OF_RANGE_ACCESS=1,
+        YOU_MADE_A_PROGRAMMING_MISTAKE=1,
+        EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT=1,
+        EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE=1,
+        YOU_CALLED_A_FIXED_SIZE_METHOD_ON_A_DYNAMIC_SIZE_MATRIX_OR_VECTOR=1,
+        YOU_CALLED_A_DYNAMIC_SIZE_METHOD_ON_A_FIXED_SIZE_MATRIX_OR_VECTOR=1,
+        UNALIGNED_LOAD_AND_STORE_OPERATIONS_UNIMPLEMENTED_ON_ALTIVEC=1,
+        THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES=1,
+        FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED=1,
+        NUMERIC_TYPE_MUST_BE_REAL=1,
+        COEFFICIENT_WRITE_ACCESS_TO_SELFADJOINT_NOT_SUPPORTED=1,
+        WRITING_TO_TRIANGULAR_PART_WITH_UNIT_DIAGONAL_IS_NOT_SUPPORTED=1,
+        THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE=1,
+        INVALID_MATRIX_PRODUCT=1,
+        INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS=1,
+        INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION=1,
+        YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY=1,
+        THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES=1,
+        THIS_METHOD_IS_ONLY_FOR_ROW_MAJOR_MATRICES=1,
+        INVALID_MATRIX_TEMPLATE_PARAMETERS=1,
+        INVALID_MATRIXBASE_TEMPLATE_PARAMETERS=1,
+        BOTH_MATRICES_MUST_HAVE_THE_SAME_STORAGE_ORDER=1,
+        THIS_METHOD_IS_ONLY_FOR_DIAGONAL_MATRIX=1,
+        THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE=1,
+        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_WITH_DIRECT_MEMORY_ACCESS_SUCH_AS_MAP_OR_PLAIN_MATRICES=1,
+        YOU_ALREADY_SPECIFIED_THIS_STRIDE=1,
+        INVALID_STORAGE_ORDER_FOR_THIS_VECTOR_EXPRESSION=1,
+        THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD=1,
+        PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1=1,
+        THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS=1,
+        YOU_CANNOT_MIX_ARRAYS_AND_MATRICES=1,
+        YOU_PERFORMED_AN_INVALID_TRANSFORMATION_CONVERSION=1,
+        THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY=1,
+        YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT=1,
+        THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS=1,
+        THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS=1,
+        THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL=1,
+        THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES=1,
+        YOU_PASSED_A_ROW_VECTOR_BUT_A_COLUMN_VECTOR_WAS_EXPECTED=1,
+        YOU_PASSED_A_COLUMN_VECTOR_BUT_A_ROW_VECTOR_WAS_EXPECTED=1,
+        THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE=1,
+        THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH=1,
+        OBJECT_ALLOCATED_ON_STACK_IS_TOO_BIG=1,
+        IMPLICIT_CONVERSION_TO_SCALAR_IS_FOR_INNER_PRODUCT_ONLY=1,
+        STORAGE_LAYOUT_DOES_NOT_MATCH=1,
+        EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT__INVALID_COST_VALUE=1,
+        THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS=1,
+        MATRIX_FREE_CONJUGATE_GRADIENT_IS_COMPATIBLE_WITH_UPPER_UNION_LOWER_MODE_ONLY=1,
+        THIS_TYPE_IS_NOT_SUPPORTED=1,
+        STORAGE_KIND_MUST_MATCH=1,
+        STORAGE_INDEX_MUST_MATCH=1,
+        CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY=1,
+        SELFADJOINTVIEW_ACCEPTS_UPPER_AND_LOWER_MODE_ONLY=1
       };
     };
 
@@ -103,15 +114,15 @@
     // Specialized implementation for MSVC to avoid "conditional
     // expression is constant" warnings.  This implementation doesn't
     // appear to work under GCC, hence the multiple implementations.
-    #ifdef _MSC_VER
+    #if EIGEN_COMP_MSVC
 
       #define EIGEN_STATIC_ASSERT(CONDITION,MSG) \
         {Eigen::internal::static_assertion<bool(CONDITION)>::MSG;}
 
     #else
-
+      // In some cases clang interprets bool(CONDITION) as function declaration
       #define EIGEN_STATIC_ASSERT(CONDITION,MSG) \
-        if (Eigen::internal::static_assertion<bool(CONDITION)>::MSG) {}
+        if (Eigen::internal::static_assertion<static_cast<bool>(CONDITION)>::MSG) {}
 
     #endif
 
@@ -122,7 +133,7 @@
   #define EIGEN_STATIC_ASSERT(CONDITION,MSG) eigen_assert((CONDITION) && #MSG);
 
 #endif // EIGEN_NO_STATIC_ASSERT
-
+#endif // EIGEN_STATIC_ASSERT
 
 // static assertion failing if the type \a TYPE is not a vector type
 #define EIGEN_STATIC_ASSERT_VECTOR_ONLY(TYPE) \
@@ -159,7 +170,7 @@
 
 #define EIGEN_PREDICATE_SAME_MATRIX_SIZE(TYPE0,TYPE1) \
      ( \
-        (int(TYPE0::SizeAtCompileTime)==0 && int(TYPE1::SizeAtCompileTime)==0) \
+        (int(Eigen::internal::size_of_xpr_at_compile_time<TYPE0>::ret)==0 && int(Eigen::internal::size_of_xpr_at_compile_time<TYPE1>::ret)==0) \
     || (\
           (int(TYPE0::RowsAtCompileTime)==Eigen::Dynamic \
         || int(TYPE1::RowsAtCompileTime)==Eigen::Dynamic \
@@ -170,13 +181,8 @@
        ) \
      )
 
-#ifdef EIGEN2_SUPPORT
-  #define EIGEN_STATIC_ASSERT_NON_INTEGER(TYPE) \
-    eigen_assert(!NumTraits<Scalar>::IsInteger);
-#else
-  #define EIGEN_STATIC_ASSERT_NON_INTEGER(TYPE) \
+#define EIGEN_STATIC_ASSERT_NON_INTEGER(TYPE) \
     EIGEN_STATIC_ASSERT(!NumTraits<TYPE>::IsInteger, THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES)
-#endif
 
 
 // static assertion failing if it is guaranteed at compile-time that the two matrix expression types have different sizes
@@ -191,18 +197,22 @@
                           THIS_METHOD_IS_ONLY_FOR_1x1_EXPRESSIONS)
 
 #define EIGEN_STATIC_ASSERT_LVALUE(Derived) \
-      EIGEN_STATIC_ASSERT(internal::is_lvalue<Derived>::value, \
+      EIGEN_STATIC_ASSERT(Eigen::internal::is_lvalue<Derived>::value, \
                           THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY)
 
 #define EIGEN_STATIC_ASSERT_ARRAYXPR(Derived) \
-      EIGEN_STATIC_ASSERT((internal::is_same<typename internal::traits<Derived>::XprKind, ArrayXpr>::value), \
+      EIGEN_STATIC_ASSERT((Eigen::internal::is_same<typename Eigen::internal::traits<Derived>::XprKind, ArrayXpr>::value), \
                           THIS_METHOD_IS_ONLY_FOR_ARRAYS_NOT_MATRICES)
 
 #define EIGEN_STATIC_ASSERT_SAME_XPR_KIND(Derived1, Derived2) \
-      EIGEN_STATIC_ASSERT((internal::is_same<typename internal::traits<Derived1>::XprKind, \
-                                             typename internal::traits<Derived2>::XprKind \
+      EIGEN_STATIC_ASSERT((Eigen::internal::is_same<typename Eigen::internal::traits<Derived1>::XprKind, \
+                                             typename Eigen::internal::traits<Derived2>::XprKind \
                                             >::value), \
                           YOU_CANNOT_MIX_ARRAYS_AND_MATRICES)
 
+// Check that a cost value is positive, and that is stay within a reasonable range
+// TODO this check could be enabled for internal debugging only
+#define EIGEN_INTERNAL_CHECK_COST_VALUE(C) \
+      EIGEN_STATIC_ASSERT((C)>=0 && (C)<=HugeCost*HugeCost, EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT__INVALID_COST_VALUE);
 
 #endif // EIGEN_STATIC_ASSERT_H
diff --git a/Eigen/src/Core/util/XprHelper.h b/Eigen/src/Core/util/XprHelper.h
index 781965d..ba5bd18 100644
--- a/Eigen/src/Core/util/XprHelper.h
+++ b/Eigen/src/Core/util/XprHelper.h
@@ -14,20 +14,77 @@
 // just a workaround because GCC seems to not really like empty structs
 // FIXME: gcc 4.3 generates bad code when strict-aliasing is enabled
 // so currently we simply disable this optimization for gcc 4.3
-#if (defined __GNUG__) && !((__GNUC__==4) && (__GNUC_MINOR__==3))
+#if EIGEN_COMP_GNUC && !EIGEN_GNUC_AT(4,3)
   #define EIGEN_EMPTY_STRUCT_CTOR(X) \
-    EIGEN_STRONG_INLINE X() {} \
-    EIGEN_STRONG_INLINE X(const X& ) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE X() {} \
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE X(const X& ) {}
 #else
   #define EIGEN_EMPTY_STRUCT_CTOR(X)
 #endif
 
 namespace Eigen {
 
-typedef EIGEN_DEFAULT_DENSE_INDEX_TYPE DenseIndex;
-
 namespace internal {
 
+template<typename IndexDest, typename IndexSrc>
+EIGEN_DEVICE_FUNC
+inline IndexDest convert_index(const IndexSrc& idx) {
+  // for sizeof(IndexDest)>=sizeof(IndexSrc) compilers should be able to optimize this away:
+  eigen_internal_assert(idx <= NumTraits<IndexDest>::highest() && "Index value to big for target type");
+  return IndexDest(idx);
+}
+
+
+// promote_scalar_arg is an helper used in operation between an expression and a scalar, like:
+//    expression * scalar
+// Its role is to determine how the type T of the scalar operand should be promoted given the scalar type ExprScalar of the given expression.
+// The IsSupported template parameter must be provided by the caller as: internal::has_ReturnType<ScalarBinaryOpTraits<ExprScalar,T,op> >::value using the proper order for ExprScalar and T.
+// Then the logic is as follows:
+//  - if the operation is natively supported as defined by IsSupported, then the scalar type is not promoted, and T is returned.
+//  - otherwise, NumTraits<ExprScalar>::Literal is returned if T is implicitly convertible to NumTraits<ExprScalar>::Literal AND that this does not imply a float to integer conversion.
+//  - otherwise, ExprScalar is returned if T is implicitly convertible to ExprScalar AND that this does not imply a float to integer conversion.
+//  - In all other cases, the promoted type is not defined, and the respective operation is thus invalid and not available (SFINAE).
+template<typename ExprScalar,typename T, bool IsSupported>
+struct promote_scalar_arg;
+
+template<typename S,typename T>
+struct promote_scalar_arg<S,T,true>
+{
+  typedef T type;
+};
+
+// Recursively check safe conversion to PromotedType, and then ExprScalar if they are different.
+template<typename ExprScalar,typename T,typename PromotedType,
+  bool ConvertibleToLiteral = internal::is_convertible<T,PromotedType>::value,
+  bool IsSafe = NumTraits<T>::IsInteger || !NumTraits<PromotedType>::IsInteger>
+struct promote_scalar_arg_unsupported;
+
+// Start recursion with NumTraits<ExprScalar>::Literal
+template<typename S,typename T>
+struct promote_scalar_arg<S,T,false> : promote_scalar_arg_unsupported<S,T,typename NumTraits<S>::Literal> {};
+
+// We found a match!
+template<typename S,typename T, typename PromotedType>
+struct promote_scalar_arg_unsupported<S,T,PromotedType,true,true>
+{
+  typedef PromotedType type;
+};
+
+// No match, but no real-to-integer issues, and ExprScalar and current PromotedType are different,
+// so let's try to promote to ExprScalar
+template<typename ExprScalar,typename T, typename PromotedType>
+struct promote_scalar_arg_unsupported<ExprScalar,T,PromotedType,false,true>
+   : promote_scalar_arg_unsupported<ExprScalar,T,ExprScalar>
+{};
+
+// Unsafe real-to-integer, let's stop.
+template<typename S,typename T, typename PromotedType, bool ConvertibleToLiteral>
+struct promote_scalar_arg_unsupported<S,T,PromotedType,ConvertibleToLiteral,false> {};
+
+// T is not even convertible to ExprScalar, let's stop.
+template<typename S,typename T>
+struct promote_scalar_arg_unsupported<S,T,S,false,true> {};
+
 //classes inheriting no_assignment_operator don't generate a default operator=.
 class no_assignment_operator
 {
@@ -50,19 +107,19 @@
 {
   public:
     EIGEN_EMPTY_STRUCT_CTOR(variable_if_dynamic)
-    explicit variable_if_dynamic(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); assert(v == T(Value)); }
-    static T value() { return T(Value); }
-    void setValue(T) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamic(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); eigen_assert(v == T(Value)); }
+    EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE T value() { return T(Value); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T) {}
 };
 
 template<typename T> class variable_if_dynamic<T, Dynamic>
 {
     T m_value;
-    variable_if_dynamic() { assert(false); }
+    EIGEN_DEVICE_FUNC variable_if_dynamic() { eigen_assert(false); }
   public:
-    explicit variable_if_dynamic(T value) : m_value(value) {}
-    T value() const { return m_value; }
-    void setValue(T value) { m_value = value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamic(T value) : m_value(value) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T value() const { return m_value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T value) { m_value = value; }
 };
 
 /** \internal like variable_if_dynamic but for DynamicIndex
@@ -71,19 +128,19 @@
 {
   public:
     EIGEN_EMPTY_STRUCT_CTOR(variable_if_dynamicindex)
-    explicit variable_if_dynamicindex(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); assert(v == T(Value)); }
-    static T value() { return T(Value); }
-    void setValue(T) {}
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamicindex(T v) { EIGEN_ONLY_USED_FOR_DEBUG(v); eigen_assert(v == T(Value)); }
+    EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE T value() { return T(Value); }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T) {}
 };
 
 template<typename T> class variable_if_dynamicindex<T, DynamicIndex>
 {
     T m_value;
-    variable_if_dynamicindex() { assert(false); }
+    EIGEN_DEVICE_FUNC variable_if_dynamicindex() { eigen_assert(false); }
   public:
-    explicit variable_if_dynamicindex(T value) : m_value(value) {}
-    T value() const { return m_value; }
-    void setValue(T value) { m_value = value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit variable_if_dynamicindex(T value) : m_value(value) {}
+    EIGEN_DEVICE_FUNC T EIGEN_STRONG_INLINE value() const { return m_value; }
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void setValue(T value) { m_value = value; }
 };
 
 template<typename T> struct functor_traits
@@ -101,7 +158,73 @@
 template<typename T> struct unpacket_traits
 {
   typedef T type;
-  enum {size=1};
+  typedef T half;
+  enum
+  {
+    size = 1,
+    alignment = 1
+  };
+};
+
+template<int Size, typename PacketType,
+         bool Stop = Size==Dynamic || (Size%unpacket_traits<PacketType>::size)==0 || is_same<PacketType,typename unpacket_traits<PacketType>::half>::value>
+struct find_best_packet_helper;
+
+template< int Size, typename PacketType>
+struct find_best_packet_helper<Size,PacketType,true>
+{
+  typedef PacketType type;
+};
+
+template<int Size, typename PacketType>
+struct find_best_packet_helper<Size,PacketType,false>
+{
+  typedef typename find_best_packet_helper<Size,typename unpacket_traits<PacketType>::half>::type type;
+};
+
+template<typename T, int Size>
+struct find_best_packet
+{
+  typedef typename find_best_packet_helper<Size,typename packet_traits<T>::type>::type type;
+};
+
+#if EIGEN_MAX_STATIC_ALIGN_BYTES>0
+template<int ArrayBytes, int AlignmentBytes,
+         bool Match     =  bool((ArrayBytes%AlignmentBytes)==0),
+         bool TryHalf   =  bool(EIGEN_MIN_ALIGN_BYTES<AlignmentBytes) >
+struct compute_default_alignment_helper
+{
+  enum { value = 0 };
+};
+
+template<int ArrayBytes, int AlignmentBytes, bool TryHalf>
+struct compute_default_alignment_helper<ArrayBytes, AlignmentBytes, true, TryHalf> // Match
+{
+  enum { value = AlignmentBytes };
+};
+
+template<int ArrayBytes, int AlignmentBytes>
+struct compute_default_alignment_helper<ArrayBytes, AlignmentBytes, false, true> // Try-half
+{
+  // current packet too large, try with an half-packet
+  enum { value = compute_default_alignment_helper<ArrayBytes, AlignmentBytes/2>::value };
+};
+#else
+// If static alignment is disabled, no need to bother.
+// This also avoids a division by zero in "bool Match =  bool((ArrayBytes%AlignmentBytes)==0)"
+template<int ArrayBytes, int AlignmentBytes>
+struct compute_default_alignment_helper
+{
+  enum { value = 0 };
+};
+#endif
+
+template<typename T, int Size> struct compute_default_alignment {
+  enum { value = compute_default_alignment_helper<Size*sizeof(T),EIGEN_MAX_STATIC_ALIGN_BYTES>::value };
+};
+
+template<typename T> struct compute_default_alignment<T,Dynamic> {
+  enum { value = EIGEN_MAX_ALIGN_BYTES };
 };
 
 template<typename _Scalar, int _Rows, int _Cols,
@@ -127,35 +250,12 @@
 template<typename Scalar, int Rows, int Cols, int Options, int MaxRows, int MaxCols>
 class compute_matrix_flags
 {
-    enum {
-      row_major_bit = Options&RowMajor ? RowMajorBit : 0,
-      is_dynamic_size_storage = MaxRows==Dynamic || MaxCols==Dynamic,
-
-      aligned_bit =
-      (
-            ((Options&DontAlign)==0)
-        && (
-#if EIGEN_ALIGN_STATICALLY
-             ((!is_dynamic_size_storage) && (((MaxCols*MaxRows*int(sizeof(Scalar))) % 16) == 0))
-#else
-             0
-#endif
-
-          ||
-
-#if EIGEN_ALIGN
-             is_dynamic_size_storage
-#else
-             0
-#endif
-
-          )
-      ) ? AlignedBit : 0,
-      packet_access_bit = packet_traits<Scalar>::Vectorizable && aligned_bit ? PacketAccessBit : 0
-    };
-
+    enum { row_major_bit = Options&RowMajor ? RowMajorBit : 0 };
   public:
-    enum { ret = LinearAccessBit | LvalueBit | DirectAccessBit | NestByRefBit | packet_access_bit | row_major_bit | aligned_bit };
+    // FIXME currently we still have to handle DirectAccessBit at the expression level to handle DenseCoeffsBase<>
+    // and then propagate this information to the evaluator's flags.
+    // However, I (Gael) think that DirectAccessBit should only matter at the evaluation stage.
+    enum { ret = DirectAccessBit | LvalueBit | NestByRefBit | row_major_bit };
 };
 
 template<int _Rows, int _Cols> struct size_at_compile_time
@@ -163,34 +263,43 @@
   enum { ret = (_Rows==Dynamic || _Cols==Dynamic) ? Dynamic : _Rows * _Cols };
 };
 
+template<typename XprType> struct size_of_xpr_at_compile_time
+{
+  enum { ret = size_at_compile_time<traits<XprType>::RowsAtCompileTime,traits<XprType>::ColsAtCompileTime>::ret };
+};
+
 /* plain_matrix_type : the difference from eval is that plain_matrix_type is always a plain matrix type,
  * whereas eval is a const reference in the case of a matrix
  */
 
 template<typename T, typename StorageKind = typename traits<T>::StorageKind> struct plain_matrix_type;
-template<typename T, typename BaseClassType> struct plain_matrix_type_dense;
+template<typename T, typename BaseClassType, int Flags> struct plain_matrix_type_dense;
 template<typename T> struct plain_matrix_type<T,Dense>
 {
-  typedef typename plain_matrix_type_dense<T,typename traits<T>::XprKind>::type type;
+  typedef typename plain_matrix_type_dense<T,typename traits<T>::XprKind, traits<T>::Flags>::type type;
+};
+template<typename T> struct plain_matrix_type<T,DiagonalShape>
+{
+  typedef typename T::PlainObject type;
 };
 
-template<typename T> struct plain_matrix_type_dense<T,MatrixXpr>
+template<typename T, int Flags> struct plain_matrix_type_dense<T,MatrixXpr,Flags>
 {
   typedef Matrix<typename traits<T>::Scalar,
                 traits<T>::RowsAtCompileTime,
                 traits<T>::ColsAtCompileTime,
-                AutoAlign | (traits<T>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                AutoAlign | (Flags&RowMajorBit ? RowMajor : ColMajor),
                 traits<T>::MaxRowsAtCompileTime,
                 traits<T>::MaxColsAtCompileTime
           > type;
 };
 
-template<typename T> struct plain_matrix_type_dense<T,ArrayXpr>
+template<typename T, int Flags> struct plain_matrix_type_dense<T,ArrayXpr,Flags>
 {
   typedef Array<typename traits<T>::Scalar,
                 traits<T>::RowsAtCompileTime,
                 traits<T>::ColsAtCompileTime,
-                AutoAlign | (traits<T>::Flags&RowMajorBit ? RowMajor : ColMajor),
+                AutoAlign | (Flags&RowMajorBit ? RowMajor : ColMajor),
                 traits<T>::MaxRowsAtCompileTime,
                 traits<T>::MaxColsAtCompileTime
           > type;
@@ -215,6 +324,11 @@
 //           > type;
 };
 
+template<typename T> struct eval<T,DiagonalShape>
+{
+  typedef typename plain_matrix_type<T>::type type;
+};
+
 // for matrices, no need to evaluate, just use a const reference to avoid a useless copy
 template<typename _Scalar, int _Rows, int _Cols, int _Options, int _MaxRows, int _MaxCols>
 struct eval<Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>, Dense>
@@ -229,6 +343,15 @@
 };
 
 
+/* similar to plain_matrix_type, but using the evaluator's Flags */
+template<typename T, typename StorageKind = typename traits<T>::StorageKind> struct plain_object_eval;
+
+template<typename T>
+struct plain_object_eval<T,Dense>
+{
+  typedef typename plain_matrix_type_dense<T,typename traits<T>::XprKind, evaluator<T>::Flags>::type type;
+};
+
 
 /* plain_matrix_type_column_major : same as plain_matrix_type but guaranteed to be column-major
  */
@@ -266,9 +389,6 @@
           > type;
 };
 
-// we should be able to get rid of this one too
-template<typename T> struct must_nest_by_value { enum { ret = false }; };
-
 /** \internal The reference selector for template expressions. The idea is that we don't
   * need to use references for expressions since they are light weight proxy
   * objects which should generate no copying overhead. */
@@ -280,6 +400,12 @@
     T const&,
     const T
   >::type type;
+  
+  typedef typename conditional<
+    bool(traits<T>::Flags & NestByRefBit),
+    T &,
+    T
+  >::type non_const_type;
 };
 
 /** \internal Adds the const qualifier on the value-type of T2 if and only if T1 is a const type */
@@ -293,54 +419,41 @@
   >::type type;
 };
 
-/** \internal Determines how a given expression should be nested into another one.
+
+// However, we still need a mechanism to detect whether an expression which is evaluated multiple time
+// has to be evaluated into a temporary.
+// That's the purpose of this new nested_eval helper:
+/** \internal Determines how a given expression should be nested when evaluated multiple times.
   * For example, when you do a * (b+c), Eigen will determine how the expression b+c should be
-  * nested into the bigger product expression. The choice is between nesting the expression b+c as-is, or
+  * evaluated into the bigger product expression. The choice is between nesting the expression b+c as-is, or
   * evaluating that expression b+c into a temporary variable d, and nest d so that the resulting expression is
   * a*d. Evaluating can be beneficial for example if every coefficient access in the resulting expression causes
   * many coefficient accesses in the nested expressions -- as is the case with matrix product for example.
   *
-  * \param T the type of the expression being nested
-  * \param n the number of coefficient accesses in the nested expression for each coefficient access in the bigger expression.
-  *
-  * Note that if no evaluation occur, then the constness of T is preserved.
-  *
-  * Example. Suppose that a, b, and c are of type Matrix3d. The user forms the expression a*(b+c).
-  * b+c is an expression "sum of matrices", which we will denote by S. In order to determine how to nest it,
-  * the Product expression uses: nested<S, 3>::ret, which turns out to be Matrix3d because the internal logic of
-  * nested determined that in this case it was better to evaluate the expression b+c into a temporary. On the other hand,
-  * since a is of type Matrix3d, the Product expression nests it as nested<Matrix3d, 3>::ret, which turns out to be
-  * const Matrix3d&, because the internal logic of nested determined that since a was already a matrix, there was no point
-  * in copying it into another matrix.
+  * \tparam T the type of the expression being nested.
+  * \tparam n the number of coefficient accesses in the nested expression for each coefficient access in the bigger expression.
+  * \tparam PlainObject the type of the temporary if needed.
   */
-template<typename T, int n=1, typename PlainObject = typename eval<T>::type> struct nested
+template<typename T, int n, typename PlainObject = typename plain_object_eval<T>::type> struct nested_eval
 {
   enum {
-    // for the purpose of this test, to keep it reasonably simple, we arbitrarily choose a value of Dynamic values.
-    // the choice of 10000 makes it larger than any practical fixed value and even most dynamic values.
-    // in extreme cases where these assumptions would be wrong, we would still at worst suffer performance issues
-    // (poor choice of temporaries).
-    // it's important that this value can still be squared without integer overflowing.
-    DynamicAsInteger = 10000,
     ScalarReadCost = NumTraits<typename traits<T>::Scalar>::ReadCost,
-    ScalarReadCostAsInteger = ScalarReadCost == Dynamic ? int(DynamicAsInteger) : int(ScalarReadCost),
-    CoeffReadCost = traits<T>::CoeffReadCost,
-    CoeffReadCostAsInteger = CoeffReadCost == Dynamic ? int(DynamicAsInteger) : int(CoeffReadCost),
-    NAsInteger = n == Dynamic ? int(DynamicAsInteger) : n,
-    CostEvalAsInteger   = (NAsInteger+1) * ScalarReadCostAsInteger + CoeffReadCostAsInteger,
-    CostNoEvalAsInteger = NAsInteger * CoeffReadCostAsInteger
+    CoeffReadCost = evaluator<T>::CoeffReadCost,  // NOTE What if an evaluator evaluate itself into a tempory?
+                                                  //      Then CoeffReadCost will be small (e.g., 1) but we still have to evaluate, especially if n>1.
+                                                  //      This situation is already taken care by the EvalBeforeNestingBit flag, which is turned ON
+                                                  //      for all evaluator creating a temporary. This flag is then propagated by the parent evaluators.
+                                                  //      Another solution could be to count the number of temps?
+    NAsInteger = n == Dynamic ? HugeCost : n,
+    CostEval   = (NAsInteger+1) * ScalarReadCost + CoeffReadCost,
+    CostNoEval = NAsInteger * CoeffReadCost,
+    Evaluate = (int(evaluator<T>::Flags) & EvalBeforeNestingBit) || (int(CostEval) < int(CostNoEval))
   };
 
-  typedef typename conditional<
-      ( (int(traits<T>::Flags) & EvalBeforeNestingBit) ||
-        int(CostEvalAsInteger) < int(CostNoEvalAsInteger)
-      ),
-      PlainObject,
-      typename ref_selector<T>::type
-  >::type type;
+  typedef typename conditional<Evaluate, PlainObject, typename ref_selector<T>::type>::type type;
 };
 
 template<typename T>
+EIGEN_DEVICE_FUNC
 inline T* const_cast_ptr(const T* ptr)
 {
   return const_cast<T*>(ptr);
@@ -364,30 +477,13 @@
   typedef ArrayBase<Derived> type;
 };
 
-/** \internal Helper base class to add a scalar multiple operator
-  * overloads for complex types */
-template<typename Derived,typename Scalar,typename OtherScalar,
-         bool EnableIt = !is_same<Scalar,OtherScalar>::value >
-struct special_scalar_op_base : public DenseCoeffsBase<Derived>
-{
-  // dummy operator* so that the
-  // "using special_scalar_op_base::operator*" compiles
-  void operator*() const;
-};
+template<typename Derived, typename XprKind = typename traits<Derived>::XprKind, typename StorageKind = typename traits<Derived>::StorageKind>
+struct generic_xpr_base;
 
-template<typename Derived,typename Scalar,typename OtherScalar>
-struct special_scalar_op_base<Derived,Scalar,OtherScalar,true>  : public DenseCoeffsBase<Derived>
+template<typename Derived, typename XprKind>
+struct generic_xpr_base<Derived, XprKind, Dense>
 {
-  const CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
-  operator*(const OtherScalar& scalar) const
-  {
-    return CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
-      (*static_cast<const Derived*>(this), scalar_multiple2_op<Scalar,OtherScalar>(scalar));
-  }
-
-  inline friend const CwiseUnaryOp<scalar_multiple2_op<Scalar,OtherScalar>, Derived>
-  operator*(const OtherScalar& scalar, const Derived& matrix)
-  { return static_cast<const special_scalar_op_base&>(matrix).operator*(scalar); }
+  typedef typename dense_xpr_base<Derived,XprKind>::type type;
 };
 
 template<typename XprType, typename CastType> struct cast_return_type
@@ -405,9 +501,79 @@
 {
   typedef A ret;
 };
+template <typename A> struct promote_storage_type<A, const A>
+{
+  typedef A ret;
+};
+template <typename A> struct promote_storage_type<const A, A>
+{
+  typedef A ret;
+};
+
+/** \internal Specify the "storage kind" of applying a coefficient-wise
+  * binary operations between two expressions of kinds A and B respectively.
+  * The template parameter Functor permits to specialize the resulting storage kind wrt to
+  * the functor.
+  * The default rules are as follows:
+  * \code
+  * A      op A      -> A
+  * A      op dense  -> dense
+  * dense  op B      -> dense
+  * sparse op dense  -> sparse
+  * dense  op sparse -> sparse
+  * \endcode
+  */
+template <typename A, typename B, typename Functor> struct cwise_promote_storage_type;
+
+template <typename A, typename Functor>                   struct cwise_promote_storage_type<A,A,Functor>                                      { typedef A      ret; };
+template <typename Functor>                               struct cwise_promote_storage_type<Dense,Dense,Functor>                              { typedef Dense  ret; };
+template <typename A, typename Functor>                   struct cwise_promote_storage_type<A,Dense,Functor>                                  { typedef Dense  ret; };
+template <typename B, typename Functor>                   struct cwise_promote_storage_type<Dense,B,Functor>                                  { typedef Dense  ret; };
+template <typename Functor>                               struct cwise_promote_storage_type<Sparse,Dense,Functor>                             { typedef Sparse ret; };
+template <typename Functor>                               struct cwise_promote_storage_type<Dense,Sparse,Functor>                             { typedef Sparse ret; };
+
+template <typename LhsKind, typename RhsKind, int LhsOrder, int RhsOrder> struct cwise_promote_storage_order {
+  enum { value = LhsOrder };
+};
+
+template <typename LhsKind, int LhsOrder, int RhsOrder>   struct cwise_promote_storage_order<LhsKind,Sparse,LhsOrder,RhsOrder>                { enum { value = RhsOrder }; };
+template <typename RhsKind, int LhsOrder, int RhsOrder>   struct cwise_promote_storage_order<Sparse,RhsKind,LhsOrder,RhsOrder>                { enum { value = LhsOrder }; };
+template <int Order>                                      struct cwise_promote_storage_order<Sparse,Sparse,Order,Order>                       { enum { value = Order }; };
+
+
+/** \internal Specify the "storage kind" of multiplying an expression of kind A with kind B.
+  * The template parameter ProductTag permits to specialize the resulting storage kind wrt to
+  * some compile-time properties of the product: GemmProduct, GemvProduct, OuterProduct, InnerProduct.
+  * The default rules are as follows:
+  * \code
+  *  K * K            -> K
+  *  dense * K        -> dense
+  *  K * dense        -> dense
+  *  diag * K         -> K
+  *  K * diag         -> K
+  *  Perm * K         -> K
+  * K * Perm          -> K
+  * \endcode
+  */
+template <typename A, typename B, int ProductTag> struct product_promote_storage_type;
+
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  A,                  ProductTag> { typedef A     ret;};
+template <int ProductTag>             struct product_promote_storage_type<Dense,              Dense,              ProductTag> { typedef Dense ret;};
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  Dense,              ProductTag> { typedef Dense ret; };
+template <typename B, int ProductTag> struct product_promote_storage_type<Dense,              B,                  ProductTag> { typedef Dense ret; };
+
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  DiagonalShape,      ProductTag> { typedef A ret; };
+template <typename B, int ProductTag> struct product_promote_storage_type<DiagonalShape,      B,                  ProductTag> { typedef B ret; };
+template <int ProductTag>             struct product_promote_storage_type<Dense,              DiagonalShape,      ProductTag> { typedef Dense ret; };
+template <int ProductTag>             struct product_promote_storage_type<DiagonalShape,      Dense,              ProductTag> { typedef Dense ret; };
+
+template <typename A, int ProductTag> struct product_promote_storage_type<A,                  PermutationStorage, ProductTag> { typedef A ret; };
+template <typename B, int ProductTag> struct product_promote_storage_type<PermutationStorage, B,                  ProductTag> { typedef B ret; };
+template <int ProductTag>             struct product_promote_storage_type<Dense,              PermutationStorage, ProductTag> { typedef Dense ret; };
+template <int ProductTag>             struct product_promote_storage_type<PermutationStorage, Dense,              ProductTag> { typedef Dense ret; };
 
 /** \internal gives the plain matrix or array type to store a row/column/diagonal of a matrix type.
-  * \param Scalar optional parameter allowing to pass a different scalar type than the one of the MatrixType.
+  * \tparam Scalar optional parameter allowing to pass a different scalar type than the one of the MatrixType.
   */
 template<typename ExpressionType, typename Scalar = typename ExpressionType::Scalar>
 struct plain_row_type
@@ -455,15 +621,201 @@
   >::type type;
 };
 
+template<typename Expr,typename Scalar = typename Expr::Scalar>
+struct plain_constant_type
+{
+  enum { Options = (traits<Expr>::Flags&RowMajorBit)?RowMajor:0 };
+
+  typedef Array<Scalar,  traits<Expr>::RowsAtCompileTime,   traits<Expr>::ColsAtCompileTime,
+                Options, traits<Expr>::MaxRowsAtCompileTime,traits<Expr>::MaxColsAtCompileTime> array_type;
+
+  typedef Matrix<Scalar,  traits<Expr>::RowsAtCompileTime,   traits<Expr>::ColsAtCompileTime,
+                 Options, traits<Expr>::MaxRowsAtCompileTime,traits<Expr>::MaxColsAtCompileTime> matrix_type;
+
+  typedef CwiseNullaryOp<scalar_constant_op<Scalar>, const typename conditional<is_same< typename traits<Expr>::XprKind, MatrixXpr >::value, matrix_type, array_type>::type > type;
+};
+
 template<typename ExpressionType>
 struct is_lvalue
 {
-  enum { value = !bool(is_const<ExpressionType>::value) &&
+  enum { value = (!bool(is_const<ExpressionType>::value)) &&
                  bool(traits<ExpressionType>::Flags & LvalueBit) };
 };
 
+template<typename T> struct is_diagonal
+{ enum { ret = false }; };
+
+template<typename T> struct is_diagonal<DiagonalBase<T> >
+{ enum { ret = true }; };
+
+template<typename T> struct is_diagonal<DiagonalWrapper<T> >
+{ enum { ret = true }; };
+
+template<typename T, int S> struct is_diagonal<DiagonalMatrix<T,S> >
+{ enum { ret = true }; };
+
+template<typename S1, typename S2> struct glue_shapes;
+template<> struct glue_shapes<DenseShape,TriangularShape> { typedef TriangularShape type;  };
+
+template<typename T1, typename T2>
+bool is_same_dense(const T1 &mat1, const T2 &mat2, typename enable_if<has_direct_access<T1>::ret&&has_direct_access<T2>::ret, T1>::type * = 0)
+{
+  return (mat1.data()==mat2.data()) && (mat1.innerStride()==mat2.innerStride()) && (mat1.outerStride()==mat2.outerStride());
+}
+
+template<typename T1, typename T2>
+bool is_same_dense(const T1 &, const T2 &, typename enable_if<!(has_direct_access<T1>::ret&&has_direct_access<T2>::ret), T1>::type * = 0)
+{
+  return false;
+}
+
+// Internal helper defining the cost of a scalar division for the type T.
+// The default heuristic can be specialized for each scalar type and architecture.
+template<typename T,bool Vectorized=false,typename EnaleIf = void>
+struct scalar_div_cost {
+  enum { value = 8*NumTraits<T>::MulCost };
+};
+
+template<typename T,bool Vectorized>
+struct scalar_div_cost<std::complex<T>, Vectorized> {
+  enum { value = 2*scalar_div_cost<T>::value
+               + 6*NumTraits<T>::MulCost
+               + 3*NumTraits<T>::AddCost
+  };
+};
+
+
+template<bool Vectorized>
+struct scalar_div_cost<signed long,Vectorized,typename conditional<sizeof(long)==8,void,false_type>::type> { enum { value = 24 }; };
+template<bool Vectorized>
+struct scalar_div_cost<unsigned long,Vectorized,typename conditional<sizeof(long)==8,void,false_type>::type> { enum { value = 21 }; };
+
+
+#ifdef EIGEN_DEBUG_ASSIGN
+std::string demangle_traversal(int t)
+{
+  if(t==DefaultTraversal) return "DefaultTraversal";
+  if(t==LinearTraversal) return "LinearTraversal";
+  if(t==InnerVectorizedTraversal) return "InnerVectorizedTraversal";
+  if(t==LinearVectorizedTraversal) return "LinearVectorizedTraversal";
+  if(t==SliceVectorizedTraversal) return "SliceVectorizedTraversal";
+  return "?";
+}
+std::string demangle_unrolling(int t)
+{
+  if(t==NoUnrolling) return "NoUnrolling";
+  if(t==InnerUnrolling) return "InnerUnrolling";
+  if(t==CompleteUnrolling) return "CompleteUnrolling";
+  return "?";
+}
+std::string demangle_flags(int f)
+{
+  std::string res;
+  if(f&RowMajorBit)                 res += " | RowMajor";
+  if(f&PacketAccessBit)             res += " | Packet";
+  if(f&LinearAccessBit)             res += " | Linear";
+  if(f&LvalueBit)                   res += " | Lvalue";
+  if(f&DirectAccessBit)             res += " | Direct";
+  if(f&NestByRefBit)                res += " | NestByRef";
+  if(f&NoPreferredStorageOrderBit)  res += " | NoPreferredStorageOrderBit";
+  
+  return res;
+}
+#endif
+
 } // end namespace internal
 
+
+/** \class ScalarBinaryOpTraits
+  * \ingroup Core_Module
+  *
+  * \brief Determines whether the given binary operation of two numeric types is allowed and what the scalar return type is.
+  *
+  * This class permits to control the scalar return type of any binary operation performed on two different scalar types through (partial) template specializations.
+  *
+  * For instance, let \c U1, \c U2 and \c U3 be three user defined scalar types for which most operations between instances of \c U1 and \c U2 returns an \c U3.
+  * You can let %Eigen knows that by defining:
+    \code
+    template<typename BinaryOp>
+    struct ScalarBinaryOpTraits<U1,U2,BinaryOp> { typedef U3 ReturnType;  };
+    template<typename BinaryOp>
+    struct ScalarBinaryOpTraits<U2,U1,BinaryOp> { typedef U3 ReturnType;  };
+    \endcode
+  * You can then explicitly disable some particular operations to get more explicit error messages:
+    \code
+    template<>
+    struct ScalarBinaryOpTraits<U1,U2,internal::scalar_max_op<U1,U2> > {};
+    \endcode
+  * Or customize the return type for individual operation:
+    \code
+    template<>
+    struct ScalarBinaryOpTraits<U1,U2,internal::scalar_sum_op<U1,U2> > { typedef U1 ReturnType; };
+    \endcode
+  *
+  * By default, the following generic combinations are supported:
+  <table class="manual">
+  <tr><th>ScalarA</th><th>ScalarB</th><th>BinaryOp</th><th>ReturnType</th><th>Note</th></tr>
+  <tr            ><td>\c T </td><td>\c T </td><td>\c * </td><td>\c T </td><td></td></tr>
+  <tr class="alt"><td>\c NumTraits<T>::Real </td><td>\c T </td><td>\c * </td><td>\c T </td><td>Only if \c NumTraits<T>::IsComplex </td></tr>
+  <tr            ><td>\c T </td><td>\c NumTraits<T>::Real </td><td>\c * </td><td>\c T </td><td>Only if \c NumTraits<T>::IsComplex </td></tr>
+  </table>
+  *
+  * \sa CwiseBinaryOp
+  */
+template<typename ScalarA, typename ScalarB, typename BinaryOp=internal::scalar_product_op<ScalarA,ScalarB> >
+struct ScalarBinaryOpTraits
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  // for backward compatibility, use the hints given by the (deprecated) internal::scalar_product_traits class.
+  : internal::scalar_product_traits<ScalarA,ScalarB>
+#endif // EIGEN_PARSED_BY_DOXYGEN
+{};
+
+template<typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<T,T,BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+template <typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<T, typename NumTraits<typename internal::enable_if<NumTraits<T>::IsComplex,T>::type>::Real, BinaryOp>
+{
+  typedef T ReturnType;
+};
+template <typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<typename NumTraits<typename internal::enable_if<NumTraits<T>::IsComplex,T>::type>::Real, T, BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+// For Matrix * Permutation
+template<typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<T,void,BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+// For Permutation * Matrix
+template<typename T, typename BinaryOp>
+struct ScalarBinaryOpTraits<void,T,BinaryOp>
+{
+  typedef T ReturnType;
+};
+
+// for Permutation*Permutation
+template<typename BinaryOp>
+struct ScalarBinaryOpTraits<void,void,BinaryOp>
+{
+  typedef void ReturnType;
+};
+
+// We require Lhs and Rhs to have "compatible" scalar types.
+// It is tempting to always allow mixing different types but remember that this is often impossible in the vectorized paths.
+// So allowing mixing different types gives very unexpected errors when enabling vectorization, when the user tries to
+// add together a float matrix and a double matrix.
+#define EIGEN_CHECK_BINARY_COMPATIBILIY(BINOP,LHS,RHS) \
+  EIGEN_STATIC_ASSERT((Eigen::internal::has_ReturnType<ScalarBinaryOpTraits<LHS, RHS,BINOP> >::value), \
+    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+    
 } // end namespace Eigen
 
 #endif // EIGEN_XPRHELPER_H
diff --git a/Eigen/src/Eigen2Support/Block.h b/Eigen/src/Eigen2Support/Block.h
deleted file mode 100644
index 604456f..0000000
--- a/Eigen/src/Eigen2Support/Block.h
+++ /dev/null
@@ -1,126 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_BLOCK2_H
-#define EIGEN_BLOCK2_H
-
-namespace Eigen { 
-
-/** \returns a dynamic-size expression of a corner of *this.
-  *
-  * \param type the type of corner. Can be \a Eigen::TopLeft, \a Eigen::TopRight,
-  * \a Eigen::BottomLeft, \a Eigen::BottomRight.
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_corner_enum_int_int.cpp
-  * Output: \verbinclude MatrixBase_corner_enum_int_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size matrix, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-template<typename Derived>
-inline Block<Derived> DenseBase<Derived>
-  ::corner(CornerType type, Index cRows, Index cCols)
-{
-  switch(type)
-  {
-    default:
-      eigen_assert(false && "Bad corner type.");
-    case TopLeft:
-      return Block<Derived>(derived(), 0, 0, cRows, cCols);
-    case TopRight:
-      return Block<Derived>(derived(), 0, cols() - cCols, cRows, cCols);
-    case BottomLeft:
-      return Block<Derived>(derived(), rows() - cRows, 0, cRows, cCols);
-    case BottomRight:
-      return Block<Derived>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
-  }
-}
-
-/** This is the const version of corner(CornerType, Index, Index).*/
-template<typename Derived>
-inline const Block<Derived>
-DenseBase<Derived>::corner(CornerType type, Index cRows, Index cCols) const
-{
-  switch(type)
-  {
-    default:
-      eigen_assert(false && "Bad corner type.");
-    case TopLeft:
-      return Block<Derived>(derived(), 0, 0, cRows, cCols);
-    case TopRight:
-      return Block<Derived>(derived(), 0, cols() - cCols, cRows, cCols);
-    case BottomLeft:
-      return Block<Derived>(derived(), rows() - cRows, 0, cRows, cCols);
-    case BottomRight:
-      return Block<Derived>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
-  }
-}
-
-/** \returns a fixed-size expression of a corner of *this.
-  *
-  * \param type the type of corner. Can be \a Eigen::TopLeft, \a Eigen::TopRight,
-  * \a Eigen::BottomLeft, \a Eigen::BottomRight.
-  *
-  * The template parameters CRows and CCols arethe number of rows and columns in the corner.
-  *
-  * Example: \include MatrixBase_template_int_int_corner_enum.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_corner_enum.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-template<typename Derived>
-template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols>
-DenseBase<Derived>::corner(CornerType type)
-{
-  switch(type)
-  {
-    default:
-      eigen_assert(false && "Bad corner type.");
-    case TopLeft:
-      return Block<Derived, CRows, CCols>(derived(), 0, 0);
-    case TopRight:
-      return Block<Derived, CRows, CCols>(derived(), 0, cols() - CCols);
-    case BottomLeft:
-      return Block<Derived, CRows, CCols>(derived(), rows() - CRows, 0);
-    case BottomRight:
-      return Block<Derived, CRows, CCols>(derived(), rows() - CRows, cols() - CCols);
-  }
-}
-
-/** This is the const version of corner<int, int>(CornerType).*/
-template<typename Derived>
-template<int CRows, int CCols>
-inline const Block<Derived, CRows, CCols>
-DenseBase<Derived>::corner(CornerType type) const
-{
-  switch(type)
-  {
-    default:
-      eigen_assert(false && "Bad corner type.");
-    case TopLeft:
-      return Block<Derived, CRows, CCols>(derived(), 0, 0);
-    case TopRight:
-      return Block<Derived, CRows, CCols>(derived(), 0, cols() - CCols);
-    case BottomLeft:
-      return Block<Derived, CRows, CCols>(derived(), rows() - CRows, 0);
-    case BottomRight:
-      return Block<Derived, CRows, CCols>(derived(), rows() - CRows, cols() - CCols);
-  }
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_BLOCK2_H
diff --git a/Eigen/src/Eigen2Support/CMakeLists.txt b/Eigen/src/Eigen2Support/CMakeLists.txt
deleted file mode 100644
index 7ae41b3..0000000
--- a/Eigen/src/Eigen2Support/CMakeLists.txt
+++ /dev/null
@@ -1,8 +0,0 @@
-FILE(GLOB Eigen_Eigen2Support_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Eigen2Support_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Eigen2Support COMPONENT Devel
-  )
-
-ADD_SUBDIRECTORY(Geometry)
\ No newline at end of file
diff --git a/Eigen/src/Eigen2Support/Cwise.h b/Eigen/src/Eigen2Support/Cwise.h
deleted file mode 100644
index d95009b..0000000
--- a/Eigen/src/Eigen2Support/Cwise.h
+++ /dev/null
@@ -1,192 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_CWISE_H
-#define EIGEN_CWISE_H
-
-namespace Eigen { 
-
-/** \internal
-  * convenient macro to defined the return type of a cwise binary operation */
-#define EIGEN_CWISE_BINOP_RETURN_TYPE(OP) \
-    CwiseBinaryOp<OP<typename internal::traits<ExpressionType>::Scalar>, ExpressionType, OtherDerived>
-
-/** \internal
-  * convenient macro to defined the return type of a cwise unary operation */
-#define EIGEN_CWISE_UNOP_RETURN_TYPE(OP) \
-    CwiseUnaryOp<OP<typename internal::traits<ExpressionType>::Scalar>, ExpressionType>
-
-/** \internal
-  * convenient macro to defined the return type of a cwise comparison to a scalar */
-#define EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(OP) \
-    CwiseBinaryOp<OP<typename internal::traits<ExpressionType>::Scalar>, ExpressionType, \
-        typename ExpressionType::ConstantReturnType >
-
-/** \class Cwise
-  *
-  * \brief Pseudo expression providing additional coefficient-wise operations
-  *
-  * \param ExpressionType the type of the object on which to do coefficient-wise operations
-  *
-  * This class represents an expression with additional coefficient-wise features.
-  * It is the return type of MatrixBase::cwise()
-  * and most of the time this is the only way it is used.
-  *
-  * Example: \include MatrixBase_cwise_const.cpp
-  * Output: \verbinclude MatrixBase_cwise_const.out
-  *
-  * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_CWISE_PLUGIN.
-  *
-  * \sa MatrixBase::cwise() const, MatrixBase::cwise()
-  */
-template<typename ExpressionType> class Cwise
-{
-  public:
-
-    typedef typename internal::traits<ExpressionType>::Scalar Scalar;
-    typedef typename internal::conditional<internal::must_nest_by_value<ExpressionType>::ret,
-        ExpressionType, const ExpressionType&>::type ExpressionTypeNested;
-    typedef CwiseUnaryOp<internal::scalar_add_op<Scalar>, ExpressionType> ScalarAddReturnType;
-
-    inline Cwise(const ExpressionType& matrix) : m_matrix(matrix) {}
-
-    /** \internal */
-    inline const ExpressionType& _expression() const { return m_matrix; }
-
-    template<typename OtherDerived>
-    const EIGEN_CWISE_PRODUCT_RETURN_TYPE(ExpressionType,OtherDerived)
-    operator*(const MatrixBase<OtherDerived> &other) const;
-
-    template<typename OtherDerived>
-    const EIGEN_CWISE_BINOP_RETURN_TYPE(internal::scalar_quotient_op)
-    operator/(const MatrixBase<OtherDerived> &other) const;
-
-    /** \deprecated ArrayBase::min() */
-    template<typename OtherDerived>
-    const EIGEN_CWISE_BINOP_RETURN_TYPE(internal::scalar_min_op)
-    (min)(const MatrixBase<OtherDerived> &other) const
-    { return EIGEN_CWISE_BINOP_RETURN_TYPE(internal::scalar_min_op)(_expression(), other.derived()); }
-
-    /** \deprecated ArrayBase::max() */
-    template<typename OtherDerived>
-    const EIGEN_CWISE_BINOP_RETURN_TYPE(internal::scalar_max_op)
-    (max)(const MatrixBase<OtherDerived> &other) const
-    { return EIGEN_CWISE_BINOP_RETURN_TYPE(internal::scalar_max_op)(_expression(), other.derived()); }
-
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_abs_op)      abs() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_abs2_op)     abs2() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_square_op)   square() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_cube_op)     cube() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_inverse_op)  inverse() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_sqrt_op)     sqrt() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_exp_op)      exp() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_log_op)      log() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_cos_op)      cos() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_sin_op)      sin() const;
-    const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_pow_op)      pow(const Scalar& exponent) const;
-
-    const ScalarAddReturnType
-    operator+(const Scalar& scalar) const;
-
-    /** \relates Cwise */
-    friend const ScalarAddReturnType
-    operator+(const Scalar& scalar, const Cwise& mat)
-    { return mat + scalar; }
-
-    ExpressionType& operator+=(const Scalar& scalar);
-
-    const ScalarAddReturnType
-    operator-(const Scalar& scalar) const;
-
-    ExpressionType& operator-=(const Scalar& scalar);
-
-    template<typename OtherDerived>
-    inline ExpressionType& operator*=(const MatrixBase<OtherDerived> &other);
-
-    template<typename OtherDerived>
-    inline ExpressionType& operator/=(const MatrixBase<OtherDerived> &other);
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less)
-    operator<(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)
-    operator<=(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater)
-    operator>(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater_equal)
-    operator>=(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::equal_to)
-    operator==(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> const EIGEN_CWISE_BINOP_RETURN_TYPE(std::not_equal_to)
-    operator!=(const MatrixBase<OtherDerived>& other) const;
-
-    // comparisons to a scalar value
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less)
-    operator<(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less_equal)
-    operator<=(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater)
-    operator>(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater_equal)
-    operator>=(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::equal_to)
-    operator==(Scalar s) const;
-
-    const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::not_equal_to)
-    operator!=(Scalar s) const;
-
-    // allow to extend Cwise outside Eigen
-    #ifdef EIGEN_CWISE_PLUGIN
-    #include EIGEN_CWISE_PLUGIN
-    #endif
-
-  protected:
-    ExpressionTypeNested m_matrix;
-};
-
-
-/** \returns a Cwise wrapper of *this providing additional coefficient-wise operations
-  *
-  * Example: \include MatrixBase_cwise_const.cpp
-  * Output: \verbinclude MatrixBase_cwise_const.out
-  *
-  * \sa class Cwise, cwise()
-  */
-template<typename Derived>
-inline const Cwise<Derived> MatrixBase<Derived>::cwise() const
-{
-  return derived();
-}
-
-/** \returns a Cwise wrapper of *this providing additional coefficient-wise operations
-  *
-  * Example: \include MatrixBase_cwise.cpp
-  * Output: \verbinclude MatrixBase_cwise.out
-  *
-  * \sa class Cwise, cwise() const
-  */
-template<typename Derived>
-inline Cwise<Derived> MatrixBase<Derived>::cwise()
-{
-  return derived();
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_CWISE_H
diff --git a/Eigen/src/Eigen2Support/CwiseOperators.h b/Eigen/src/Eigen2Support/CwiseOperators.h
deleted file mode 100644
index 482f306..0000000
--- a/Eigen/src/Eigen2Support/CwiseOperators.h
+++ /dev/null
@@ -1,298 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_ARRAY_CWISE_OPERATORS_H
-#define EIGEN_ARRAY_CWISE_OPERATORS_H
-
-namespace Eigen { 
-
-/***************************************************************************
-* The following functions were defined in Core
-***************************************************************************/
-
-
-/** \deprecated ArrayBase::abs() */
-template<typename ExpressionType>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_abs_op)
-Cwise<ExpressionType>::abs() const
-{
-  return _expression();
-}
-
-/** \deprecated ArrayBase::abs2() */
-template<typename ExpressionType>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_abs2_op)
-Cwise<ExpressionType>::abs2() const
-{
-  return _expression();
-}
-
-/** \deprecated ArrayBase::exp() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_exp_op)
-Cwise<ExpressionType>::exp() const
-{
-  return _expression();
-}
-
-/** \deprecated ArrayBase::log() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_log_op)
-Cwise<ExpressionType>::log() const
-{
-  return _expression();
-}
-
-/** \deprecated ArrayBase::operator*() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_PRODUCT_RETURN_TYPE(ExpressionType,OtherDerived)
-Cwise<ExpressionType>::operator*(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_PRODUCT_RETURN_TYPE(ExpressionType,OtherDerived)(_expression(), other.derived());
-}
-
-/** \deprecated ArrayBase::operator/() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_BINOP_RETURN_TYPE(internal::scalar_quotient_op)
-Cwise<ExpressionType>::operator/(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(internal::scalar_quotient_op)(_expression(), other.derived());
-}
-
-/** \deprecated ArrayBase::operator*=() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline ExpressionType& Cwise<ExpressionType>::operator*=(const MatrixBase<OtherDerived> &other)
-{
-  return m_matrix.const_cast_derived() = *this * other;
-}
-
-/** \deprecated ArrayBase::operator/=() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline ExpressionType& Cwise<ExpressionType>::operator/=(const MatrixBase<OtherDerived> &other)
-{
-  return m_matrix.const_cast_derived() = *this / other;
-}
-
-/***************************************************************************
-* The following functions were defined in Array
-***************************************************************************/
-
-// -- unary operators --
-
-/** \deprecated ArrayBase::sqrt() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_sqrt_op)
-Cwise<ExpressionType>::sqrt() const
-{
-  return _expression();
-}
-
-/** \deprecated ArrayBase::cos() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_cos_op)
-Cwise<ExpressionType>::cos() const
-{
-  return _expression();
-}
-
-
-/** \deprecated ArrayBase::sin() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_sin_op)
-Cwise<ExpressionType>::sin() const
-{
-  return _expression();
-}
-
-
-/** \deprecated ArrayBase::log() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_pow_op)
-Cwise<ExpressionType>::pow(const Scalar& exponent) const
-{
-  return EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_pow_op)(_expression(), internal::scalar_pow_op<Scalar>(exponent));
-}
-
-
-/** \deprecated ArrayBase::inverse() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_inverse_op)
-Cwise<ExpressionType>::inverse() const
-{
-  return _expression();
-}
-
-/** \deprecated ArrayBase::square() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_square_op)
-Cwise<ExpressionType>::square() const
-{
-  return _expression();
-}
-
-/** \deprecated ArrayBase::cube() */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_UNOP_RETURN_TYPE(internal::scalar_cube_op)
-Cwise<ExpressionType>::cube() const
-{
-  return _expression();
-}
-
-
-// -- binary operators --
-
-/** \deprecated ArrayBase::operator<() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less)
-Cwise<ExpressionType>::operator<(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::less)(_expression(), other.derived());
-}
-
-/** \deprecated ArrayBase::<=() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)
-Cwise<ExpressionType>::operator<=(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::less_equal)(_expression(), other.derived());
-}
-
-/** \deprecated ArrayBase::operator>() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater)
-Cwise<ExpressionType>::operator>(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater)(_expression(), other.derived());
-}
-
-/** \deprecated ArrayBase::operator>=() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater_equal)
-Cwise<ExpressionType>::operator>=(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::greater_equal)(_expression(), other.derived());
-}
-
-/** \deprecated ArrayBase::operator==() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::equal_to)
-Cwise<ExpressionType>::operator==(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::equal_to)(_expression(), other.derived());
-}
-
-/** \deprecated ArrayBase::operator!=() */
-template<typename ExpressionType>
-template<typename OtherDerived>
-inline const EIGEN_CWISE_BINOP_RETURN_TYPE(std::not_equal_to)
-Cwise<ExpressionType>::operator!=(const MatrixBase<OtherDerived> &other) const
-{
-  return EIGEN_CWISE_BINOP_RETURN_TYPE(std::not_equal_to)(_expression(), other.derived());
-}
-
-// comparisons to scalar value
-
-/** \deprecated ArrayBase::operator<(Scalar) */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less)
-Cwise<ExpressionType>::operator<(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \deprecated ArrayBase::operator<=(Scalar) */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less_equal)
-Cwise<ExpressionType>::operator<=(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::less_equal)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \deprecated ArrayBase::operator>(Scalar) */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater)
-Cwise<ExpressionType>::operator>(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \deprecated ArrayBase::operator>=(Scalar) */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater_equal)
-Cwise<ExpressionType>::operator>=(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::greater_equal)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \deprecated ArrayBase::operator==(Scalar) */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::equal_to)
-Cwise<ExpressionType>::operator==(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::equal_to)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-/** \deprecated ArrayBase::operator!=(Scalar) */
-template<typename ExpressionType>
-inline const EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::not_equal_to)
-Cwise<ExpressionType>::operator!=(Scalar s) const
-{
-  return EIGEN_CWISE_COMP_TO_SCALAR_RETURN_TYPE(std::not_equal_to)(_expression(),
-            typename ExpressionType::ConstantReturnType(_expression().rows(), _expression().cols(), s));
-}
-
-// scalar addition
-
-/** \deprecated ArrayBase::operator+(Scalar) */
-template<typename ExpressionType>
-inline const typename Cwise<ExpressionType>::ScalarAddReturnType
-Cwise<ExpressionType>::operator+(const Scalar& scalar) const
-{
-  return typename Cwise<ExpressionType>::ScalarAddReturnType(m_matrix, internal::scalar_add_op<Scalar>(scalar));
-}
-
-/** \deprecated ArrayBase::operator+=(Scalar) */
-template<typename ExpressionType>
-inline ExpressionType& Cwise<ExpressionType>::operator+=(const Scalar& scalar)
-{
-  return m_matrix.const_cast_derived() = *this + scalar;
-}
-
-/** \deprecated ArrayBase::operator-(Scalar) */
-template<typename ExpressionType>
-inline const typename Cwise<ExpressionType>::ScalarAddReturnType
-Cwise<ExpressionType>::operator-(const Scalar& scalar) const
-{
-  return *this + (-scalar);
-}
-
-/** \deprecated ArrayBase::operator-=(Scalar) */
-template<typename ExpressionType>
-inline ExpressionType& Cwise<ExpressionType>::operator-=(const Scalar& scalar)
-{
-  return m_matrix.const_cast_derived() = *this - scalar;
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_ARRAY_CWISE_OPERATORS_H
diff --git a/Eigen/src/Eigen2Support/Geometry/AlignedBox.h b/Eigen/src/Eigen2Support/Geometry/AlignedBox.h
deleted file mode 100644
index 2e4309d..0000000
--- a/Eigen/src/Eigen2Support/Geometry/AlignedBox.h
+++ /dev/null
@@ -1,159 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-/** \geometry_module \ingroup Geometry_Module
-  * \nonstableyet
-  *
-  * \class AlignedBox
-  *
-  * \brief An axis aligned box
-  *
-  * \param _Scalar the type of the scalar coefficients
-  * \param _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
-  *
-  * This class represents an axis aligned box as a pair of the minimal and maximal corners.
-  */
-template <typename _Scalar, int _AmbientDim>
-class AlignedBox
-{
-public:
-EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim==Dynamic ? Dynamic : _AmbientDim+1)
-  enum { AmbientDimAtCompileTime = _AmbientDim };
-  typedef _Scalar Scalar;
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef Matrix<Scalar,AmbientDimAtCompileTime,1> VectorType;
-
-  /** Default constructor initializing a null box. */
-  inline AlignedBox()
-  { if (AmbientDimAtCompileTime!=Dynamic) setNull(); }
-
-  /** Constructs a null box with \a _dim the dimension of the ambient space. */
-  inline explicit AlignedBox(int _dim) : m_min(_dim), m_max(_dim)
-  { setNull(); }
-
-  /** Constructs a box with extremities \a _min and \a _max. */
-  inline AlignedBox(const VectorType& _min, const VectorType& _max) : m_min(_min), m_max(_max) {}
-
-  /** Constructs a box containing a single point \a p. */
-  inline explicit AlignedBox(const VectorType& p) : m_min(p), m_max(p) {}
-
-  ~AlignedBox() {}
-
-  /** \returns the dimension in which the box holds */
-  inline int dim() const { return AmbientDimAtCompileTime==Dynamic ? m_min.size()-1 : AmbientDimAtCompileTime; }
-
-  /** \returns true if the box is null, i.e, empty. */
-  inline bool isNull() const { return (m_min.cwise() > m_max).any(); }
-
-  /** Makes \c *this a null/empty box. */
-  inline void setNull()
-  {
-    m_min.setConstant( (std::numeric_limits<Scalar>::max)());
-    m_max.setConstant(-(std::numeric_limits<Scalar>::max)());
-  }
-
-  /** \returns the minimal corner */
-  inline const VectorType& (min)() const { return m_min; }
-  /** \returns a non const reference to the minimal corner */
-  inline VectorType& (min)() { return m_min; }
-  /** \returns the maximal corner */
-  inline const VectorType& (max)() const { return m_max; }
-  /** \returns a non const reference to the maximal corner */
-  inline VectorType& (max)() { return m_max; }
-
-  /** \returns true if the point \a p is inside the box \c *this. */
-  inline bool contains(const VectorType& p) const
-  { return (m_min.cwise()<=p).all() && (p.cwise()<=m_max).all(); }
-
-  /** \returns true if the box \a b is entirely inside the box \c *this. */
-  inline bool contains(const AlignedBox& b) const
-  { return (m_min.cwise()<=(b.min)()).all() && ((b.max)().cwise()<=m_max).all(); }
-
-  /** Extends \c *this such that it contains the point \a p and returns a reference to \c *this. */
-  inline AlignedBox& extend(const VectorType& p)
-  { m_min = (m_min.cwise().min)(p); m_max = (m_max.cwise().max)(p); return *this; }
-
-  /** Extends \c *this such that it contains the box \a b and returns a reference to \c *this. */
-  inline AlignedBox& extend(const AlignedBox& b)
-  { m_min = (m_min.cwise().min)(b.m_min); m_max = (m_max.cwise().max)(b.m_max); return *this; }
-
-  /** Clamps \c *this by the box \a b and returns a reference to \c *this. */
-  inline AlignedBox& clamp(const AlignedBox& b)
-  { m_min = (m_min.cwise().max)(b.m_min); m_max = (m_max.cwise().min)(b.m_max); return *this; }
-
-  /** Translate \c *this by the vector \a t and returns a reference to \c *this. */
-  inline AlignedBox& translate(const VectorType& t)
-  { m_min += t; m_max += t; return *this; }
-
-  /** \returns the squared distance between the point \a p and the box \c *this,
-    * and zero if \a p is inside the box.
-    * \sa exteriorDistance()
-    */
-  inline Scalar squaredExteriorDistance(const VectorType& p) const;
-
-  /** \returns the distance between the point \a p and the box \c *this,
-    * and zero if \a p is inside the box.
-    * \sa squaredExteriorDistance()
-    */
-  inline Scalar exteriorDistance(const VectorType& p) const
-  { return ei_sqrt(squaredExteriorDistance(p)); }
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<AlignedBox,
-           AlignedBox<NewScalarType,AmbientDimAtCompileTime> >::type cast() const
-  {
-    return typename internal::cast_return_type<AlignedBox,
-                    AlignedBox<NewScalarType,AmbientDimAtCompileTime> >::type(*this);
-  }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit AlignedBox(const AlignedBox<OtherScalarType,AmbientDimAtCompileTime>& other)
-  {
-    m_min = (other.min)().template cast<Scalar>();
-    m_max = (other.max)().template cast<Scalar>();
-  }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const AlignedBox& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_min.isApprox(other.m_min, prec) && m_max.isApprox(other.m_max, prec); }
-
-protected:
-
-  VectorType m_min, m_max;
-};
-
-template<typename Scalar,int AmbiantDim>
-inline Scalar AlignedBox<Scalar,AmbiantDim>::squaredExteriorDistance(const VectorType& p) const
-{
-  Scalar dist2(0);
-  Scalar aux;
-  for (int k=0; k<dim(); ++k)
-  {
-    if ((aux = (p[k]-m_min[k]))<Scalar(0))
-      dist2 += aux*aux;
-    else if ( (aux = (m_max[k]-p[k]))<Scalar(0))
-      dist2 += aux*aux;
-  }
-  return dist2;
-}
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/All.h b/Eigen/src/Eigen2Support/Geometry/All.h
deleted file mode 100644
index e0b00fc..0000000
--- a/Eigen/src/Eigen2Support/Geometry/All.h
+++ /dev/null
@@ -1,115 +0,0 @@
-#ifndef EIGEN2_GEOMETRY_MODULE_H
-#define EIGEN2_GEOMETRY_MODULE_H
-
-#include <limits>
-
-#ifndef M_PI
-#define M_PI 3.14159265358979323846
-#endif
-
-#if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS
-#include "RotationBase.h"
-#include "Rotation2D.h"
-#include "Quaternion.h"
-#include "AngleAxis.h"
-#include "Transform.h"
-#include "Translation.h"
-#include "Scaling.h"
-#include "AlignedBox.h"
-#include "Hyperplane.h"
-#include "ParametrizedLine.h"
-#endif
-
-
-#define RotationBase eigen2_RotationBase
-#define Rotation2D eigen2_Rotation2D
-#define Rotation2Df eigen2_Rotation2Df
-#define Rotation2Dd eigen2_Rotation2Dd
-
-#define Quaternion  eigen2_Quaternion
-#define Quaternionf eigen2_Quaternionf
-#define Quaterniond eigen2_Quaterniond
-
-#define AngleAxis eigen2_AngleAxis
-#define AngleAxisf eigen2_AngleAxisf
-#define AngleAxisd eigen2_AngleAxisd
-
-#define Transform   eigen2_Transform
-#define Transform2f eigen2_Transform2f
-#define Transform2d eigen2_Transform2d
-#define Transform3f eigen2_Transform3f
-#define Transform3d eigen2_Transform3d
-
-#define Translation eigen2_Translation
-#define Translation2f eigen2_Translation2f
-#define Translation2d eigen2_Translation2d
-#define Translation3f eigen2_Translation3f
-#define Translation3d eigen2_Translation3d
-
-#define Scaling eigen2_Scaling
-#define Scaling2f eigen2_Scaling2f
-#define Scaling2d eigen2_Scaling2d
-#define Scaling3f eigen2_Scaling3f
-#define Scaling3d eigen2_Scaling3d
-
-#define AlignedBox eigen2_AlignedBox
-
-#define Hyperplane eigen2_Hyperplane
-#define ParametrizedLine eigen2_ParametrizedLine
-
-#define ei_toRotationMatrix eigen2_ei_toRotationMatrix
-#define ei_quaternion_assign_impl eigen2_ei_quaternion_assign_impl
-#define ei_transform_product_impl eigen2_ei_transform_product_impl
-
-#include "RotationBase.h"
-#include "Rotation2D.h"
-#include "Quaternion.h"
-#include "AngleAxis.h"
-#include "Transform.h"
-#include "Translation.h"
-#include "Scaling.h"
-#include "AlignedBox.h"
-#include "Hyperplane.h"
-#include "ParametrizedLine.h"
-
-#undef ei_toRotationMatrix
-#undef ei_quaternion_assign_impl
-#undef ei_transform_product_impl
-
-#undef RotationBase
-#undef Rotation2D
-#undef Rotation2Df
-#undef Rotation2Dd
-
-#undef Quaternion
-#undef Quaternionf
-#undef Quaterniond
-
-#undef AngleAxis
-#undef AngleAxisf
-#undef AngleAxisd
-
-#undef Transform
-#undef Transform2f
-#undef Transform2d
-#undef Transform3f
-#undef Transform3d
-
-#undef Translation
-#undef Translation2f
-#undef Translation2d
-#undef Translation3f
-#undef Translation3d
-
-#undef Scaling
-#undef Scaling2f
-#undef Scaling2d
-#undef Scaling3f
-#undef Scaling3d
-
-#undef AlignedBox
-
-#undef Hyperplane
-#undef ParametrizedLine
-
-#endif // EIGEN2_GEOMETRY_MODULE_H
diff --git a/Eigen/src/Eigen2Support/Geometry/AngleAxis.h b/Eigen/src/Eigen2Support/Geometry/AngleAxis.h
deleted file mode 100644
index af598a4..0000000
--- a/Eigen/src/Eigen2Support/Geometry/AngleAxis.h
+++ /dev/null
@@ -1,214 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class AngleAxis
-  *
-  * \brief Represents a 3D rotation as a rotation angle around an arbitrary 3D axis
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients.
-  *
-  * The following two typedefs are provided for convenience:
-  * \li \c AngleAxisf for \c float
-  * \li \c AngleAxisd for \c double
-  *
-  * \addexample AngleAxisForEuler \label How to define a rotation from Euler-angles
-  *
-  * Combined with MatrixBase::Unit{X,Y,Z}, AngleAxis can be used to easily
-  * mimic Euler-angles. Here is an example:
-  * \include AngleAxis_mimic_euler.cpp
-  * Output: \verbinclude AngleAxis_mimic_euler.out
-  *
-  * \note This class is not aimed to be used to store a rotation transformation,
-  * but rather to make easier the creation of other rotation (Quaternion, rotation Matrix)
-  * and transformation objects.
-  *
-  * \sa class Quaternion, class Transform, MatrixBase::UnitX()
-  */
-
-template<typename _Scalar> struct ei_traits<AngleAxis<_Scalar> >
-{
-  typedef _Scalar Scalar;
-};
-
-template<typename _Scalar>
-class AngleAxis : public RotationBase<AngleAxis<_Scalar>,3>
-{
-  typedef RotationBase<AngleAxis<_Scalar>,3> Base;
-
-public:
-
-  using Base::operator*;
-
-  enum { Dim = 3 };
-  /** the scalar type of the coefficients */
-  typedef _Scalar Scalar;
-  typedef Matrix<Scalar,3,3> Matrix3;
-  typedef Matrix<Scalar,3,1> Vector3;
-  typedef Quaternion<Scalar> QuaternionType;
-
-protected:
-
-  Vector3 m_axis;
-  Scalar m_angle;
-
-public:
-
-  /** Default constructor without initialization. */
-  AngleAxis() {}
-  /** Constructs and initialize the angle-axis rotation from an \a angle in radian
-    * and an \a axis which must be normalized. */
-  template<typename Derived>
-  inline AngleAxis(Scalar angle, const MatrixBase<Derived>& axis) : m_axis(axis), m_angle(angle) {}
-  /** Constructs and initialize the angle-axis rotation from a quaternion \a q. */
-  inline AngleAxis(const QuaternionType& q) { *this = q; }
-  /** Constructs and initialize the angle-axis rotation from a 3x3 rotation matrix. */
-  template<typename Derived>
-  inline explicit AngleAxis(const MatrixBase<Derived>& m) { *this = m; }
-
-  Scalar angle() const { return m_angle; }
-  Scalar& angle() { return m_angle; }
-
-  const Vector3& axis() const { return m_axis; }
-  Vector3& axis() { return m_axis; }
-
-  /** Concatenates two rotations */
-  inline QuaternionType operator* (const AngleAxis& other) const
-  { return QuaternionType(*this) * QuaternionType(other); }
-
-  /** Concatenates two rotations */
-  inline QuaternionType operator* (const QuaternionType& other) const
-  { return QuaternionType(*this) * other; }
-
-  /** Concatenates two rotations */
-  friend inline QuaternionType operator* (const QuaternionType& a, const AngleAxis& b)
-  { return a * QuaternionType(b); }
-
-  /** Concatenates two rotations */
-  inline Matrix3 operator* (const Matrix3& other) const
-  { return toRotationMatrix() * other; }
-
-  /** Concatenates two rotations */
-  inline friend Matrix3 operator* (const Matrix3& a, const AngleAxis& b)
-  { return a * b.toRotationMatrix(); }
-
-  /** Applies rotation to vector */
-  inline Vector3 operator* (const Vector3& other) const
-  { return toRotationMatrix() * other; }
-
-  /** \returns the inverse rotation, i.e., an angle-axis with opposite rotation angle */
-  AngleAxis inverse() const
-  { return AngleAxis(-m_angle, m_axis); }
-
-  AngleAxis& operator=(const QuaternionType& q);
-  template<typename Derived>
-  AngleAxis& operator=(const MatrixBase<Derived>& m);
-
-  template<typename Derived>
-  AngleAxis& fromRotationMatrix(const MatrixBase<Derived>& m);
-  Matrix3 toRotationMatrix(void) const;
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type cast() const
-  { return typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type(*this); }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit AngleAxis(const AngleAxis<OtherScalarType>& other)
-  {
-    m_axis = other.axis().template cast<Scalar>();
-    m_angle = Scalar(other.angle());
-  }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const AngleAxis& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_axis.isApprox(other.m_axis, prec) && ei_isApprox(m_angle,other.m_angle, prec); }
-};
-
-/** \ingroup Geometry_Module
-  * single precision angle-axis type */
-typedef AngleAxis<float> AngleAxisf;
-/** \ingroup Geometry_Module
-  * double precision angle-axis type */
-typedef AngleAxis<double> AngleAxisd;
-
-/** Set \c *this from a quaternion.
-  * The axis is normalized.
-  */
-template<typename Scalar>
-AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const QuaternionType& q)
-{
-  Scalar n2 = q.vec().squaredNorm();
-  if (n2 < precision<Scalar>()*precision<Scalar>())
-  {
-    m_angle = 0;
-    m_axis << 1, 0, 0;
-  }
-  else
-  {
-    m_angle = 2*std::acos(q.w());
-    m_axis = q.vec() / ei_sqrt(n2);
-  }
-  return *this;
-}
-
-/** Set \c *this from a 3x3 rotation matrix \a mat.
-  */
-template<typename Scalar>
-template<typename Derived>
-AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const MatrixBase<Derived>& mat)
-{
-  // Since a direct conversion would not be really faster,
-  // let's use the robust Quaternion implementation:
-  return *this = QuaternionType(mat);
-}
-
-/** Constructs and \returns an equivalent 3x3 rotation matrix.
-  */
-template<typename Scalar>
-typename AngleAxis<Scalar>::Matrix3
-AngleAxis<Scalar>::toRotationMatrix(void) const
-{
-  Matrix3 res;
-  Vector3 sin_axis  = ei_sin(m_angle) * m_axis;
-  Scalar c = ei_cos(m_angle);
-  Vector3 cos1_axis = (Scalar(1)-c) * m_axis;
-
-  Scalar tmp;
-  tmp = cos1_axis.x() * m_axis.y();
-  res.coeffRef(0,1) = tmp - sin_axis.z();
-  res.coeffRef(1,0) = tmp + sin_axis.z();
-
-  tmp = cos1_axis.x() * m_axis.z();
-  res.coeffRef(0,2) = tmp + sin_axis.y();
-  res.coeffRef(2,0) = tmp - sin_axis.y();
-
-  tmp = cos1_axis.y() * m_axis.z();
-  res.coeffRef(1,2) = tmp - sin_axis.x();
-  res.coeffRef(2,1) = tmp + sin_axis.x();
-
-  res.diagonal() = (cos1_axis.cwise() * m_axis).cwise() + c;
-
-  return res;
-}
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/CMakeLists.txt b/Eigen/src/Eigen2Support/Geometry/CMakeLists.txt
deleted file mode 100644
index c347a8f..0000000
--- a/Eigen/src/Eigen2Support/Geometry/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Eigen2Support_Geometry_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Eigen2Support_Geometry_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Eigen2Support/Geometry
-  )
diff --git a/Eigen/src/Eigen2Support/Geometry/Hyperplane.h b/Eigen/src/Eigen2Support/Geometry/Hyperplane.h
deleted file mode 100644
index b95bf00..0000000
--- a/Eigen/src/Eigen2Support/Geometry/Hyperplane.h
+++ /dev/null
@@ -1,254 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class Hyperplane
-  *
-  * \brief A hyperplane
-  *
-  * A hyperplane is an affine subspace of dimension n-1 in a space of dimension n.
-  * For example, a hyperplane in a plane is a line; a hyperplane in 3-space is a plane.
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  * \param _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
-  *             Notice that the dimension of the hyperplane is _AmbientDim-1.
-  *
-  * This class represents an hyperplane as the zero set of the implicit equation
-  * \f$ n \cdot x + d = 0 \f$ where \f$ n \f$ is a unit normal vector of the plane (linear part)
-  * and \f$ d \f$ is the distance (offset) to the origin.
-  */
-template <typename _Scalar, int _AmbientDim>
-class Hyperplane
-{
-public:
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim==Dynamic ? Dynamic : _AmbientDim+1)
-  enum { AmbientDimAtCompileTime = _AmbientDim };
-  typedef _Scalar Scalar;
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef Matrix<Scalar,AmbientDimAtCompileTime,1> VectorType;
-  typedef Matrix<Scalar,int(AmbientDimAtCompileTime)==Dynamic
-                        ? Dynamic
-                        : int(AmbientDimAtCompileTime)+1,1> Coefficients;
-  typedef Block<Coefficients,AmbientDimAtCompileTime,1> NormalReturnType;
-
-  /** Default constructor without initialization */
-  inline Hyperplane() {}
-
-  /** Constructs a dynamic-size hyperplane with \a _dim the dimension
-    * of the ambient space */
-  inline explicit Hyperplane(int _dim) : m_coeffs(_dim+1) {}
-
-  /** Construct a plane from its normal \a n and a point \a e onto the plane.
-    * \warning the vector normal is assumed to be normalized.
-    */
-  inline Hyperplane(const VectorType& n, const VectorType& e)
-    : m_coeffs(n.size()+1)
-  {
-    normal() = n;
-    offset() = -e.eigen2_dot(n);
-  }
-
-  /** Constructs a plane from its normal \a n and distance to the origin \a d
-    * such that the algebraic equation of the plane is \f$ n \cdot x + d = 0 \f$.
-    * \warning the vector normal is assumed to be normalized.
-    */
-  inline Hyperplane(const VectorType& n, Scalar d)
-    : m_coeffs(n.size()+1)
-  {
-    normal() = n;
-    offset() = d;
-  }
-
-  /** Constructs a hyperplane passing through the two points. If the dimension of the ambient space
-    * is greater than 2, then there isn't uniqueness, so an arbitrary choice is made.
-    */
-  static inline Hyperplane Through(const VectorType& p0, const VectorType& p1)
-  {
-    Hyperplane result(p0.size());
-    result.normal() = (p1 - p0).unitOrthogonal();
-    result.offset() = -result.normal().eigen2_dot(p0);
-    return result;
-  }
-
-  /** Constructs a hyperplane passing through the three points. The dimension of the ambient space
-    * is required to be exactly 3.
-    */
-  static inline Hyperplane Through(const VectorType& p0, const VectorType& p1, const VectorType& p2)
-  {
-    EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 3)
-    Hyperplane result(p0.size());
-    result.normal() = (p2 - p0).cross(p1 - p0).normalized();
-    result.offset() = -result.normal().eigen2_dot(p0);
-    return result;
-  }
-
-  /** Constructs a hyperplane passing through the parametrized line \a parametrized.
-    * If the dimension of the ambient space is greater than 2, then there isn't uniqueness,
-    * so an arbitrary choice is made.
-    */
-  // FIXME to be consitent with the rest this could be implemented as a static Through function ??
-  explicit Hyperplane(const ParametrizedLine<Scalar, AmbientDimAtCompileTime>& parametrized)
-  {
-    normal() = parametrized.direction().unitOrthogonal();
-    offset() = -normal().eigen2_dot(parametrized.origin());
-  }
-
-  ~Hyperplane() {}
-
-  /** \returns the dimension in which the plane holds */
-  inline int dim() const { return int(AmbientDimAtCompileTime)==Dynamic ? m_coeffs.size()-1 : int(AmbientDimAtCompileTime); }
-
-  /** normalizes \c *this */
-  void normalize(void)
-  {
-    m_coeffs /= normal().norm();
-  }
-
-  /** \returns the signed distance between the plane \c *this and a point \a p.
-    * \sa absDistance()
-    */
-  inline Scalar signedDistance(const VectorType& p) const { return p.eigen2_dot(normal()) + offset(); }
-
-  /** \returns the absolute distance between the plane \c *this and a point \a p.
-    * \sa signedDistance()
-    */
-  inline Scalar absDistance(const VectorType& p) const { return ei_abs(signedDistance(p)); }
-
-  /** \returns the projection of a point \a p onto the plane \c *this.
-    */
-  inline VectorType projection(const VectorType& p) const { return p - signedDistance(p) * normal(); }
-
-  /** \returns a constant reference to the unit normal vector of the plane, which corresponds
-    * to the linear part of the implicit equation.
-    */
-  inline const NormalReturnType normal() const { return NormalReturnType(*const_cast<Coefficients*>(&m_coeffs),0,0,dim(),1); }
-
-  /** \returns a non-constant reference to the unit normal vector of the plane, which corresponds
-    * to the linear part of the implicit equation.
-    */
-  inline NormalReturnType normal() { return NormalReturnType(m_coeffs,0,0,dim(),1); }
-
-  /** \returns the distance to the origin, which is also the "constant term" of the implicit equation
-    * \warning the vector normal is assumed to be normalized.
-    */
-  inline const Scalar& offset() const { return m_coeffs.coeff(dim()); }
-
-  /** \returns a non-constant reference to the distance to the origin, which is also the constant part
-    * of the implicit equation */
-  inline Scalar& offset() { return m_coeffs(dim()); }
-
-  /** \returns a constant reference to the coefficients c_i of the plane equation:
-    * \f$ c_0*x_0 + ... + c_{d-1}*x_{d-1} + c_d = 0 \f$
-    */
-  inline const Coefficients& coeffs() const { return m_coeffs; }
-
-  /** \returns a non-constant reference to the coefficients c_i of the plane equation:
-    * \f$ c_0*x_0 + ... + c_{d-1}*x_{d-1} + c_d = 0 \f$
-    */
-  inline Coefficients& coeffs() { return m_coeffs; }
-
-  /** \returns the intersection of *this with \a other.
-    *
-    * \warning The ambient space must be a plane, i.e. have dimension 2, so that \c *this and \a other are lines.
-    *
-    * \note If \a other is approximately parallel to *this, this method will return any point on *this.
-    */
-  VectorType intersection(const Hyperplane& other)
-  {
-    EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 2)
-    Scalar det = coeffs().coeff(0) * other.coeffs().coeff(1) - coeffs().coeff(1) * other.coeffs().coeff(0);
-    // since the line equations ax+by=c are normalized with a^2+b^2=1, the following tests
-    // whether the two lines are approximately parallel.
-    if(ei_isMuchSmallerThan(det, Scalar(1)))
-    {   // special case where the two lines are approximately parallel. Pick any point on the first line.
-        if(ei_abs(coeffs().coeff(1))>ei_abs(coeffs().coeff(0)))
-            return VectorType(coeffs().coeff(1), -coeffs().coeff(2)/coeffs().coeff(1)-coeffs().coeff(0));
-        else
-            return VectorType(-coeffs().coeff(2)/coeffs().coeff(0)-coeffs().coeff(1), coeffs().coeff(0));
-    }
-    else
-    {   // general case
-        Scalar invdet = Scalar(1) / det;
-        return VectorType(invdet*(coeffs().coeff(1)*other.coeffs().coeff(2)-other.coeffs().coeff(1)*coeffs().coeff(2)),
-                          invdet*(other.coeffs().coeff(0)*coeffs().coeff(2)-coeffs().coeff(0)*other.coeffs().coeff(2)));
-    }
-  }
-
-  /** Applies the transformation matrix \a mat to \c *this and returns a reference to \c *this.
-    *
-    * \param mat the Dim x Dim transformation matrix
-    * \param traits specifies whether the matrix \a mat represents an Isometry
-    *               or a more generic Affine transformation. The default is Affine.
-    */
-  template<typename XprType>
-  inline Hyperplane& transform(const MatrixBase<XprType>& mat, TransformTraits traits = Affine)
-  {
-    if (traits==Affine)
-      normal() = mat.inverse().transpose() * normal();
-    else if (traits==Isometry)
-      normal() = mat * normal();
-    else
-    {
-      ei_assert("invalid traits value in Hyperplane::transform()");
-    }
-    return *this;
-  }
-
-  /** Applies the transformation \a t to \c *this and returns a reference to \c *this.
-    *
-    * \param t the transformation of dimension Dim
-    * \param traits specifies whether the transformation \a t represents an Isometry
-    *               or a more generic Affine transformation. The default is Affine.
-    *               Other kind of transformations are not supported.
-    */
-  inline Hyperplane& transform(const Transform<Scalar,AmbientDimAtCompileTime>& t,
-                                TransformTraits traits = Affine)
-  {
-    transform(t.linear(), traits);
-    offset() -= t.translation().eigen2_dot(normal());
-    return *this;
-  }
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<Hyperplane,
-           Hyperplane<NewScalarType,AmbientDimAtCompileTime> >::type cast() const
-  {
-    return typename internal::cast_return_type<Hyperplane,
-                    Hyperplane<NewScalarType,AmbientDimAtCompileTime> >::type(*this);
-  }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit Hyperplane(const Hyperplane<OtherScalarType,AmbientDimAtCompileTime>& other)
-  { m_coeffs = other.coeffs().template cast<Scalar>(); }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const Hyperplane& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_coeffs.isApprox(other.m_coeffs, prec); }
-
-protected:
-
-  Coefficients m_coeffs;
-};
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/ParametrizedLine.h b/Eigen/src/Eigen2Support/Geometry/ParametrizedLine.h
deleted file mode 100644
index 9b57b7e..0000000
--- a/Eigen/src/Eigen2Support/Geometry/ParametrizedLine.h
+++ /dev/null
@@ -1,141 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class ParametrizedLine
-  *
-  * \brief A parametrized line
-  *
-  * A parametrized line is defined by an origin point \f$ \mathbf{o} \f$ and a unit
-  * direction vector \f$ \mathbf{d} \f$ such that the line corresponds to
-  * the set \f$ l(t) = \mathbf{o} + t \mathbf{d} \f$, \f$ l \in \mathbf{R} \f$.
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  * \param _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
-  */
-template <typename _Scalar, int _AmbientDim>
-class ParametrizedLine
-{
-public:
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_AmbientDim)
-  enum { AmbientDimAtCompileTime = _AmbientDim };
-  typedef _Scalar Scalar;
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef Matrix<Scalar,AmbientDimAtCompileTime,1> VectorType;
-
-  /** Default constructor without initialization */
-  inline ParametrizedLine() {}
-
-  /** Constructs a dynamic-size line with \a _dim the dimension
-    * of the ambient space */
-  inline explicit ParametrizedLine(int _dim) : m_origin(_dim), m_direction(_dim) {}
-
-  /** Initializes a parametrized line of direction \a direction and origin \a origin.
-    * \warning the vector direction is assumed to be normalized.
-    */
-  ParametrizedLine(const VectorType& origin, const VectorType& direction)
-    : m_origin(origin), m_direction(direction) {}
-
-  explicit ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim>& hyperplane);
-
-  /** Constructs a parametrized line going from \a p0 to \a p1. */
-  static inline ParametrizedLine Through(const VectorType& p0, const VectorType& p1)
-  { return ParametrizedLine(p0, (p1-p0).normalized()); }
-
-  ~ParametrizedLine() {}
-
-  /** \returns the dimension in which the line holds */
-  inline int dim() const { return m_direction.size(); }
-
-  const VectorType& origin() const { return m_origin; }
-  VectorType& origin() { return m_origin; }
-
-  const VectorType& direction() const { return m_direction; }
-  VectorType& direction() { return m_direction; }
-
-  /** \returns the squared distance of a point \a p to its projection onto the line \c *this.
-    * \sa distance()
-    */
-  RealScalar squaredDistance(const VectorType& p) const
-  {
-    VectorType diff = p-origin();
-    return (diff - diff.eigen2_dot(direction())* direction()).squaredNorm();
-  }
-  /** \returns the distance of a point \a p to its projection onto the line \c *this.
-    * \sa squaredDistance()
-    */
-  RealScalar distance(const VectorType& p) const { return ei_sqrt(squaredDistance(p)); }
-
-  /** \returns the projection of a point \a p onto the line \c *this. */
-  VectorType projection(const VectorType& p) const
-  { return origin() + (p-origin()).eigen2_dot(direction()) * direction(); }
-
-  Scalar intersection(const Hyperplane<_Scalar, _AmbientDim>& hyperplane);
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<ParametrizedLine,
-           ParametrizedLine<NewScalarType,AmbientDimAtCompileTime> >::type cast() const
-  {
-    return typename internal::cast_return_type<ParametrizedLine,
-                    ParametrizedLine<NewScalarType,AmbientDimAtCompileTime> >::type(*this);
-  }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit ParametrizedLine(const ParametrizedLine<OtherScalarType,AmbientDimAtCompileTime>& other)
-  {
-    m_origin = other.origin().template cast<Scalar>();
-    m_direction = other.direction().template cast<Scalar>();
-  }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const ParametrizedLine& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_origin.isApprox(other.m_origin, prec) && m_direction.isApprox(other.m_direction, prec); }
-
-protected:
-
-  VectorType m_origin, m_direction;
-};
-
-/** Constructs a parametrized line from a 2D hyperplane
-  *
-  * \warning the ambient space must have dimension 2 such that the hyperplane actually describes a line
-  */
-template <typename _Scalar, int _AmbientDim>
-inline ParametrizedLine<_Scalar, _AmbientDim>::ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim>& hyperplane)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 2)
-  direction() = hyperplane.normal().unitOrthogonal();
-  origin() = -hyperplane.normal()*hyperplane.offset();
-}
-
-/** \returns the parameter value of the intersection between \c *this and the given hyperplane
-  */
-template <typename _Scalar, int _AmbientDim>
-inline _Scalar ParametrizedLine<_Scalar, _AmbientDim>::intersection(const Hyperplane<_Scalar, _AmbientDim>& hyperplane)
-{
-  return -(hyperplane.offset()+origin().eigen2_dot(hyperplane.normal()))
-          /(direction().eigen2_dot(hyperplane.normal()));
-}
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/Quaternion.h b/Eigen/src/Eigen2Support/Geometry/Quaternion.h
deleted file mode 100644
index 4b6390c..0000000
--- a/Eigen/src/Eigen2Support/Geometry/Quaternion.h
+++ /dev/null
@@ -1,495 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-template<typename Other,
-         int OtherRows=Other::RowsAtCompileTime,
-         int OtherCols=Other::ColsAtCompileTime>
-struct ei_quaternion_assign_impl;
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class Quaternion
-  *
-  * \brief The quaternion class used to represent 3D orientations and rotations
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  *
-  * This class represents a quaternion \f$ w+xi+yj+zk \f$ that is a convenient representation of
-  * orientations and rotations of objects in three dimensions. Compared to other representations
-  * like Euler angles or 3x3 matrices, quatertions offer the following advantages:
-  * \li \b compact storage (4 scalars)
-  * \li \b efficient to compose (28 flops),
-  * \li \b stable spherical interpolation
-  *
-  * The following two typedefs are provided for convenience:
-  * \li \c Quaternionf for \c float
-  * \li \c Quaterniond for \c double
-  *
-  * \sa  class AngleAxis, class Transform
-  */
-
-template<typename _Scalar> struct ei_traits<Quaternion<_Scalar> >
-{
-  typedef _Scalar Scalar;
-};
-
-template<typename _Scalar>
-class Quaternion : public RotationBase<Quaternion<_Scalar>,3>
-{
-  typedef RotationBase<Quaternion<_Scalar>,3> Base;
-
-public:
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,4)
-
-  using Base::operator*;
-
-  /** the scalar type of the coefficients */
-  typedef _Scalar Scalar;
-
-  /** the type of the Coefficients 4-vector */
-  typedef Matrix<Scalar, 4, 1> Coefficients;
-  /** the type of a 3D vector */
-  typedef Matrix<Scalar,3,1> Vector3;
-  /** the equivalent rotation matrix type */
-  typedef Matrix<Scalar,3,3> Matrix3;
-  /** the equivalent angle-axis type */
-  typedef AngleAxis<Scalar> AngleAxisType;
-
-  /** \returns the \c x coefficient */
-  inline Scalar x() const { return m_coeffs.coeff(0); }
-  /** \returns the \c y coefficient */
-  inline Scalar y() const { return m_coeffs.coeff(1); }
-  /** \returns the \c z coefficient */
-  inline Scalar z() const { return m_coeffs.coeff(2); }
-  /** \returns the \c w coefficient */
-  inline Scalar w() const { return m_coeffs.coeff(3); }
-
-  /** \returns a reference to the \c x coefficient */
-  inline Scalar& x() { return m_coeffs.coeffRef(0); }
-  /** \returns a reference to the \c y coefficient */
-  inline Scalar& y() { return m_coeffs.coeffRef(1); }
-  /** \returns a reference to the \c z coefficient */
-  inline Scalar& z() { return m_coeffs.coeffRef(2); }
-  /** \returns a reference to the \c w coefficient */
-  inline Scalar& w() { return m_coeffs.coeffRef(3); }
-
-  /** \returns a read-only vector expression of the imaginary part (x,y,z) */
-  inline const Block<const Coefficients,3,1> vec() const { return m_coeffs.template start<3>(); }
-
-  /** \returns a vector expression of the imaginary part (x,y,z) */
-  inline Block<Coefficients,3,1> vec() { return m_coeffs.template start<3>(); }
-
-  /** \returns a read-only vector expression of the coefficients (x,y,z,w) */
-  inline const Coefficients& coeffs() const { return m_coeffs; }
-
-  /** \returns a vector expression of the coefficients (x,y,z,w) */
-  inline Coefficients& coeffs() { return m_coeffs; }
-
-  /** Default constructor leaving the quaternion uninitialized. */
-  inline Quaternion() {}
-
-  /** Constructs and initializes the quaternion \f$ w+xi+yj+zk \f$ from
-    * its four coefficients \a w, \a x, \a y and \a z.
-    *
-    * \warning Note the order of the arguments: the real \a w coefficient first,
-    * while internally the coefficients are stored in the following order:
-    * [\c x, \c y, \c z, \c w]
-    */
-  inline Quaternion(Scalar w, Scalar x, Scalar y, Scalar z)
-  { m_coeffs << x, y, z, w; }
-
-  /** Copy constructor */
-  inline Quaternion(const Quaternion& other) { m_coeffs = other.m_coeffs; }
-
-  /** Constructs and initializes a quaternion from the angle-axis \a aa */
-  explicit inline Quaternion(const AngleAxisType& aa) { *this = aa; }
-
-  /** Constructs and initializes a quaternion from either:
-    *  - a rotation matrix expression,
-    *  - a 4D vector expression representing quaternion coefficients.
-    * \sa operator=(MatrixBase<Derived>)
-    */
-  template<typename Derived>
-  explicit inline Quaternion(const MatrixBase<Derived>& other) { *this = other; }
-
-  Quaternion& operator=(const Quaternion& other);
-  Quaternion& operator=(const AngleAxisType& aa);
-  template<typename Derived>
-  Quaternion& operator=(const MatrixBase<Derived>& m);
-
-  /** \returns a quaternion representing an identity rotation
-    * \sa MatrixBase::Identity()
-    */
-  static inline Quaternion Identity() { return Quaternion(1, 0, 0, 0); }
-
-  /** \sa Quaternion::Identity(), MatrixBase::setIdentity()
-    */
-  inline Quaternion& setIdentity() { m_coeffs << 0, 0, 0, 1; return *this; }
-
-  /** \returns the squared norm of the quaternion's coefficients
-    * \sa Quaternion::norm(), MatrixBase::squaredNorm()
-    */
-  inline Scalar squaredNorm() const { return m_coeffs.squaredNorm(); }
-
-  /** \returns the norm of the quaternion's coefficients
-    * \sa Quaternion::squaredNorm(), MatrixBase::norm()
-    */
-  inline Scalar norm() const { return m_coeffs.norm(); }
-
-  /** Normalizes the quaternion \c *this
-    * \sa normalized(), MatrixBase::normalize() */
-  inline void normalize() { m_coeffs.normalize(); }
-  /** \returns a normalized version of \c *this
-    * \sa normalize(), MatrixBase::normalized() */
-  inline Quaternion normalized() const { return Quaternion(m_coeffs.normalized()); }
-
-  /** \returns the dot product of \c *this and \a other
-    * Geometrically speaking, the dot product of two unit quaternions
-    * corresponds to the cosine of half the angle between the two rotations.
-    * \sa angularDistance()
-    */
-  inline Scalar eigen2_dot(const Quaternion& other) const { return m_coeffs.eigen2_dot(other.m_coeffs); }
-
-  inline Scalar angularDistance(const Quaternion& other) const;
-
-  Matrix3 toRotationMatrix(void) const;
-
-  template<typename Derived1, typename Derived2>
-  Quaternion& setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
-
-  inline Quaternion operator* (const Quaternion& q) const;
-  inline Quaternion& operator*= (const Quaternion& q);
-
-  Quaternion inverse(void) const;
-  Quaternion conjugate(void) const;
-
-  Quaternion slerp(Scalar t, const Quaternion& other) const;
-
-  template<typename Derived>
-  Vector3 operator* (const MatrixBase<Derived>& vec) const;
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<Quaternion,Quaternion<NewScalarType> >::type cast() const
-  { return typename internal::cast_return_type<Quaternion,Quaternion<NewScalarType> >::type(*this); }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit Quaternion(const Quaternion<OtherScalarType>& other)
-  { m_coeffs = other.coeffs().template cast<Scalar>(); }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const Quaternion& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_coeffs.isApprox(other.m_coeffs, prec); }
-
-protected:
-  Coefficients m_coeffs;
-};
-
-/** \ingroup Geometry_Module
-  * single precision quaternion type */
-typedef Quaternion<float> Quaternionf;
-/** \ingroup Geometry_Module
-  * double precision quaternion type */
-typedef Quaternion<double> Quaterniond;
-
-// Generic Quaternion * Quaternion product
-template<typename Scalar> inline Quaternion<Scalar>
-ei_quaternion_product(const Quaternion<Scalar>& a, const Quaternion<Scalar>& b)
-{
-  return Quaternion<Scalar>
-  (
-    a.w() * b.w() - a.x() * b.x() - a.y() * b.y() - a.z() * b.z(),
-    a.w() * b.x() + a.x() * b.w() + a.y() * b.z() - a.z() * b.y(),
-    a.w() * b.y() + a.y() * b.w() + a.z() * b.x() - a.x() * b.z(),
-    a.w() * b.z() + a.z() * b.w() + a.x() * b.y() - a.y() * b.x()
-  );
-}
-
-/** \returns the concatenation of two rotations as a quaternion-quaternion product */
-template <typename Scalar>
-inline Quaternion<Scalar> Quaternion<Scalar>::operator* (const Quaternion& other) const
-{
-  return ei_quaternion_product(*this,other);
-}
-
-/** \sa operator*(Quaternion) */
-template <typename Scalar>
-inline Quaternion<Scalar>& Quaternion<Scalar>::operator*= (const Quaternion& other)
-{
-  return (*this = *this * other);
-}
-
-/** Rotation of a vector by a quaternion.
-  * \remarks If the quaternion is used to rotate several points (>1)
-  * then it is much more efficient to first convert it to a 3x3 Matrix.
-  * Comparison of the operation cost for n transformations:
-  *   - Quaternion:    30n
-  *   - Via a Matrix3: 24 + 15n
-  */
-template <typename Scalar>
-template<typename Derived>
-inline typename Quaternion<Scalar>::Vector3
-Quaternion<Scalar>::operator* (const MatrixBase<Derived>& v) const
-{
-    // Note that this algorithm comes from the optimization by hand
-    // of the conversion to a Matrix followed by a Matrix/Vector product.
-    // It appears to be much faster than the common algorithm found
-    // in the litterature (30 versus 39 flops). It also requires two
-    // Vector3 as temporaries.
-    Vector3 uv;
-    uv = 2 * this->vec().cross(v);
-    return v + this->w() * uv + this->vec().cross(uv);
-}
-
-template<typename Scalar>
-inline Quaternion<Scalar>& Quaternion<Scalar>::operator=(const Quaternion& other)
-{
-  m_coeffs = other.m_coeffs;
-  return *this;
-}
-
-/** Set \c *this from an angle-axis \a aa and returns a reference to \c *this
-  */
-template<typename Scalar>
-inline Quaternion<Scalar>& Quaternion<Scalar>::operator=(const AngleAxisType& aa)
-{
-  Scalar ha = Scalar(0.5)*aa.angle(); // Scalar(0.5) to suppress precision loss warnings
-  this->w() = ei_cos(ha);
-  this->vec() = ei_sin(ha) * aa.axis();
-  return *this;
-}
-
-/** Set \c *this from the expression \a xpr:
-  *   - if \a xpr is a 4x1 vector, then \a xpr is assumed to be a quaternion
-  *   - if \a xpr is a 3x3 matrix, then \a xpr is assumed to be rotation matrix
-  *     and \a xpr is converted to a quaternion
-  */
-template<typename Scalar>
-template<typename Derived>
-inline Quaternion<Scalar>& Quaternion<Scalar>::operator=(const MatrixBase<Derived>& xpr)
-{
-  ei_quaternion_assign_impl<Derived>::run(*this, xpr.derived());
-  return *this;
-}
-
-/** Convert the quaternion to a 3x3 rotation matrix */
-template<typename Scalar>
-inline typename Quaternion<Scalar>::Matrix3
-Quaternion<Scalar>::toRotationMatrix(void) const
-{
-  // NOTE if inlined, then gcc 4.2 and 4.4 get rid of the temporary (not gcc 4.3 !!)
-  // if not inlined then the cost of the return by value is huge ~ +35%,
-  // however, not inlining this function is an order of magnitude slower, so
-  // it has to be inlined, and so the return by value is not an issue
-  Matrix3 res;
-
-  const Scalar tx  = Scalar(2)*this->x();
-  const Scalar ty  = Scalar(2)*this->y();
-  const Scalar tz  = Scalar(2)*this->z();
-  const Scalar twx = tx*this->w();
-  const Scalar twy = ty*this->w();
-  const Scalar twz = tz*this->w();
-  const Scalar txx = tx*this->x();
-  const Scalar txy = ty*this->x();
-  const Scalar txz = tz*this->x();
-  const Scalar tyy = ty*this->y();
-  const Scalar tyz = tz*this->y();
-  const Scalar tzz = tz*this->z();
-
-  res.coeffRef(0,0) = Scalar(1)-(tyy+tzz);
-  res.coeffRef(0,1) = txy-twz;
-  res.coeffRef(0,2) = txz+twy;
-  res.coeffRef(1,0) = txy+twz;
-  res.coeffRef(1,1) = Scalar(1)-(txx+tzz);
-  res.coeffRef(1,2) = tyz-twx;
-  res.coeffRef(2,0) = txz-twy;
-  res.coeffRef(2,1) = tyz+twx;
-  res.coeffRef(2,2) = Scalar(1)-(txx+tyy);
-
-  return res;
-}
-
-/** Sets *this to be a quaternion representing a rotation sending the vector \a a to the vector \a b.
-  *
-  * \returns a reference to *this.
-  *
-  * Note that the two input vectors do \b not have to be normalized.
-  */
-template<typename Scalar>
-template<typename Derived1, typename Derived2>
-inline Quaternion<Scalar>& Quaternion<Scalar>::setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
-{
-  Vector3 v0 = a.normalized();
-  Vector3 v1 = b.normalized();
-  Scalar c = v0.eigen2_dot(v1);
-
-  // if dot == 1, vectors are the same
-  if (ei_isApprox(c,Scalar(1)))
-  {
-    // set to identity
-    this->w() = 1; this->vec().setZero();
-    return *this;
-  }
-  // if dot == -1, vectors are opposites
-  if (ei_isApprox(c,Scalar(-1)))
-  {
-    this->vec() = v0.unitOrthogonal();
-    this->w() = 0;
-    return *this;
-  }
-
-  Vector3 axis = v0.cross(v1);
-  Scalar s = ei_sqrt((Scalar(1)+c)*Scalar(2));
-  Scalar invs = Scalar(1)/s;
-  this->vec() = axis * invs;
-  this->w() = s * Scalar(0.5);
-
-  return *this;
-}
-
-/** \returns the multiplicative inverse of \c *this
-  * Note that in most cases, i.e., if you simply want the opposite rotation,
-  * and/or the quaternion is normalized, then it is enough to use the conjugate.
-  *
-  * \sa Quaternion::conjugate()
-  */
-template <typename Scalar>
-inline Quaternion<Scalar> Quaternion<Scalar>::inverse() const
-{
-  // FIXME should this function be called multiplicativeInverse and conjugate() be called inverse() or opposite()  ??
-  Scalar n2 = this->squaredNorm();
-  if (n2 > 0)
-    return Quaternion(conjugate().coeffs() / n2);
-  else
-  {
-    // return an invalid result to flag the error
-    return Quaternion(Coefficients::Zero());
-  }
-}
-
-/** \returns the conjugate of the \c *this which is equal to the multiplicative inverse
-  * if the quaternion is normalized.
-  * The conjugate of a quaternion represents the opposite rotation.
-  *
-  * \sa Quaternion::inverse()
-  */
-template <typename Scalar>
-inline Quaternion<Scalar> Quaternion<Scalar>::conjugate() const
-{
-  return Quaternion(this->w(),-this->x(),-this->y(),-this->z());
-}
-
-/** \returns the angle (in radian) between two rotations
-  * \sa eigen2_dot()
-  */
-template <typename Scalar>
-inline Scalar Quaternion<Scalar>::angularDistance(const Quaternion& other) const
-{
-  double d = ei_abs(this->eigen2_dot(other));
-  if (d>=1.0)
-    return 0;
-  return Scalar(2) * std::acos(d);
-}
-
-/** \returns the spherical linear interpolation between the two quaternions
-  * \c *this and \a other at the parameter \a t
-  */
-template <typename Scalar>
-Quaternion<Scalar> Quaternion<Scalar>::slerp(Scalar t, const Quaternion& other) const
-{
-  static const Scalar one = Scalar(1) - machine_epsilon<Scalar>();
-  Scalar d = this->eigen2_dot(other);
-  Scalar absD = ei_abs(d);
-
-  Scalar scale0;
-  Scalar scale1;
-
-  if (absD>=one)
-  {
-    scale0 = Scalar(1) - t;
-    scale1 = t;
-  }
-  else
-  {
-    // theta is the angle between the 2 quaternions
-    Scalar theta = std::acos(absD);
-    Scalar sinTheta = ei_sin(theta);
-
-    scale0 = ei_sin( ( Scalar(1) - t ) * theta) / sinTheta;
-    scale1 = ei_sin( ( t * theta) ) / sinTheta;
-    if (d<0)
-      scale1 = -scale1;
-  }
-
-  return Quaternion<Scalar>(scale0 * coeffs() + scale1 * other.coeffs());
-}
-
-// set from a rotation matrix
-template<typename Other>
-struct ei_quaternion_assign_impl<Other,3,3>
-{
-  typedef typename Other::Scalar Scalar;
-  static inline void run(Quaternion<Scalar>& q, const Other& mat)
-  {
-    // This algorithm comes from  "Quaternion Calculus and Fast Animation",
-    // Ken Shoemake, 1987 SIGGRAPH course notes
-    Scalar t = mat.trace();
-    if (t > 0)
-    {
-      t = ei_sqrt(t + Scalar(1.0));
-      q.w() = Scalar(0.5)*t;
-      t = Scalar(0.5)/t;
-      q.x() = (mat.coeff(2,1) - mat.coeff(1,2)) * t;
-      q.y() = (mat.coeff(0,2) - mat.coeff(2,0)) * t;
-      q.z() = (mat.coeff(1,0) - mat.coeff(0,1)) * t;
-    }
-    else
-    {
-      int i = 0;
-      if (mat.coeff(1,1) > mat.coeff(0,0))
-        i = 1;
-      if (mat.coeff(2,2) > mat.coeff(i,i))
-        i = 2;
-      int j = (i+1)%3;
-      int k = (j+1)%3;
-
-      t = ei_sqrt(mat.coeff(i,i)-mat.coeff(j,j)-mat.coeff(k,k) + Scalar(1.0));
-      q.coeffs().coeffRef(i) = Scalar(0.5) * t;
-      t = Scalar(0.5)/t;
-      q.w() = (mat.coeff(k,j)-mat.coeff(j,k))*t;
-      q.coeffs().coeffRef(j) = (mat.coeff(j,i)+mat.coeff(i,j))*t;
-      q.coeffs().coeffRef(k) = (mat.coeff(k,i)+mat.coeff(i,k))*t;
-    }
-  }
-};
-
-// set from a vector of coefficients assumed to be a quaternion
-template<typename Other>
-struct ei_quaternion_assign_impl<Other,4,1>
-{
-  typedef typename Other::Scalar Scalar;
-  static inline void run(Quaternion<Scalar>& q, const Other& vec)
-  {
-    q.coeffs() = vec;
-  }
-};
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/Rotation2D.h b/Eigen/src/Eigen2Support/Geometry/Rotation2D.h
deleted file mode 100644
index 19b8582..0000000
--- a/Eigen/src/Eigen2Support/Geometry/Rotation2D.h
+++ /dev/null
@@ -1,145 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class Rotation2D
-  *
-  * \brief Represents a rotation/orientation in a 2 dimensional space.
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  *
-  * This class is equivalent to a single scalar representing a counter clock wise rotation
-  * as a single angle in radian. It provides some additional features such as the automatic
-  * conversion from/to a 2x2 rotation matrix. Moreover this class aims to provide a similar
-  * interface to Quaternion in order to facilitate the writing of generic algorithms
-  * dealing with rotations.
-  *
-  * \sa class Quaternion, class Transform
-  */
-template<typename _Scalar> struct ei_traits<Rotation2D<_Scalar> >
-{
-  typedef _Scalar Scalar;
-};
-
-template<typename _Scalar>
-class Rotation2D : public RotationBase<Rotation2D<_Scalar>,2>
-{
-  typedef RotationBase<Rotation2D<_Scalar>,2> Base;
-
-public:
-
-  using Base::operator*;
-
-  enum { Dim = 2 };
-  /** the scalar type of the coefficients */
-  typedef _Scalar Scalar;
-  typedef Matrix<Scalar,2,1> Vector2;
-  typedef Matrix<Scalar,2,2> Matrix2;
-
-protected:
-
-  Scalar m_angle;
-
-public:
-
-  /** Construct a 2D counter clock wise rotation from the angle \a a in radian. */
-  inline Rotation2D(Scalar a) : m_angle(a) {}
-
-  /** \returns the rotation angle */
-  inline Scalar angle() const { return m_angle; }
-
-  /** \returns a read-write reference to the rotation angle */
-  inline Scalar& angle() { return m_angle; }
-
-  /** \returns the inverse rotation */
-  inline Rotation2D inverse() const { return -m_angle; }
-
-  /** Concatenates two rotations */
-  inline Rotation2D operator*(const Rotation2D& other) const
-  { return m_angle + other.m_angle; }
-
-  /** Concatenates two rotations */
-  inline Rotation2D& operator*=(const Rotation2D& other)
-  { return m_angle += other.m_angle; return *this; }
-
-  /** Applies the rotation to a 2D vector */
-  Vector2 operator* (const Vector2& vec) const
-  { return toRotationMatrix() * vec; }
-
-  template<typename Derived>
-  Rotation2D& fromRotationMatrix(const MatrixBase<Derived>& m);
-  Matrix2 toRotationMatrix(void) const;
-
-  /** \returns the spherical interpolation between \c *this and \a other using
-    * parameter \a t. It is in fact equivalent to a linear interpolation.
-    */
-  inline Rotation2D slerp(Scalar t, const Rotation2D& other) const
-  { return m_angle * (1-t) + other.angle() * t; }
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type cast() const
-  { return typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type(*this); }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit Rotation2D(const Rotation2D<OtherScalarType>& other)
-  {
-    m_angle = Scalar(other.angle());
-  }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const Rotation2D& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return ei_isApprox(m_angle,other.m_angle, prec); }
-};
-
-/** \ingroup Geometry_Module
-  * single precision 2D rotation type */
-typedef Rotation2D<float> Rotation2Df;
-/** \ingroup Geometry_Module
-  * double precision 2D rotation type */
-typedef Rotation2D<double> Rotation2Dd;
-
-/** Set \c *this from a 2x2 rotation matrix \a mat.
-  * In other words, this function extract the rotation angle
-  * from the rotation matrix.
-  */
-template<typename Scalar>
-template<typename Derived>
-Rotation2D<Scalar>& Rotation2D<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
-{
-  EIGEN_STATIC_ASSERT(Derived::RowsAtCompileTime==2 && Derived::ColsAtCompileTime==2,YOU_MADE_A_PROGRAMMING_MISTAKE)
-  m_angle = ei_atan2(mat.coeff(1,0), mat.coeff(0,0));
-  return *this;
-}
-
-/** Constructs and \returns an equivalent 2x2 rotation matrix.
-  */
-template<typename Scalar>
-typename Rotation2D<Scalar>::Matrix2
-Rotation2D<Scalar>::toRotationMatrix(void) const
-{
-  Scalar sinA = ei_sin(m_angle);
-  Scalar cosA = ei_cos(m_angle);
-  return (Matrix2() << cosA, -sinA, sinA, cosA).finished();
-}
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/RotationBase.h b/Eigen/src/Eigen2Support/Geometry/RotationBase.h
deleted file mode 100644
index b1c8f38..0000000
--- a/Eigen/src/Eigen2Support/Geometry/RotationBase.h
+++ /dev/null
@@ -1,123 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-// this file aims to contains the various representations of rotation/orientation
-// in 2D and 3D space excepted Matrix and Quaternion.
-
-/** \class RotationBase
-  *
-  * \brief Common base class for compact rotation representations
-  *
-  * \param Derived is the derived type, i.e., a rotation type
-  * \param _Dim the dimension of the space
-  */
-template<typename Derived, int _Dim>
-class RotationBase
-{
-  public:
-    enum { Dim = _Dim };
-    /** the scalar type of the coefficients */
-    typedef typename ei_traits<Derived>::Scalar Scalar;
-    
-    /** corresponding linear transformation matrix type */
-    typedef Matrix<Scalar,Dim,Dim> RotationMatrixType;
-
-    inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
-    inline Derived& derived() { return *static_cast<Derived*>(this); }
-
-    /** \returns an equivalent rotation matrix */
-    inline RotationMatrixType toRotationMatrix() const { return derived().toRotationMatrix(); }
-
-    /** \returns the inverse rotation */
-    inline Derived inverse() const { return derived().inverse(); }
-
-    /** \returns the concatenation of the rotation \c *this with a translation \a t */
-    inline Transform<Scalar,Dim> operator*(const Translation<Scalar,Dim>& t) const
-    { return toRotationMatrix() * t; }
-
-    /** \returns the concatenation of the rotation \c *this with a scaling \a s */
-    inline RotationMatrixType operator*(const Scaling<Scalar,Dim>& s) const
-    { return toRotationMatrix() * s; }
-
-    /** \returns the concatenation of the rotation \c *this with an affine transformation \a t */
-    inline Transform<Scalar,Dim> operator*(const Transform<Scalar,Dim>& t) const
-    { return toRotationMatrix() * t; }
-};
-
-/** \geometry_module
-  *
-  * Constructs a Dim x Dim rotation matrix from the rotation \a r
-  */
-template<typename _Scalar, int _Rows, int _Cols, int _Storage, int _MaxRows, int _MaxCols>
-template<typename OtherDerived>
-Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
-::Matrix(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
-{
-  EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(Matrix,int(OtherDerived::Dim),int(OtherDerived::Dim))
-  *this = r.toRotationMatrix();
-}
-
-/** \geometry_module
-  *
-  * Set a Dim x Dim rotation matrix from the rotation \a r
-  */
-template<typename _Scalar, int _Rows, int _Cols, int _Storage, int _MaxRows, int _MaxCols>
-template<typename OtherDerived>
-Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>&
-Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
-::operator=(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
-{
-  EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(Matrix,int(OtherDerived::Dim),int(OtherDerived::Dim))
-  return *this = r.toRotationMatrix();
-}
-
-/** \internal
-  *
-  * Helper function to return an arbitrary rotation object to a rotation matrix.
-  *
-  * \param Scalar the numeric type of the matrix coefficients
-  * \param Dim the dimension of the current space
-  *
-  * It returns a Dim x Dim fixed size matrix.
-  *
-  * Default specializations are provided for:
-  *   - any scalar type (2D),
-  *   - any matrix expression,
-  *   - any type based on RotationBase (e.g., Quaternion, AngleAxis, Rotation2D)
-  *
-  * Currently ei_toRotationMatrix is only used by Transform.
-  *
-  * \sa class Transform, class Rotation2D, class Quaternion, class AngleAxis
-  */
-template<typename Scalar, int Dim>
-static inline Matrix<Scalar,2,2> ei_toRotationMatrix(const Scalar& s)
-{
-  EIGEN_STATIC_ASSERT(Dim==2,YOU_MADE_A_PROGRAMMING_MISTAKE)
-  return Rotation2D<Scalar>(s).toRotationMatrix();
-}
-
-template<typename Scalar, int Dim, typename OtherDerived>
-static inline Matrix<Scalar,Dim,Dim> ei_toRotationMatrix(const RotationBase<OtherDerived,Dim>& r)
-{
-  return r.toRotationMatrix();
-}
-
-template<typename Scalar, int Dim, typename OtherDerived>
-static inline const MatrixBase<OtherDerived>& ei_toRotationMatrix(const MatrixBase<OtherDerived>& mat)
-{
-  EIGEN_STATIC_ASSERT(OtherDerived::RowsAtCompileTime==Dim && OtherDerived::ColsAtCompileTime==Dim,
-    YOU_MADE_A_PROGRAMMING_MISTAKE)
-  return mat;
-}
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/Scaling.h b/Eigen/src/Eigen2Support/Geometry/Scaling.h
deleted file mode 100644
index b8fa6cd..0000000
--- a/Eigen/src/Eigen2Support/Geometry/Scaling.h
+++ /dev/null
@@ -1,167 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class Scaling
-  *
-  * \brief Represents a possibly non uniform scaling transformation
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients.
-  * \param _Dim the  dimension of the space, can be a compile time value or Dynamic
-  *
-  * \note This class is not aimed to be used to store a scaling transformation,
-  * but rather to make easier the constructions and updates of Transform objects.
-  *
-  * \sa class Translation, class Transform
-  */
-template<typename _Scalar, int _Dim>
-class Scaling
-{
-public:
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_Dim)
-  /** dimension of the space */
-  enum { Dim = _Dim };
-  /** the scalar type of the coefficients */
-  typedef _Scalar Scalar;
-  /** corresponding vector type */
-  typedef Matrix<Scalar,Dim,1> VectorType;
-  /** corresponding linear transformation matrix type */
-  typedef Matrix<Scalar,Dim,Dim> LinearMatrixType;
-  /** corresponding translation type */
-  typedef Translation<Scalar,Dim> TranslationType;
-  /** corresponding affine transformation type */
-  typedef Transform<Scalar,Dim> TransformType;
-
-protected:
-
-  VectorType m_coeffs;
-
-public:
-
-  /** Default constructor without initialization. */
-  Scaling() {}
-  /** Constructs and initialize a uniform scaling transformation */
-  explicit inline Scaling(const Scalar& s) { m_coeffs.setConstant(s); }
-  /** 2D only */
-  inline Scaling(const Scalar& sx, const Scalar& sy)
-  {
-    ei_assert(Dim==2);
-    m_coeffs.x() = sx;
-    m_coeffs.y() = sy;
-  }
-  /** 3D only */
-  inline Scaling(const Scalar& sx, const Scalar& sy, const Scalar& sz)
-  {
-    ei_assert(Dim==3);
-    m_coeffs.x() = sx;
-    m_coeffs.y() = sy;
-    m_coeffs.z() = sz;
-  }
-  /** Constructs and initialize the scaling transformation from a vector of scaling coefficients */
-  explicit inline Scaling(const VectorType& coeffs) : m_coeffs(coeffs) {}
-
-  const VectorType& coeffs() const { return m_coeffs; }
-  VectorType& coeffs() { return m_coeffs; }
-
-  /** Concatenates two scaling */
-  inline Scaling operator* (const Scaling& other) const
-  { return Scaling(coeffs().cwise() * other.coeffs()); }
-
-  /** Concatenates a scaling and a translation */
-  inline TransformType operator* (const TranslationType& t) const;
-
-  /** Concatenates a scaling and an affine transformation */
-  inline TransformType operator* (const TransformType& t) const;
-
-  /** Concatenates a scaling and a linear transformation matrix */
-  // TODO returns an expression
-  inline LinearMatrixType operator* (const LinearMatrixType& other) const
-  { return coeffs().asDiagonal() * other; }
-
-  /** Concatenates a linear transformation matrix and a scaling */
-  // TODO returns an expression
-  friend inline LinearMatrixType operator* (const LinearMatrixType& other, const Scaling& s)
-  { return other * s.coeffs().asDiagonal(); }
-
-  template<typename Derived>
-  inline LinearMatrixType operator*(const RotationBase<Derived,Dim>& r) const
-  { return *this * r.toRotationMatrix(); }
-
-  /** Applies scaling to vector */
-  inline VectorType operator* (const VectorType& other) const
-  { return coeffs().asDiagonal() * other; }
-
-  /** \returns the inverse scaling */
-  inline Scaling inverse() const
-  { return Scaling(coeffs().cwise().inverse()); }
-
-  inline Scaling& operator=(const Scaling& other)
-  {
-    m_coeffs = other.m_coeffs;
-    return *this;
-  }
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<Scaling,Scaling<NewScalarType,Dim> >::type cast() const
-  { return typename internal::cast_return_type<Scaling,Scaling<NewScalarType,Dim> >::type(*this); }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit Scaling(const Scaling<OtherScalarType,Dim>& other)
-  { m_coeffs = other.coeffs().template cast<Scalar>(); }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const Scaling& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_coeffs.isApprox(other.m_coeffs, prec); }
-
-};
-
-/** \addtogroup Geometry_Module */
-//@{
-typedef Scaling<float, 2> Scaling2f;
-typedef Scaling<double,2> Scaling2d;
-typedef Scaling<float, 3> Scaling3f;
-typedef Scaling<double,3> Scaling3d;
-//@}
-
-template<typename Scalar, int Dim>
-inline typename Scaling<Scalar,Dim>::TransformType
-Scaling<Scalar,Dim>::operator* (const TranslationType& t) const
-{
-  TransformType res;
-  res.matrix().setZero();
-  res.linear().diagonal() = coeffs();
-  res.translation() = m_coeffs.cwise() * t.vector();
-  res(Dim,Dim) = Scalar(1);
-  return res;
-}
-
-template<typename Scalar, int Dim>
-inline typename Scaling<Scalar,Dim>::TransformType
-Scaling<Scalar,Dim>::operator* (const TransformType& t) const
-{
-  TransformType res = t;
-  res.prescale(m_coeffs);
-  return res;
-}
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/Transform.h b/Eigen/src/Eigen2Support/Geometry/Transform.h
deleted file mode 100644
index fab60b2..0000000
--- a/Eigen/src/Eigen2Support/Geometry/Transform.h
+++ /dev/null
@@ -1,786 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-// Note that we have to pass Dim and HDim because it is not allowed to use a template
-// parameter to define a template specialization. To be more precise, in the following
-// specializations, it is not allowed to use Dim+1 instead of HDim.
-template< typename Other,
-          int Dim,
-          int HDim,
-          int OtherRows=Other::RowsAtCompileTime,
-          int OtherCols=Other::ColsAtCompileTime>
-struct ei_transform_product_impl;
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class Transform
-  *
-  * \brief Represents an homogeneous transformation in a N dimensional space
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  * \param _Dim the dimension of the space
-  *
-  * The homography is internally represented and stored as a (Dim+1)^2 matrix which
-  * is available through the matrix() method.
-  *
-  * Conversion methods from/to Qt's QMatrix and QTransform are available if the
-  * preprocessor token EIGEN_QT_SUPPORT is defined.
-  *
-  * \sa class Matrix, class Quaternion
-  */
-template<typename _Scalar, int _Dim>
-class Transform
-{
-public:
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_Dim==Dynamic ? Dynamic : (_Dim+1)*(_Dim+1))
-  enum {
-    Dim = _Dim,     ///< space dimension in which the transformation holds
-    HDim = _Dim+1   ///< size of a respective homogeneous vector
-  };
-  /** the scalar type of the coefficients */
-  typedef _Scalar Scalar;
-  /** type of the matrix used to represent the transformation */
-  typedef Matrix<Scalar,HDim,HDim> MatrixType;
-  /** type of the matrix used to represent the linear part of the transformation */
-  typedef Matrix<Scalar,Dim,Dim> LinearMatrixType;
-  /** type of read/write reference to the linear part of the transformation */
-  typedef Block<MatrixType,Dim,Dim> LinearPart;
-  /** type of read/write reference to the linear part of the transformation */
-  typedef const Block<const MatrixType,Dim,Dim> ConstLinearPart;
-  /** type of a vector */
-  typedef Matrix<Scalar,Dim,1> VectorType;
-  /** type of a read/write reference to the translation part of the rotation */
-  typedef Block<MatrixType,Dim,1> TranslationPart;
-  /** type of a read/write reference to the translation part of the rotation */
-  typedef const Block<const MatrixType,Dim,1> ConstTranslationPart;
-  /** corresponding translation type */
-  typedef Translation<Scalar,Dim> TranslationType;
-  /** corresponding scaling transformation type */
-  typedef Scaling<Scalar,Dim> ScalingType;
-
-protected:
-
-  MatrixType m_matrix;
-
-public:
-
-  /** Default constructor without initialization of the coefficients. */
-  inline Transform() { }
-
-  inline Transform(const Transform& other)
-  {
-    m_matrix = other.m_matrix;
-  }
-
-  inline explicit Transform(const TranslationType& t) { *this = t; }
-  inline explicit Transform(const ScalingType& s) { *this = s; }
-  template<typename Derived>
-  inline explicit Transform(const RotationBase<Derived, Dim>& r) { *this = r; }
-
-  inline Transform& operator=(const Transform& other)
-  { m_matrix = other.m_matrix; return *this; }
-
-  template<typename OtherDerived, bool BigMatrix> // MSVC 2005 will commit suicide if BigMatrix has a default value
-  struct construct_from_matrix
-  {
-    static inline void run(Transform *transform, const MatrixBase<OtherDerived>& other)
-    {
-      transform->matrix() = other;
-    }
-  };
-
-  template<typename OtherDerived> struct construct_from_matrix<OtherDerived, true>
-  {
-    static inline void run(Transform *transform, const MatrixBase<OtherDerived>& other)
-    {
-      transform->linear() = other;
-      transform->translation().setZero();
-      transform->matrix()(Dim,Dim) = Scalar(1);
-      transform->matrix().template block<1,Dim>(Dim,0).setZero();
-    }
-  };
-
-  /** Constructs and initializes a transformation from a Dim^2 or a (Dim+1)^2 matrix. */
-  template<typename OtherDerived>
-  inline explicit Transform(const MatrixBase<OtherDerived>& other)
-  {
-    construct_from_matrix<OtherDerived, int(OtherDerived::RowsAtCompileTime) == Dim>::run(this, other);
-  }
-
-  /** Set \c *this from a (Dim+1)^2 matrix. */
-  template<typename OtherDerived>
-  inline Transform& operator=(const MatrixBase<OtherDerived>& other)
-  { m_matrix = other; return *this; }
-
-  #ifdef EIGEN_QT_SUPPORT
-  inline Transform(const QMatrix& other);
-  inline Transform& operator=(const QMatrix& other);
-  inline QMatrix toQMatrix(void) const;
-  inline Transform(const QTransform& other);
-  inline Transform& operator=(const QTransform& other);
-  inline QTransform toQTransform(void) const;
-  #endif
-
-  /** shortcut for m_matrix(row,col);
-    * \sa MatrixBase::operaror(int,int) const */
-  inline Scalar operator() (int row, int col) const { return m_matrix(row,col); }
-  /** shortcut for m_matrix(row,col);
-    * \sa MatrixBase::operaror(int,int) */
-  inline Scalar& operator() (int row, int col) { return m_matrix(row,col); }
-
-  /** \returns a read-only expression of the transformation matrix */
-  inline const MatrixType& matrix() const { return m_matrix; }
-  /** \returns a writable expression of the transformation matrix */
-  inline MatrixType& matrix() { return m_matrix; }
-
-  /** \returns a read-only expression of the linear (linear) part of the transformation */
-  inline ConstLinearPart linear() const { return m_matrix.template block<Dim,Dim>(0,0); }
-  /** \returns a writable expression of the linear (linear) part of the transformation */
-  inline LinearPart linear() { return m_matrix.template block<Dim,Dim>(0,0); }
-
-  /** \returns a read-only expression of the translation vector of the transformation */
-  inline ConstTranslationPart translation() const { return m_matrix.template block<Dim,1>(0,Dim); }
-  /** \returns a writable expression of the translation vector of the transformation */
-  inline TranslationPart translation() { return m_matrix.template block<Dim,1>(0,Dim); }
-
-  /** \returns an expression of the product between the transform \c *this and a matrix expression \a other
-  *
-  * The right hand side \a other might be either:
-  * \li a vector of size Dim,
-  * \li an homogeneous vector of size Dim+1,
-  * \li a transformation matrix of size Dim+1 x Dim+1.
-  */
-  // note: this function is defined here because some compilers cannot find the respective declaration
-  template<typename OtherDerived>
-  inline const typename ei_transform_product_impl<OtherDerived,_Dim,_Dim+1>::ResultType
-  operator * (const MatrixBase<OtherDerived> &other) const
-  { return ei_transform_product_impl<OtherDerived,Dim,HDim>::run(*this,other.derived()); }
-
-  /** \returns the product expression of a transformation matrix \a a times a transform \a b
-    * The transformation matrix \a a must have a Dim+1 x Dim+1 sizes. */
-  template<typename OtherDerived>
-  friend inline const typename ProductReturnType<OtherDerived,MatrixType>::Type
-  operator * (const MatrixBase<OtherDerived> &a, const Transform &b)
-  { return a.derived() * b.matrix(); }
-
-  /** Contatenates two transformations */
-  inline const Transform
-  operator * (const Transform& other) const
-  { return Transform(m_matrix * other.matrix()); }
-
-  /** \sa MatrixBase::setIdentity() */
-  void setIdentity() { m_matrix.setIdentity(); }
-  static const typename MatrixType::IdentityReturnType Identity()
-  {
-    return MatrixType::Identity();
-  }
-
-  template<typename OtherDerived>
-  inline Transform& scale(const MatrixBase<OtherDerived> &other);
-
-  template<typename OtherDerived>
-  inline Transform& prescale(const MatrixBase<OtherDerived> &other);
-
-  inline Transform& scale(Scalar s);
-  inline Transform& prescale(Scalar s);
-
-  template<typename OtherDerived>
-  inline Transform& translate(const MatrixBase<OtherDerived> &other);
-
-  template<typename OtherDerived>
-  inline Transform& pretranslate(const MatrixBase<OtherDerived> &other);
-
-  template<typename RotationType>
-  inline Transform& rotate(const RotationType& rotation);
-
-  template<typename RotationType>
-  inline Transform& prerotate(const RotationType& rotation);
-
-  Transform& shear(Scalar sx, Scalar sy);
-  Transform& preshear(Scalar sx, Scalar sy);
-
-  inline Transform& operator=(const TranslationType& t);
-  inline Transform& operator*=(const TranslationType& t) { return translate(t.vector()); }
-  inline Transform operator*(const TranslationType& t) const;
-
-  inline Transform& operator=(const ScalingType& t);
-  inline Transform& operator*=(const ScalingType& s) { return scale(s.coeffs()); }
-  inline Transform operator*(const ScalingType& s) const;
-  friend inline Transform operator*(const LinearMatrixType& mat, const Transform& t)
-  {
-    Transform res = t;
-    res.matrix().row(Dim) = t.matrix().row(Dim);
-    res.matrix().template block<Dim,HDim>(0,0) = (mat * t.matrix().template block<Dim,HDim>(0,0)).lazy();
-    return res;
-  }
-
-  template<typename Derived>
-  inline Transform& operator=(const RotationBase<Derived,Dim>& r);
-  template<typename Derived>
-  inline Transform& operator*=(const RotationBase<Derived,Dim>& r) { return rotate(r.toRotationMatrix()); }
-  template<typename Derived>
-  inline Transform operator*(const RotationBase<Derived,Dim>& r) const;
-
-  LinearMatrixType rotation() const;
-  template<typename RotationMatrixType, typename ScalingMatrixType>
-  void computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const;
-  template<typename ScalingMatrixType, typename RotationMatrixType>
-  void computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const;
-
-  template<typename PositionDerived, typename OrientationType, typename ScaleDerived>
-  Transform& fromPositionOrientationScale(const MatrixBase<PositionDerived> &position,
-    const OrientationType& orientation, const MatrixBase<ScaleDerived> &scale);
-
-  inline const MatrixType inverse(TransformTraits traits = Affine) const;
-
-  /** \returns a const pointer to the column major internal matrix */
-  const Scalar* data() const { return m_matrix.data(); }
-  /** \returns a non-const pointer to the column major internal matrix */
-  Scalar* data() { return m_matrix.data(); }
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim> >::type cast() const
-  { return typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim> >::type(*this); }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit Transform(const Transform<OtherScalarType,Dim>& other)
-  { m_matrix = other.matrix().template cast<Scalar>(); }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const Transform& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_matrix.isApprox(other.m_matrix, prec); }
-
-  #ifdef EIGEN_TRANSFORM_PLUGIN
-  #include EIGEN_TRANSFORM_PLUGIN
-  #endif
-
-protected:
-
-};
-
-/** \ingroup Geometry_Module */
-typedef Transform<float,2> Transform2f;
-/** \ingroup Geometry_Module */
-typedef Transform<float,3> Transform3f;
-/** \ingroup Geometry_Module */
-typedef Transform<double,2> Transform2d;
-/** \ingroup Geometry_Module */
-typedef Transform<double,3> Transform3d;
-
-/**************************
-*** Optional QT support ***
-**************************/
-
-#ifdef EIGEN_QT_SUPPORT
-/** Initialises \c *this from a QMatrix assuming the dimension is 2.
-  *
-  * This function is available only if the token EIGEN_QT_SUPPORT is defined.
-  */
-template<typename Scalar, int Dim>
-Transform<Scalar,Dim>::Transform(const QMatrix& other)
-{
-  *this = other;
-}
-
-/** Set \c *this from a QMatrix assuming the dimension is 2.
-  *
-  * This function is available only if the token EIGEN_QT_SUPPORT is defined.
-  */
-template<typename Scalar, int Dim>
-Transform<Scalar,Dim>& Transform<Scalar,Dim>::operator=(const QMatrix& other)
-{
-  EIGEN_STATIC_ASSERT(Dim==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  m_matrix << other.m11(), other.m21(), other.dx(),
-              other.m12(), other.m22(), other.dy(),
-              0, 0, 1;
-   return *this;
-}
-
-/** \returns a QMatrix from \c *this assuming the dimension is 2.
-  *
-  * \warning this convertion might loss data if \c *this is not affine
-  *
-  * This function is available only if the token EIGEN_QT_SUPPORT is defined.
-  */
-template<typename Scalar, int Dim>
-QMatrix Transform<Scalar,Dim>::toQMatrix(void) const
-{
-  EIGEN_STATIC_ASSERT(Dim==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  return QMatrix(m_matrix.coeff(0,0), m_matrix.coeff(1,0),
-                 m_matrix.coeff(0,1), m_matrix.coeff(1,1),
-                 m_matrix.coeff(0,2), m_matrix.coeff(1,2));
-}
-
-/** Initialises \c *this from a QTransform assuming the dimension is 2.
-  *
-  * This function is available only if the token EIGEN_QT_SUPPORT is defined.
-  */
-template<typename Scalar, int Dim>
-Transform<Scalar,Dim>::Transform(const QTransform& other)
-{
-  *this = other;
-}
-
-/** Set \c *this from a QTransform assuming the dimension is 2.
-  *
-  * This function is available only if the token EIGEN_QT_SUPPORT is defined.
-  */
-template<typename Scalar, int Dim>
-Transform<Scalar,Dim>& Transform<Scalar,Dim>::operator=(const QTransform& other)
-{
-  EIGEN_STATIC_ASSERT(Dim==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  m_matrix << other.m11(), other.m21(), other.dx(),
-              other.m12(), other.m22(), other.dy(),
-              other.m13(), other.m23(), other.m33();
-   return *this;
-}
-
-/** \returns a QTransform from \c *this assuming the dimension is 2.
-  *
-  * This function is available only if the token EIGEN_QT_SUPPORT is defined.
-  */
-template<typename Scalar, int Dim>
-QTransform Transform<Scalar,Dim>::toQTransform(void) const
-{
-  EIGEN_STATIC_ASSERT(Dim==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  return QTransform(m_matrix.coeff(0,0), m_matrix.coeff(1,0), m_matrix.coeff(2,0),
-                    m_matrix.coeff(0,1), m_matrix.coeff(1,1), m_matrix.coeff(2,1),
-                    m_matrix.coeff(0,2), m_matrix.coeff(1,2), m_matrix.coeff(2,2));
-}
-#endif
-
-/*********************
-*** Procedural API ***
-*********************/
-
-/** Applies on the right the non uniform scale transformation represented
-  * by the vector \a other to \c *this and returns a reference to \c *this.
-  * \sa prescale()
-  */
-template<typename Scalar, int Dim>
-template<typename OtherDerived>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::scale(const MatrixBase<OtherDerived> &other)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
-  linear() = (linear() * other.asDiagonal()).lazy();
-  return *this;
-}
-
-/** Applies on the right a uniform scale of a factor \a c to \c *this
-  * and returns a reference to \c *this.
-  * \sa prescale(Scalar)
-  */
-template<typename Scalar, int Dim>
-inline Transform<Scalar,Dim>& Transform<Scalar,Dim>::scale(Scalar s)
-{
-  linear() *= s;
-  return *this;
-}
-
-/** Applies on the left the non uniform scale transformation represented
-  * by the vector \a other to \c *this and returns a reference to \c *this.
-  * \sa scale()
-  */
-template<typename Scalar, int Dim>
-template<typename OtherDerived>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::prescale(const MatrixBase<OtherDerived> &other)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
-  m_matrix.template block<Dim,HDim>(0,0) = (other.asDiagonal() * m_matrix.template block<Dim,HDim>(0,0)).lazy();
-  return *this;
-}
-
-/** Applies on the left a uniform scale of a factor \a c to \c *this
-  * and returns a reference to \c *this.
-  * \sa scale(Scalar)
-  */
-template<typename Scalar, int Dim>
-inline Transform<Scalar,Dim>& Transform<Scalar,Dim>::prescale(Scalar s)
-{
-  m_matrix.template corner<Dim,HDim>(TopLeft) *= s;
-  return *this;
-}
-
-/** Applies on the right the translation matrix represented by the vector \a other
-  * to \c *this and returns a reference to \c *this.
-  * \sa pretranslate()
-  */
-template<typename Scalar, int Dim>
-template<typename OtherDerived>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::translate(const MatrixBase<OtherDerived> &other)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
-  translation() += linear() * other;
-  return *this;
-}
-
-/** Applies on the left the translation matrix represented by the vector \a other
-  * to \c *this and returns a reference to \c *this.
-  * \sa translate()
-  */
-template<typename Scalar, int Dim>
-template<typename OtherDerived>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::pretranslate(const MatrixBase<OtherDerived> &other)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
-  translation() += other;
-  return *this;
-}
-
-/** Applies on the right the rotation represented by the rotation \a rotation
-  * to \c *this and returns a reference to \c *this.
-  *
-  * The template parameter \a RotationType is the type of the rotation which
-  * must be known by ei_toRotationMatrix<>.
-  *
-  * Natively supported types includes:
-  *   - any scalar (2D),
-  *   - a Dim x Dim matrix expression,
-  *   - a Quaternion (3D),
-  *   - a AngleAxis (3D)
-  *
-  * This mechanism is easily extendable to support user types such as Euler angles,
-  * or a pair of Quaternion for 4D rotations.
-  *
-  * \sa rotate(Scalar), class Quaternion, class AngleAxis, prerotate(RotationType)
-  */
-template<typename Scalar, int Dim>
-template<typename RotationType>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::rotate(const RotationType& rotation)
-{
-  linear() *= ei_toRotationMatrix<Scalar,Dim>(rotation);
-  return *this;
-}
-
-/** Applies on the left the rotation represented by the rotation \a rotation
-  * to \c *this and returns a reference to \c *this.
-  *
-  * See rotate() for further details.
-  *
-  * \sa rotate()
-  */
-template<typename Scalar, int Dim>
-template<typename RotationType>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::prerotate(const RotationType& rotation)
-{
-  m_matrix.template block<Dim,HDim>(0,0) = ei_toRotationMatrix<Scalar,Dim>(rotation)
-                                         * m_matrix.template block<Dim,HDim>(0,0);
-  return *this;
-}
-
-/** Applies on the right the shear transformation represented
-  * by the vector \a other to \c *this and returns a reference to \c *this.
-  * \warning 2D only.
-  * \sa preshear()
-  */
-template<typename Scalar, int Dim>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::shear(Scalar sx, Scalar sy)
-{
-  EIGEN_STATIC_ASSERT(int(Dim)==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  VectorType tmp = linear().col(0)*sy + linear().col(1);
-  linear() << linear().col(0) + linear().col(1)*sx, tmp;
-  return *this;
-}
-
-/** Applies on the left the shear transformation represented
-  * by the vector \a other to \c *this and returns a reference to \c *this.
-  * \warning 2D only.
-  * \sa shear()
-  */
-template<typename Scalar, int Dim>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::preshear(Scalar sx, Scalar sy)
-{
-  EIGEN_STATIC_ASSERT(int(Dim)==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  m_matrix.template block<Dim,HDim>(0,0) = LinearMatrixType(1, sx, sy, 1) * m_matrix.template block<Dim,HDim>(0,0);
-  return *this;
-}
-
-/******************************************************
-*** Scaling, Translation and Rotation compatibility ***
-******************************************************/
-
-template<typename Scalar, int Dim>
-inline Transform<Scalar,Dim>& Transform<Scalar,Dim>::operator=(const TranslationType& t)
-{
-  linear().setIdentity();
-  translation() = t.vector();
-  m_matrix.template block<1,Dim>(Dim,0).setZero();
-  m_matrix(Dim,Dim) = Scalar(1);
-  return *this;
-}
-
-template<typename Scalar, int Dim>
-inline Transform<Scalar,Dim> Transform<Scalar,Dim>::operator*(const TranslationType& t) const
-{
-  Transform res = *this;
-  res.translate(t.vector());
-  return res;
-}
-
-template<typename Scalar, int Dim>
-inline Transform<Scalar,Dim>& Transform<Scalar,Dim>::operator=(const ScalingType& s)
-{
-  m_matrix.setZero();
-  linear().diagonal() = s.coeffs();
-  m_matrix.coeffRef(Dim,Dim) = Scalar(1);
-  return *this;
-}
-
-template<typename Scalar, int Dim>
-inline Transform<Scalar,Dim> Transform<Scalar,Dim>::operator*(const ScalingType& s) const
-{
-  Transform res = *this;
-  res.scale(s.coeffs());
-  return res;
-}
-
-template<typename Scalar, int Dim>
-template<typename Derived>
-inline Transform<Scalar,Dim>& Transform<Scalar,Dim>::operator=(const RotationBase<Derived,Dim>& r)
-{
-  linear() = ei_toRotationMatrix<Scalar,Dim>(r);
-  translation().setZero();
-  m_matrix.template block<1,Dim>(Dim,0).setZero();
-  m_matrix.coeffRef(Dim,Dim) = Scalar(1);
-  return *this;
-}
-
-template<typename Scalar, int Dim>
-template<typename Derived>
-inline Transform<Scalar,Dim> Transform<Scalar,Dim>::operator*(const RotationBase<Derived,Dim>& r) const
-{
-  Transform res = *this;
-  res.rotate(r.derived());
-  return res;
-}
-
-/************************
-*** Special functions ***
-************************/
-
-/** \returns the rotation part of the transformation
-  * \nonstableyet
-  *
-  * \svd_module
-  *
-  * \sa computeRotationScaling(), computeScalingRotation(), class SVD
-  */
-template<typename Scalar, int Dim>
-typename Transform<Scalar,Dim>::LinearMatrixType
-Transform<Scalar,Dim>::rotation() const
-{
-  LinearMatrixType result;
-  computeRotationScaling(&result, (LinearMatrixType*)0);
-  return result;
-}
-
-
-/** decomposes the linear part of the transformation as a product rotation x scaling, the scaling being
-  * not necessarily positive.
-  *
-  * If either pointer is zero, the corresponding computation is skipped.
-  *
-  * \nonstableyet
-  *
-  * \svd_module
-  *
-  * \sa computeScalingRotation(), rotation(), class SVD
-  */
-template<typename Scalar, int Dim>
-template<typename RotationMatrixType, typename ScalingMatrixType>
-void Transform<Scalar,Dim>::computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const
-{
-  JacobiSVD<LinearMatrixType> svd(linear(), ComputeFullU|ComputeFullV);
-  Scalar x = (svd.matrixU() * svd.matrixV().adjoint()).determinant(); // so x has absolute value 1
-  Matrix<Scalar, Dim, 1> sv(svd.singularValues());
-  sv.coeffRef(0) *= x;
-  if(scaling)
-  {
-    scaling->noalias() = svd.matrixV() * sv.asDiagonal() * svd.matrixV().adjoint();
-  }
-  if(rotation)
-  {
-    LinearMatrixType m(svd.matrixU());
-    m.col(0) /= x;
-    rotation->noalias() = m * svd.matrixV().adjoint();
-  }
-}
-
-/** decomposes the linear part of the transformation as a product rotation x scaling, the scaling being
-  * not necessarily positive.
-  *
-  * If either pointer is zero, the corresponding computation is skipped.
-  *
-  * \nonstableyet
-  *
-  * \svd_module
-  *
-  * \sa computeRotationScaling(), rotation(), class SVD
-  */
-template<typename Scalar, int Dim>
-template<typename ScalingMatrixType, typename RotationMatrixType>
-void Transform<Scalar,Dim>::computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const
-{
-  JacobiSVD<LinearMatrixType> svd(linear(), ComputeFullU|ComputeFullV);
-  Scalar x = (svd.matrixU() * svd.matrixV().adjoint()).determinant(); // so x has absolute value 1
-  Matrix<Scalar, Dim, 1> sv(svd.singularValues());
-  sv.coeffRef(0) *= x;
-  if(scaling)
-  {
-    scaling->noalias() = svd.matrixU() * sv.asDiagonal() * svd.matrixU().adjoint();
-  }
-  if(rotation)
-  {
-    LinearMatrixType m(svd.matrixU());
-    m.col(0) /= x;
-    rotation->noalias() = m * svd.matrixV().adjoint();
-  }
-}
-
-/** Convenient method to set \c *this from a position, orientation and scale
-  * of a 3D object.
-  */
-template<typename Scalar, int Dim>
-template<typename PositionDerived, typename OrientationType, typename ScaleDerived>
-Transform<Scalar,Dim>&
-Transform<Scalar,Dim>::fromPositionOrientationScale(const MatrixBase<PositionDerived> &position,
-  const OrientationType& orientation, const MatrixBase<ScaleDerived> &scale)
-{
-  linear() = ei_toRotationMatrix<Scalar,Dim>(orientation);
-  linear() *= scale.asDiagonal();
-  translation() = position;
-  m_matrix.template block<1,Dim>(Dim,0).setZero();
-  m_matrix(Dim,Dim) = Scalar(1);
-  return *this;
-}
-
-/** \nonstableyet
-  *
-  * \returns the inverse transformation matrix according to some given knowledge
-  * on \c *this.
-  *
-  * \param traits allows to optimize the inversion process when the transformion
-  * is known to be not a general transformation. The possible values are:
-  *  - Projective if the transformation is not necessarily affine, i.e., if the
-  *    last row is not guaranteed to be [0 ... 0 1]
-  *  - Affine is the default, the last row is assumed to be [0 ... 0 1]
-  *  - Isometry if the transformation is only a concatenations of translations
-  *    and rotations.
-  *
-  * \warning unless \a traits is always set to NoShear or NoScaling, this function
-  * requires the generic inverse method of MatrixBase defined in the LU module. If
-  * you forget to include this module, then you will get hard to debug linking errors.
-  *
-  * \sa MatrixBase::inverse()
-  */
-template<typename Scalar, int Dim>
-inline const typename Transform<Scalar,Dim>::MatrixType
-Transform<Scalar,Dim>::inverse(TransformTraits traits) const
-{
-  if (traits == Projective)
-  {
-    return m_matrix.inverse();
-  }
-  else
-  {
-    MatrixType res;
-    if (traits == Affine)
-    {
-      res.template corner<Dim,Dim>(TopLeft) = linear().inverse();
-    }
-    else if (traits == Isometry)
-    {
-      res.template corner<Dim,Dim>(TopLeft) = linear().transpose();
-    }
-    else
-    {
-      ei_assert("invalid traits value in Transform::inverse()");
-    }
-    // translation and remaining parts
-    res.template corner<Dim,1>(TopRight) = - res.template corner<Dim,Dim>(TopLeft) * translation();
-    res.template corner<1,Dim>(BottomLeft).setZero();
-    res.coeffRef(Dim,Dim) = Scalar(1);
-    return res;
-  }
-}
-
-/*****************************************************
-*** Specializations of operator* with a MatrixBase ***
-*****************************************************/
-
-template<typename Other, int Dim, int HDim>
-struct ei_transform_product_impl<Other,Dim,HDim, HDim,HDim>
-{
-  typedef Transform<typename Other::Scalar,Dim> TransformType;
-  typedef typename TransformType::MatrixType MatrixType;
-  typedef typename ProductReturnType<MatrixType,Other>::Type ResultType;
-  static ResultType run(const TransformType& tr, const Other& other)
-  { return tr.matrix() * other; }
-};
-
-template<typename Other, int Dim, int HDim>
-struct ei_transform_product_impl<Other,Dim,HDim, Dim,Dim>
-{
-  typedef Transform<typename Other::Scalar,Dim> TransformType;
-  typedef typename TransformType::MatrixType MatrixType;
-  typedef TransformType ResultType;
-  static ResultType run(const TransformType& tr, const Other& other)
-  {
-    TransformType res;
-    res.translation() = tr.translation();
-    res.matrix().row(Dim) = tr.matrix().row(Dim);
-    res.linear() = (tr.linear() * other).lazy();
-    return res;
-  }
-};
-
-template<typename Other, int Dim, int HDim>
-struct ei_transform_product_impl<Other,Dim,HDim, HDim,1>
-{
-  typedef Transform<typename Other::Scalar,Dim> TransformType;
-  typedef typename TransformType::MatrixType MatrixType;
-  typedef typename ProductReturnType<MatrixType,Other>::Type ResultType;
-  static ResultType run(const TransformType& tr, const Other& other)
-  { return tr.matrix() * other; }
-};
-
-template<typename Other, int Dim, int HDim>
-struct ei_transform_product_impl<Other,Dim,HDim, Dim,1>
-{
-  typedef typename Other::Scalar Scalar;
-  typedef Transform<Scalar,Dim> TransformType;
-  typedef Matrix<Scalar,Dim,1> ResultType;
-  static ResultType run(const TransformType& tr, const Other& other)
-  { return ((tr.linear() * other) + tr.translation())
-          * (Scalar(1) / ( (tr.matrix().template block<1,Dim>(Dim,0) * other).coeff(0) + tr.matrix().coeff(Dim,Dim))); }
-};
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/Geometry/Translation.h b/Eigen/src/Eigen2Support/Geometry/Translation.h
deleted file mode 100644
index 2b9859f..0000000
--- a/Eigen/src/Eigen2Support/Geometry/Translation.h
+++ /dev/null
@@ -1,184 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-// no include guard, we'll include this twice from All.h from Eigen2Support, and it's internal anyway
-
-namespace Eigen { 
-
-/** \geometry_module \ingroup Geometry_Module
-  *
-  * \class Translation
-  *
-  * \brief Represents a translation transformation
-  *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients.
-  * \param _Dim the  dimension of the space, can be a compile time value or Dynamic
-  *
-  * \note This class is not aimed to be used to store a translation transformation,
-  * but rather to make easier the constructions and updates of Transform objects.
-  *
-  * \sa class Scaling, class Transform
-  */
-template<typename _Scalar, int _Dim>
-class Translation
-{
-public:
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF_VECTORIZABLE_FIXED_SIZE(_Scalar,_Dim)
-  /** dimension of the space */
-  enum { Dim = _Dim };
-  /** the scalar type of the coefficients */
-  typedef _Scalar Scalar;
-  /** corresponding vector type */
-  typedef Matrix<Scalar,Dim,1> VectorType;
-  /** corresponding linear transformation matrix type */
-  typedef Matrix<Scalar,Dim,Dim> LinearMatrixType;
-  /** corresponding scaling transformation type */
-  typedef Scaling<Scalar,Dim> ScalingType;
-  /** corresponding affine transformation type */
-  typedef Transform<Scalar,Dim> TransformType;
-
-protected:
-
-  VectorType m_coeffs;
-
-public:
-
-  /** Default constructor without initialization. */
-  Translation() {}
-  /**  */
-  inline Translation(const Scalar& sx, const Scalar& sy)
-  {
-    ei_assert(Dim==2);
-    m_coeffs.x() = sx;
-    m_coeffs.y() = sy;
-  }
-  /**  */
-  inline Translation(const Scalar& sx, const Scalar& sy, const Scalar& sz)
-  {
-    ei_assert(Dim==3);
-    m_coeffs.x() = sx;
-    m_coeffs.y() = sy;
-    m_coeffs.z() = sz;
-  }
-  /** Constructs and initialize the scaling transformation from a vector of scaling coefficients */
-  explicit inline Translation(const VectorType& vector) : m_coeffs(vector) {}
-
-  const VectorType& vector() const { return m_coeffs; }
-  VectorType& vector() { return m_coeffs; }
-
-  /** Concatenates two translation */
-  inline Translation operator* (const Translation& other) const
-  { return Translation(m_coeffs + other.m_coeffs); }
-
-  /** Concatenates a translation and a scaling */
-  inline TransformType operator* (const ScalingType& other) const;
-
-  /** Concatenates a translation and a linear transformation */
-  inline TransformType operator* (const LinearMatrixType& linear) const;
-
-  template<typename Derived>
-  inline TransformType operator*(const RotationBase<Derived,Dim>& r) const
-  { return *this * r.toRotationMatrix(); }
-
-  /** Concatenates a linear transformation and a translation */
-  // its a nightmare to define a templated friend function outside its declaration
-  friend inline TransformType operator* (const LinearMatrixType& linear, const Translation& t)
-  {
-    TransformType res;
-    res.matrix().setZero();
-    res.linear() = linear;
-    res.translation() = linear * t.m_coeffs;
-    res.matrix().row(Dim).setZero();
-    res(Dim,Dim) = Scalar(1);
-    return res;
-  }
-
-  /** Concatenates a translation and an affine transformation */
-  inline TransformType operator* (const TransformType& t) const;
-
-  /** Applies translation to vector */
-  inline VectorType operator* (const VectorType& other) const
-  { return m_coeffs + other; }
-
-  /** \returns the inverse translation (opposite) */
-  Translation inverse() const { return Translation(-m_coeffs); }
-
-  Translation& operator=(const Translation& other)
-  {
-    m_coeffs = other.m_coeffs;
-    return *this;
-  }
-
-  /** \returns \c *this with scalar type casted to \a NewScalarType
-    *
-    * Note that if \a NewScalarType is equal to the current scalar type of \c *this
-    * then this function smartly returns a const reference to \c *this.
-    */
-  template<typename NewScalarType>
-  inline typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type cast() const
-  { return typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type(*this); }
-
-  /** Copy constructor with scalar type conversion */
-  template<typename OtherScalarType>
-  inline explicit Translation(const Translation<OtherScalarType,Dim>& other)
-  { m_coeffs = other.vector().template cast<Scalar>(); }
-
-  /** \returns \c true if \c *this is approximately equal to \a other, within the precision
-    * determined by \a prec.
-    *
-    * \sa MatrixBase::isApprox() */
-  bool isApprox(const Translation& other, typename NumTraits<Scalar>::Real prec = precision<Scalar>()) const
-  { return m_coeffs.isApprox(other.m_coeffs, prec); }
-
-};
-
-/** \addtogroup Geometry_Module */
-//@{
-typedef Translation<float, 2> Translation2f;
-typedef Translation<double,2> Translation2d;
-typedef Translation<float, 3> Translation3f;
-typedef Translation<double,3> Translation3d;
-//@}
-
-
-template<typename Scalar, int Dim>
-inline typename Translation<Scalar,Dim>::TransformType
-Translation<Scalar,Dim>::operator* (const ScalingType& other) const
-{
-  TransformType res;
-  res.matrix().setZero();
-  res.linear().diagonal() = other.coeffs();
-  res.translation() = m_coeffs;
-  res(Dim,Dim) = Scalar(1);
-  return res;
-}
-
-template<typename Scalar, int Dim>
-inline typename Translation<Scalar,Dim>::TransformType
-Translation<Scalar,Dim>::operator* (const LinearMatrixType& linear) const
-{
-  TransformType res;
-  res.matrix().setZero();
-  res.linear() = linear;
-  res.translation() = m_coeffs;
-  res.matrix().row(Dim).setZero();
-  res(Dim,Dim) = Scalar(1);
-  return res;
-}
-
-template<typename Scalar, int Dim>
-inline typename Translation<Scalar,Dim>::TransformType
-Translation<Scalar,Dim>::operator* (const TransformType& t) const
-{
-  TransformType res = t;
-  res.pretranslate(m_coeffs);
-  return res;
-}
-
-} // end namespace Eigen
diff --git a/Eigen/src/Eigen2Support/LU.h b/Eigen/src/Eigen2Support/LU.h
deleted file mode 100644
index 49f19ad..0000000
--- a/Eigen/src/Eigen2Support/LU.h
+++ /dev/null
@@ -1,120 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_LU_H
-#define EIGEN2_LU_H
-
-namespace Eigen { 
-
-template<typename MatrixType>
-class LU : public FullPivLU<MatrixType>
-{
-  public:
-
-    typedef typename MatrixType::Scalar Scalar;
-    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef Matrix<int, 1, MatrixType::ColsAtCompileTime, MatrixType::Options, 1, MatrixType::MaxColsAtCompileTime> IntRowVectorType;
-    typedef Matrix<int, MatrixType::RowsAtCompileTime, 1, MatrixType::Options, MatrixType::MaxRowsAtCompileTime, 1> IntColVectorType;
-    typedef Matrix<Scalar, 1, MatrixType::ColsAtCompileTime, MatrixType::Options, 1, MatrixType::MaxColsAtCompileTime> RowVectorType;
-    typedef Matrix<Scalar, MatrixType::RowsAtCompileTime, 1, MatrixType::Options, MatrixType::MaxRowsAtCompileTime, 1> ColVectorType;
-
-    typedef Matrix<typename MatrixType::Scalar,
-                  MatrixType::ColsAtCompileTime, // the number of rows in the "kernel matrix" is the number of cols of the original matrix
-                                                 // so that the product "matrix * kernel = zero" makes sense
-                  Dynamic,                       // we don't know at compile-time the dimension of the kernel
-                  MatrixType::Options,
-                  MatrixType::MaxColsAtCompileTime, // see explanation for 2nd template parameter
-                  MatrixType::MaxColsAtCompileTime // the kernel is a subspace of the domain space, whose dimension is the number
-                                                   // of columns of the original matrix
-    > KernelResultType;
-
-    typedef Matrix<typename MatrixType::Scalar,
-                   MatrixType::RowsAtCompileTime, // the image is a subspace of the destination space, whose dimension is the number
-                                                  // of rows of the original matrix
-                   Dynamic,                       // we don't know at compile time the dimension of the image (the rank)
-                   MatrixType::Options,
-                   MatrixType::MaxRowsAtCompileTime, // the image matrix will consist of columns from the original matrix,
-                   MatrixType::MaxColsAtCompileTime  // so it has the same number of rows and at most as many columns.
-    > ImageResultType;
-
-    typedef FullPivLU<MatrixType> Base;
-
-    template<typename T>
-    explicit LU(const T& t) : Base(t), m_originalMatrix(t) {}
-
-    template<typename OtherDerived, typename ResultType>
-    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
-    {
-      *result = static_cast<const Base*>(this)->solve(b);
-      return true;
-    }
-
-    template<typename ResultType>
-    inline void computeInverse(ResultType *result) const
-    {
-      solve(MatrixType::Identity(this->rows(), this->cols()), result);
-    }
-    
-    template<typename KernelMatrixType>
-    void computeKernel(KernelMatrixType *result) const
-    {
-      *result = static_cast<const Base*>(this)->kernel();
-    }
-    
-    template<typename ImageMatrixType>
-    void computeImage(ImageMatrixType *result) const
-    {
-      *result = static_cast<const Base*>(this)->image(m_originalMatrix);
-    }
-    
-    const ImageResultType image() const
-    {
-      return static_cast<const Base*>(this)->image(m_originalMatrix);
-    }
-    
-    const MatrixType& m_originalMatrix;
-};
-
-#if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS
-/** \lu_module
-  *
-  * Synonym of partialPivLu().
-  *
-  * \return the partial-pivoting LU decomposition of \c *this.
-  *
-  * \sa class PartialPivLU
-  */
-template<typename Derived>
-inline const LU<typename MatrixBase<Derived>::PlainObject>
-MatrixBase<Derived>::lu() const
-{
-  return LU<PlainObject>(eval());
-}
-#endif
-
-#ifdef EIGEN2_SUPPORT
-/** \lu_module
-  *
-  * Synonym of partialPivLu().
-  *
-  * \return the partial-pivoting LU decomposition of \c *this.
-  *
-  * \sa class PartialPivLU
-  */
-template<typename Derived>
-inline const LU<typename MatrixBase<Derived>::PlainObject>
-MatrixBase<Derived>::eigen2_lu() const
-{
-  return LU<PlainObject>(eval());
-}
-#endif
-
-} // end namespace Eigen
-
-#endif // EIGEN2_LU_H
diff --git a/Eigen/src/Eigen2Support/Lazy.h b/Eigen/src/Eigen2Support/Lazy.h
deleted file mode 100644
index 593fc78..0000000
--- a/Eigen/src/Eigen2Support/Lazy.h
+++ /dev/null
@@ -1,71 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_LAZY_H
-#define EIGEN_LAZY_H
-
-namespace Eigen { 
-
-/** \deprecated it is only used by lazy() which is deprecated
-  *
-  * \returns an expression of *this with added flags
-  *
-  * Example: \include MatrixBase_marked.cpp
-  * Output: \verbinclude MatrixBase_marked.out
-  *
-  * \sa class Flagged, extract(), part()
-  */
-template<typename Derived>
-template<unsigned int Added>
-inline const Flagged<Derived, Added, 0>
-MatrixBase<Derived>::marked() const
-{
-  return derived();
-}
-
-/** \deprecated use MatrixBase::noalias()
-  *
-  * \returns an expression of *this with the EvalBeforeAssigningBit flag removed.
-  *
-  * Example: \include MatrixBase_lazy.cpp
-  * Output: \verbinclude MatrixBase_lazy.out
-  *
-  * \sa class Flagged, marked()
-  */
-template<typename Derived>
-inline const Flagged<Derived, 0, EvalBeforeAssigningBit>
-MatrixBase<Derived>::lazy() const
-{
-  return derived();
-}
-
-
-/** \internal
-  * Overloaded to perform an efficient C += (A*B).lazy() */
-template<typename Derived>
-template<typename ProductDerived, typename Lhs, typename Rhs>
-Derived& MatrixBase<Derived>::operator+=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                                       EvalBeforeAssigningBit>& other)
-{
-  other._expression().derived().addTo(derived()); return derived();
-}
-
-/** \internal
-  * Overloaded to perform an efficient C -= (A*B).lazy() */
-template<typename Derived>
-template<typename ProductDerived, typename Lhs, typename Rhs>
-Derived& MatrixBase<Derived>::operator-=(const Flagged<ProductBase<ProductDerived, Lhs,Rhs>, 0,
-                                                       EvalBeforeAssigningBit>& other)
-{
-  other._expression().derived().subTo(derived()); return derived();
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_LAZY_H
diff --git a/Eigen/src/Eigen2Support/LeastSquares.h b/Eigen/src/Eigen2Support/LeastSquares.h
deleted file mode 100644
index 7992d49..0000000
--- a/Eigen/src/Eigen2Support/LeastSquares.h
+++ /dev/null
@@ -1,169 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2006-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_LEASTSQUARES_H
-#define EIGEN2_LEASTSQUARES_H
-
-namespace Eigen { 
-
-/** \ingroup LeastSquares_Module
-  *
-  * \leastsquares_module
-  *
-  * For a set of points, this function tries to express
-  * one of the coords as a linear (affine) function of the other coords.
-  *
-  * This is best explained by an example. This function works in full
-  * generality, for points in a space of arbitrary dimension, and also over
-  * the complex numbers, but for this example we will work in dimension 3
-  * over the real numbers (doubles).
-  *
-  * So let us work with the following set of 5 points given by their
-  * \f$(x,y,z)\f$ coordinates:
-  * @code
-    Vector3d points[5];
-    points[0] = Vector3d( 3.02, 6.89, -4.32 );
-    points[1] = Vector3d( 2.01, 5.39, -3.79 );
-    points[2] = Vector3d( 2.41, 6.01, -4.01 );
-    points[3] = Vector3d( 2.09, 5.55, -3.86 );
-    points[4] = Vector3d( 2.58, 6.32, -4.10 );
-  * @endcode
-  * Suppose that we want to express the second coordinate (\f$y\f$) as a linear
-  * expression in \f$x\f$ and \f$z\f$, that is,
-  * \f[ y=ax+bz+c \f]
-  * for some constants \f$a,b,c\f$. Thus, we want to find the best possible
-  * constants \f$a,b,c\f$ so that the plane of equation \f$y=ax+bz+c\f$ fits
-  * best the five above points. To do that, call this function as follows:
-  * @code
-    Vector3d coeffs; // will store the coefficients a, b, c
-    linearRegression(
-      5,
-      &points,
-      &coeffs,
-      1 // the coord to express as a function of
-        // the other ones. 0 means x, 1 means y, 2 means z.
-    );
-  * @endcode
-  * Now the vector \a coeffs is approximately
-  * \f$( 0.495 ,  -1.927 ,  -2.906 )\f$.
-  * Thus, we get \f$a=0.495, b = -1.927, c = -2.906\f$. Let us check for
-  * instance how near points[0] is from the plane of equation \f$y=ax+bz+c\f$.
-  * Looking at the coords of points[0], we see that:
-  * \f[ax+bz+c = 0.495 * 3.02 + (-1.927) * (-4.32) + (-2.906) = 6.91.\f]
-  * On the other hand, we have \f$y=6.89\f$. We see that the values
-  * \f$6.91\f$ and \f$6.89\f$
-  * are near, so points[0] is very near the plane of equation \f$y=ax+bz+c\f$.
-  *
-  * Let's now describe precisely the parameters:
-  * @param numPoints the number of points
-  * @param points the array of pointers to the points on which to perform the linear regression
-  * @param result pointer to the vector in which to store the result.
-                  This vector must be of the same type and size as the
-                  data points. The meaning of its coords is as follows.
-                  For brevity, let \f$n=Size\f$,
-                  \f$r_i=result[i]\f$,
-                  and \f$f=funcOfOthers\f$. Denote by
-                  \f$x_0,\ldots,x_{n-1}\f$
-                  the n coordinates in the n-dimensional space.
-                  Then the resulting equation is:
-                  \f[ x_f = r_0 x_0 + \cdots + r_{f-1}x_{f-1}
-                   + r_{f+1}x_{f+1} + \cdots + r_{n-1}x_{n-1} + r_n. \f]
-  * @param funcOfOthers Determines which coord to express as a function of the
-                        others. Coords are numbered starting from 0, so that a
-                        value of 0 means \f$x\f$, 1 means \f$y\f$,
-                        2 means \f$z\f$, ...
-  *
-  * \sa fitHyperplane()
-  */
-template<typename VectorType>
-void linearRegression(int numPoints,
-                      VectorType **points,
-                      VectorType *result,
-                      int funcOfOthers )
-{
-  typedef typename VectorType::Scalar Scalar;
-  typedef Hyperplane<Scalar, VectorType::SizeAtCompileTime> HyperplaneType;
-  const int size = points[0]->size();
-  result->resize(size);
-  HyperplaneType h(size);
-  fitHyperplane(numPoints, points, &h);
-  for(int i = 0; i < funcOfOthers; i++)
-    result->coeffRef(i) = - h.coeffs()[i] / h.coeffs()[funcOfOthers];
-  for(int i = funcOfOthers; i < size; i++)
-    result->coeffRef(i) = - h.coeffs()[i+1] / h.coeffs()[funcOfOthers];
-}
-
-/** \ingroup LeastSquares_Module
-  *
-  * \leastsquares_module
-  *
-  * This function is quite similar to linearRegression(), so we refer to the
-  * documentation of this function and only list here the differences.
-  *
-  * The main difference from linearRegression() is that this function doesn't
-  * take a \a funcOfOthers argument. Instead, it finds a general equation
-  * of the form
-  * \f[ r_0 x_0 + \cdots + r_{n-1}x_{n-1} + r_n = 0, \f]
-  * where \f$n=Size\f$, \f$r_i=retCoefficients[i]\f$, and we denote by
-  * \f$x_0,\ldots,x_{n-1}\f$ the n coordinates in the n-dimensional space.
-  *
-  * Thus, the vector \a retCoefficients has size \f$n+1\f$, which is another
-  * difference from linearRegression().
-  *
-  * In practice, this function performs an hyper-plane fit in a total least square sense
-  * via the following steps:
-  *  1 - center the data to the mean
-  *  2 - compute the covariance matrix
-  *  3 - pick the eigenvector corresponding to the smallest eigenvalue of the covariance matrix
-  * The ratio of the smallest eigenvalue and the second one gives us a hint about the relevance
-  * of the solution. This value is optionally returned in \a soundness.
-  *
-  * \sa linearRegression()
-  */
-template<typename VectorType, typename HyperplaneType>
-void fitHyperplane(int numPoints,
-                   VectorType **points,
-                   HyperplaneType *result,
-                   typename NumTraits<typename VectorType::Scalar>::Real* soundness = 0)
-{
-  typedef typename VectorType::Scalar Scalar;
-  typedef Matrix<Scalar,VectorType::SizeAtCompileTime,VectorType::SizeAtCompileTime> CovMatrixType;
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType)
-  ei_assert(numPoints >= 1);
-  int size = points[0]->size();
-  ei_assert(size+1 == result->coeffs().size());
-
-  // compute the mean of the data
-  VectorType mean = VectorType::Zero(size);
-  for(int i = 0; i < numPoints; ++i)
-    mean += *(points[i]);
-  mean /= numPoints;
-
-  // compute the covariance matrix
-  CovMatrixType covMat = CovMatrixType::Zero(size, size);
-  for(int i = 0; i < numPoints; ++i)
-  {
-    VectorType diff = (*(points[i]) - mean).conjugate();
-    covMat += diff * diff.adjoint();
-  }
-
-  // now we just have to pick the eigen vector with smallest eigen value
-  SelfAdjointEigenSolver<CovMatrixType> eig(covMat);
-  result->normal() = eig.eigenvectors().col(0);
-  if (soundness)
-    *soundness = eig.eigenvalues().coeff(0)/eig.eigenvalues().coeff(1);
-
-  // let's compute the constant coefficient such that the
-  // plane pass trough the mean point:
-  result->offset() = - (result->normal().cwise()* mean).sum();
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN2_LEASTSQUARES_H
diff --git a/Eigen/src/Eigen2Support/Macros.h b/Eigen/src/Eigen2Support/Macros.h
deleted file mode 100644
index 351c32a..0000000
--- a/Eigen/src/Eigen2Support/Macros.h
+++ /dev/null
@@ -1,20 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_MACROS_H
-#define EIGEN2_MACROS_H
-
-#define ei_assert eigen_assert
-#define ei_internal_assert eigen_internal_assert
-
-#define EIGEN_ALIGN_128 EIGEN_ALIGN16
-
-#define EIGEN_ARCH_WANTS_ALIGNMENT EIGEN_ALIGN_STATICALLY
-
-#endif // EIGEN2_MACROS_H
diff --git a/Eigen/src/Eigen2Support/MathFunctions.h b/Eigen/src/Eigen2Support/MathFunctions.h
deleted file mode 100644
index 3544af2..0000000
--- a/Eigen/src/Eigen2Support/MathFunctions.h
+++ /dev/null
@@ -1,57 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_MATH_FUNCTIONS_H
-#define EIGEN2_MATH_FUNCTIONS_H
-
-namespace Eigen { 
-
-template<typename T> inline typename NumTraits<T>::Real ei_real(const T& x) { return numext::real(x); }
-template<typename T> inline typename NumTraits<T>::Real ei_imag(const T& x) { return numext::imag(x); }
-template<typename T> inline T ei_conj(const T& x) { return numext::conj(x); }
-template<typename T> inline typename NumTraits<T>::Real ei_abs (const T& x) { using std::abs; return abs(x); }
-template<typename T> inline typename NumTraits<T>::Real ei_abs2(const T& x) { return numext::abs2(x); }
-template<typename T> inline T ei_sqrt(const T& x) { using std::sqrt; return sqrt(x); }
-template<typename T> inline T ei_exp (const T& x) { using std::exp;  return exp(x); }
-template<typename T> inline T ei_log (const T& x) { using std::log;  return log(x); }
-template<typename T> inline T ei_sin (const T& x) { using std::sin;  return sin(x); }
-template<typename T> inline T ei_cos (const T& x) { using std::cos;  return cos(x); }
-template<typename T> inline T ei_atan2(const T& x,const T& y) { using std::atan2; return atan2(x,y); }
-template<typename T> inline T ei_pow (const T& x,const T& y) { return numext::pow(x,y); }
-template<typename T> inline T ei_random () { return internal::random<T>(); }
-template<typename T> inline T ei_random (const T& x, const T& y) { return internal::random(x, y); }
-
-template<typename T> inline T precision () { return NumTraits<T>::dummy_precision(); }
-template<typename T> inline T machine_epsilon () { return NumTraits<T>::epsilon(); }
-
-
-template<typename Scalar, typename OtherScalar>
-inline bool ei_isMuchSmallerThan(const Scalar& x, const OtherScalar& y,
-                                   typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
-{
-  return internal::isMuchSmallerThan(x, y, precision);
-}
-
-template<typename Scalar>
-inline bool ei_isApprox(const Scalar& x, const Scalar& y,
-                          typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
-{
-  return internal::isApprox(x, y, precision);
-}
-
-template<typename Scalar>
-inline bool ei_isApproxOrLessThan(const Scalar& x, const Scalar& y,
-                                    typename NumTraits<Scalar>::Real precision = NumTraits<Scalar>::dummy_precision())
-{
-  return internal::isApproxOrLessThan(x, y, precision);
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN2_MATH_FUNCTIONS_H
diff --git a/Eigen/src/Eigen2Support/Memory.h b/Eigen/src/Eigen2Support/Memory.h
deleted file mode 100644
index f86372b..0000000
--- a/Eigen/src/Eigen2Support/Memory.h
+++ /dev/null
@@ -1,45 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_MEMORY_H
-#define EIGEN2_MEMORY_H
-
-namespace Eigen { 
-
-inline void* ei_aligned_malloc(size_t size) { return internal::aligned_malloc(size); }
-inline void  ei_aligned_free(void *ptr) { internal::aligned_free(ptr); }
-inline void* ei_aligned_realloc(void *ptr, size_t new_size, size_t old_size) { return internal::aligned_realloc(ptr, new_size, old_size); }
-inline void* ei_handmade_aligned_malloc(size_t size) { return internal::handmade_aligned_malloc(size); }
-inline void  ei_handmade_aligned_free(void *ptr) { internal::handmade_aligned_free(ptr); }
-
-template<bool Align> inline void* ei_conditional_aligned_malloc(size_t size)
-{
-  return internal::conditional_aligned_malloc<Align>(size);
-}
-template<bool Align> inline void ei_conditional_aligned_free(void *ptr)
-{
-  internal::conditional_aligned_free<Align>(ptr);
-}
-template<bool Align> inline void* ei_conditional_aligned_realloc(void* ptr, size_t new_size, size_t old_size)
-{
-  return internal::conditional_aligned_realloc<Align>(ptr, new_size, old_size);
-}
-
-template<typename T> inline T* ei_aligned_new(size_t size)
-{
-  return internal::aligned_new<T>(size);
-}
-template<typename T> inline void ei_aligned_delete(T *ptr, size_t size)
-{
-  return internal::aligned_delete(ptr, size);
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN2_MACROS_H
diff --git a/Eigen/src/Eigen2Support/Meta.h b/Eigen/src/Eigen2Support/Meta.h
deleted file mode 100644
index fa37cfc..0000000
--- a/Eigen/src/Eigen2Support/Meta.h
+++ /dev/null
@@ -1,75 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_META_H
-#define EIGEN2_META_H
-
-namespace Eigen { 
-
-template<typename T>
-struct ei_traits : internal::traits<T>
-{};
-
-struct ei_meta_true {  enum { ret = 1 }; };
-struct ei_meta_false { enum { ret = 0 }; };
-
-template<bool Condition, typename Then, typename Else>
-struct ei_meta_if { typedef Then ret; };
-
-template<typename Then, typename Else>
-struct ei_meta_if <false, Then, Else> { typedef Else ret; };
-
-template<typename T, typename U> struct ei_is_same_type { enum { ret = 0 }; };
-template<typename T> struct ei_is_same_type<T,T> { enum { ret = 1 }; };
-
-template<typename T> struct ei_unref { typedef T type; };
-template<typename T> struct ei_unref<T&> { typedef T type; };
-
-template<typename T> struct ei_unpointer { typedef T type; };
-template<typename T> struct ei_unpointer<T*> { typedef T type; };
-template<typename T> struct ei_unpointer<T*const> { typedef T type; };
-
-template<typename T> struct ei_unconst { typedef T type; };
-template<typename T> struct ei_unconst<const T> { typedef T type; };
-template<typename T> struct ei_unconst<T const &> { typedef T & type; };
-template<typename T> struct ei_unconst<T const *> { typedef T * type; };
-
-template<typename T> struct ei_cleantype { typedef T type; };
-template<typename T> struct ei_cleantype<const T>   { typedef typename ei_cleantype<T>::type type; };
-template<typename T> struct ei_cleantype<const T&>  { typedef typename ei_cleantype<T>::type type; };
-template<typename T> struct ei_cleantype<T&>        { typedef typename ei_cleantype<T>::type type; };
-template<typename T> struct ei_cleantype<const T*>  { typedef typename ei_cleantype<T>::type type; };
-template<typename T> struct ei_cleantype<T*>        { typedef typename ei_cleantype<T>::type type; };
-
-/** \internal In short, it computes int(sqrt(\a Y)) with \a Y an integer.
-  * Usage example: \code ei_meta_sqrt<1023>::ret \endcode
-  */
-template<int Y,
-         int InfX = 0,
-         int SupX = ((Y==1) ? 1 : Y/2),
-         bool Done = ((SupX-InfX)<=1 ? true : ((SupX*SupX <= Y) && ((SupX+1)*(SupX+1) > Y))) >
-                                // use ?: instead of || just to shut up a stupid gcc 4.3 warning
-class ei_meta_sqrt
-{
-    enum {
-      MidX = (InfX+SupX)/2,
-      TakeInf = MidX*MidX > Y ? 1 : 0,
-      NewInf = int(TakeInf) ? InfX : int(MidX),
-      NewSup = int(TakeInf) ? int(MidX) : SupX
-    };
-  public:
-    enum { ret = ei_meta_sqrt<Y,NewInf,NewSup>::ret };
-};
-
-template<int Y, int InfX, int SupX>
-class ei_meta_sqrt<Y, InfX, SupX, true> { public:  enum { ret = (SupX*SupX <= Y) ? SupX : InfX }; };
-
-} // end namespace Eigen
-
-#endif // EIGEN2_META_H
diff --git a/Eigen/src/Eigen2Support/Minor.h b/Eigen/src/Eigen2Support/Minor.h
deleted file mode 100644
index 4cded57..0000000
--- a/Eigen/src/Eigen2Support/Minor.h
+++ /dev/null
@@ -1,117 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2006-2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_MINOR_H
-#define EIGEN_MINOR_H
-
-namespace Eigen { 
-
-/**
-  * \class Minor
-  *
-  * \brief Expression of a minor
-  *
-  * \param MatrixType the type of the object in which we are taking a minor
-  *
-  * This class represents an expression of a minor. It is the return
-  * type of MatrixBase::minor() and most of the time this is the only way it
-  * is used.
-  *
-  * \sa MatrixBase::minor()
-  */
-
-namespace internal {
-template<typename MatrixType>
-struct traits<Minor<MatrixType> >
- : traits<MatrixType>
-{
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
-  typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
-  typedef typename MatrixType::StorageKind StorageKind;
-  enum {
-    RowsAtCompileTime = (MatrixType::RowsAtCompileTime != Dynamic) ?
-                          int(MatrixType::RowsAtCompileTime) - 1 : Dynamic,
-    ColsAtCompileTime = (MatrixType::ColsAtCompileTime != Dynamic) ?
-                          int(MatrixType::ColsAtCompileTime) - 1 : Dynamic,
-    MaxRowsAtCompileTime = (MatrixType::MaxRowsAtCompileTime != Dynamic) ?
-                             int(MatrixType::MaxRowsAtCompileTime) - 1 : Dynamic,
-    MaxColsAtCompileTime = (MatrixType::MaxColsAtCompileTime != Dynamic) ?
-                             int(MatrixType::MaxColsAtCompileTime) - 1 : Dynamic,
-    Flags = _MatrixTypeNested::Flags & (HereditaryBits | LvalueBit),
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost // minor is used typically on tiny matrices,
-      // where loops are unrolled and the 'if' evaluates at compile time
-  };
-};
-}
-
-template<typename MatrixType> class Minor
-  : public MatrixBase<Minor<MatrixType> >
-{
-  public:
-
-    typedef MatrixBase<Minor> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(Minor)
-
-    inline Minor(const MatrixType& matrix,
-                       Index row, Index col)
-      : m_matrix(matrix), m_row(row), m_col(col)
-    {
-      eigen_assert(row >= 0 && row < matrix.rows()
-          && col >= 0 && col < matrix.cols());
-    }
-
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Minor)
-
-    inline Index rows() const { return m_matrix.rows() - 1; }
-    inline Index cols() const { return m_matrix.cols() - 1; }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      return m_matrix.const_cast_derived().coeffRef(row + (row >= m_row), col + (col >= m_col));
-    }
-
-    inline const Scalar coeff(Index row, Index col) const
-    {
-      return m_matrix.coeff(row + (row >= m_row), col + (col >= m_col));
-    }
-
-  protected:
-    const typename MatrixType::Nested m_matrix;
-    const Index m_row, m_col;
-};
-
-/**
-  * \return an expression of the (\a row, \a col)-minor of *this,
-  * i.e. an expression constructed from *this by removing the specified
-  * row and column.
-  *
-  * Example: \include MatrixBase_minor.cpp
-  * Output: \verbinclude MatrixBase_minor.out
-  *
-  * \sa class Minor
-  */
-template<typename Derived>
-inline Minor<Derived>
-MatrixBase<Derived>::minor(Index row, Index col)
-{
-  return Minor<Derived>(derived(), row, col);
-}
-
-/**
-  * This is the const version of minor(). */
-template<typename Derived>
-inline const Minor<Derived>
-MatrixBase<Derived>::minor(Index row, Index col) const
-{
-  return Minor<Derived>(derived(), row, col);
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN_MINOR_H
diff --git a/Eigen/src/Eigen2Support/QR.h b/Eigen/src/Eigen2Support/QR.h
deleted file mode 100644
index 2042c98..0000000
--- a/Eigen/src/Eigen2Support/QR.h
+++ /dev/null
@@ -1,67 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-// Copyright (C) 2011 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_QR_H
-#define EIGEN2_QR_H
-
-namespace Eigen { 
-
-template<typename MatrixType>
-class QR : public HouseholderQR<MatrixType>
-{
-  public:
-
-    typedef HouseholderQR<MatrixType> Base;
-    typedef Block<const MatrixType, MatrixType::ColsAtCompileTime, MatrixType::ColsAtCompileTime> MatrixRBlockType;
-
-    QR() : Base() {}
-
-    template<typename T>
-    explicit QR(const T& t) : Base(t) {}
-
-    template<typename OtherDerived, typename ResultType>
-    bool solve(const MatrixBase<OtherDerived>& b, ResultType *result) const
-    {
-      *result = static_cast<const Base*>(this)->solve(b);
-      return true;
-    }
-
-    MatrixType matrixQ(void) const {
-      MatrixType ret = MatrixType::Identity(this->rows(), this->cols());
-      ret = this->householderQ() * ret;
-      return ret;
-    }
-
-    bool isFullRank() const {
-      return true;
-    }
-    
-    const TriangularView<MatrixRBlockType, UpperTriangular>
-    matrixR(void) const
-    {
-      int cols = this->cols();
-      return MatrixRBlockType(this->matrixQR(), 0, 0, cols, cols).template triangularView<UpperTriangular>();
-    }
-};
-
-/** \return the QR decomposition of \c *this.
-  *
-  * \sa class QR
-  */
-template<typename Derived>
-const QR<typename MatrixBase<Derived>::PlainObject>
-MatrixBase<Derived>::qr() const
-{
-  return QR<PlainObject>(eval());
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN2_QR_H
diff --git a/Eigen/src/Eigen2Support/SVD.h b/Eigen/src/Eigen2Support/SVD.h
deleted file mode 100644
index 3d03d22..0000000
--- a/Eigen/src/Eigen2Support/SVD.h
+++ /dev/null
@@ -1,637 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008 Gael Guennebaud <g.gael@free.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_SVD_H
-#define EIGEN2_SVD_H
-
-namespace Eigen {
-
-/** \ingroup SVD_Module
-  * \nonstableyet
-  *
-  * \class SVD
-  *
-  * \brief Standard SVD decomposition of a matrix and associated features
-  *
-  * \param MatrixType the type of the matrix of which we are computing the SVD decomposition
-  *
-  * This class performs a standard SVD decomposition of a real matrix A of size \c M x \c N
-  * with \c M \>= \c N.
-  *
-  *
-  * \sa MatrixBase::SVD()
-  */
-template<typename MatrixType> class SVD
-{
-  private:
-    typedef typename MatrixType::Scalar Scalar;
-    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-
-    enum {
-      PacketSize = internal::packet_traits<Scalar>::size,
-      AlignmentMask = int(PacketSize)-1,
-      MinSize = EIGEN_SIZE_MIN_PREFER_DYNAMIC(MatrixType::RowsAtCompileTime, MatrixType::ColsAtCompileTime)
-    };
-
-    typedef Matrix<Scalar, MatrixType::RowsAtCompileTime, 1> ColVector;
-    typedef Matrix<Scalar, MatrixType::ColsAtCompileTime, 1> RowVector;
-
-    typedef Matrix<Scalar, MatrixType::RowsAtCompileTime, MinSize> MatrixUType;
-    typedef Matrix<Scalar, MatrixType::ColsAtCompileTime, MatrixType::ColsAtCompileTime> MatrixVType;
-    typedef Matrix<Scalar, MinSize, 1> SingularValuesType;
-
-  public:
-
-    SVD() {} // a user who relied on compiler-generated default compiler reported problems with MSVC in 2.0.7
-    
-    SVD(const MatrixType& matrix)
-      : m_matU(matrix.rows(), (std::min)(matrix.rows(), matrix.cols())),
-        m_matV(matrix.cols(),matrix.cols()),
-        m_sigma((std::min)(matrix.rows(),matrix.cols()))
-    {
-      compute(matrix);
-    }
-
-    template<typename OtherDerived, typename ResultType>
-    bool solve(const MatrixBase<OtherDerived> &b, ResultType* result) const;
-
-    const MatrixUType& matrixU() const { return m_matU; }
-    const SingularValuesType& singularValues() const { return m_sigma; }
-    const MatrixVType& matrixV() const { return m_matV; }
-
-    void compute(const MatrixType& matrix);
-    SVD& sort();
-
-    template<typename UnitaryType, typename PositiveType>
-    void computeUnitaryPositive(UnitaryType *unitary, PositiveType *positive) const;
-    template<typename PositiveType, typename UnitaryType>
-    void computePositiveUnitary(PositiveType *positive, UnitaryType *unitary) const;
-    template<typename RotationType, typename ScalingType>
-    void computeRotationScaling(RotationType *unitary, ScalingType *positive) const;
-    template<typename ScalingType, typename RotationType>
-    void computeScalingRotation(ScalingType *positive, RotationType *unitary) const;
-
-  protected:
-    /** \internal */
-    MatrixUType m_matU;
-    /** \internal */
-    MatrixVType m_matV;
-    /** \internal */
-    SingularValuesType m_sigma;
-};
-
-/** Computes / recomputes the SVD decomposition A = U S V^* of \a matrix
-  *
-  * \note this code has been adapted from JAMA (public domain)
-  */
-template<typename MatrixType>
-void SVD<MatrixType>::compute(const MatrixType& matrix)
-{
-  const int m = matrix.rows();
-  const int n = matrix.cols();
-  const int nu = (std::min)(m,n);
-  ei_assert(m>=n && "In Eigen 2.0, SVD only works for MxN matrices with M>=N. Sorry!");
-  ei_assert(m>1 && "In Eigen 2.0, SVD doesn't work on 1x1 matrices");
-
-  m_matU.resize(m, nu);
-  m_matU.setZero();
-  m_sigma.resize((std::min)(m,n));
-  m_matV.resize(n,n);
-
-  RowVector e(n);
-  ColVector work(m);
-  MatrixType matA(matrix);
-  const bool wantu = true;
-  const bool wantv = true;
-  int i=0, j=0, k=0;
-
-  // Reduce A to bidiagonal form, storing the diagonal elements
-  // in s and the super-diagonal elements in e.
-  int nct = (std::min)(m-1,n);
-  int nrt = (std::max)(0,(std::min)(n-2,m));
-  for (k = 0; k < (std::max)(nct,nrt); ++k)
-  {
-    if (k < nct)
-    {
-      // Compute the transformation for the k-th column and
-      // place the k-th diagonal in m_sigma[k].
-      m_sigma[k] = matA.col(k).end(m-k).norm();
-      if (m_sigma[k] != 0.0) // FIXME
-      {
-        if (matA(k,k) < 0.0)
-          m_sigma[k] = -m_sigma[k];
-        matA.col(k).end(m-k) /= m_sigma[k];
-        matA(k,k) += 1.0;
-      }
-      m_sigma[k] = -m_sigma[k];
-    }
-
-    for (j = k+1; j < n; ++j)
-    {
-      if ((k < nct) && (m_sigma[k] != 0.0))
-      {
-        // Apply the transformation.
-        Scalar t = matA.col(k).end(m-k).eigen2_dot(matA.col(j).end(m-k)); // FIXME dot product or cwise prod + .sum() ??
-        t = -t/matA(k,k);
-        matA.col(j).end(m-k) += t * matA.col(k).end(m-k);
-      }
-
-      // Place the k-th row of A into e for the
-      // subsequent calculation of the row transformation.
-      e[j] = matA(k,j);
-    }
-
-    // Place the transformation in U for subsequent back multiplication.
-    if (wantu & (k < nct))
-      m_matU.col(k).end(m-k) = matA.col(k).end(m-k);
-
-    if (k < nrt)
-    {
-      // Compute the k-th row transformation and place the
-      // k-th super-diagonal in e[k].
-      e[k] = e.end(n-k-1).norm();
-      if (e[k] != 0.0)
-      {
-          if (e[k+1] < 0.0)
-            e[k] = -e[k];
-          e.end(n-k-1) /= e[k];
-          e[k+1] += 1.0;
-      }
-      e[k] = -e[k];
-      if ((k+1 < m) & (e[k] != 0.0))
-      {
-        // Apply the transformation.
-        work.end(m-k-1) = matA.corner(BottomRight,m-k-1,n-k-1) * e.end(n-k-1);
-        for (j = k+1; j < n; ++j)
-          matA.col(j).end(m-k-1) += (-e[j]/e[k+1]) * work.end(m-k-1);
-      }
-
-      // Place the transformation in V for subsequent back multiplication.
-      if (wantv)
-        m_matV.col(k).end(n-k-1) = e.end(n-k-1);
-    }
-  }
-
-
-  // Set up the final bidiagonal matrix or order p.
-  int p = (std::min)(n,m+1);
-  if (nct < n)
-    m_sigma[nct] = matA(nct,nct);
-  if (m < p)
-    m_sigma[p-1] = 0.0;
-  if (nrt+1 < p)
-    e[nrt] = matA(nrt,p-1);
-  e[p-1] = 0.0;
-
-  // If required, generate U.
-  if (wantu)
-  {
-    for (j = nct; j < nu; ++j)
-    {
-      m_matU.col(j).setZero();
-      m_matU(j,j) = 1.0;
-    }
-    for (k = nct-1; k >= 0; k--)
-    {
-      if (m_sigma[k] != 0.0)
-      {
-        for (j = k+1; j < nu; ++j)
-        {
-          Scalar t = m_matU.col(k).end(m-k).eigen2_dot(m_matU.col(j).end(m-k)); // FIXME is it really a dot product we want ?
-          t = -t/m_matU(k,k);
-          m_matU.col(j).end(m-k) += t * m_matU.col(k).end(m-k);
-        }
-        m_matU.col(k).end(m-k) = - m_matU.col(k).end(m-k);
-        m_matU(k,k) = Scalar(1) + m_matU(k,k);
-        if (k-1>0)
-          m_matU.col(k).start(k-1).setZero();
-      }
-      else
-      {
-        m_matU.col(k).setZero();
-        m_matU(k,k) = 1.0;
-      }
-    }
-  }
-
-  // If required, generate V.
-  if (wantv)
-  {
-    for (k = n-1; k >= 0; k--)
-    {
-      if ((k < nrt) & (e[k] != 0.0))
-      {
-        for (j = k+1; j < nu; ++j)
-        {
-          Scalar t = m_matV.col(k).end(n-k-1).eigen2_dot(m_matV.col(j).end(n-k-1)); // FIXME is it really a dot product we want ?
-          t = -t/m_matV(k+1,k);
-          m_matV.col(j).end(n-k-1) += t * m_matV.col(k).end(n-k-1);
-        }
-      }
-      m_matV.col(k).setZero();
-      m_matV(k,k) = 1.0;
-    }
-  }
-
-  // Main iteration loop for the singular values.
-  int pp = p-1;
-  int iter = 0;
-  Scalar eps = ei_pow(Scalar(2),ei_is_same_type<Scalar,float>::ret ? Scalar(-23) : Scalar(-52));
-  while (p > 0)
-  {
-    int k=0;
-    int kase=0;
-
-    // Here is where a test for too many iterations would go.
-
-    // This section of the program inspects for
-    // negligible elements in the s and e arrays.  On
-    // completion the variables kase and k are set as follows.
-
-    // kase = 1     if s(p) and e[k-1] are negligible and k<p
-    // kase = 2     if s(k) is negligible and k<p
-    // kase = 3     if e[k-1] is negligible, k<p, and
-    //              s(k), ..., s(p) are not negligible (qr step).
-    // kase = 4     if e(p-1) is negligible (convergence).
-
-    for (k = p-2; k >= -1; --k)
-    {
-      if (k == -1)
-          break;
-      if (ei_abs(e[k]) <= eps*(ei_abs(m_sigma[k]) + ei_abs(m_sigma[k+1])))
-      {
-          e[k] = 0.0;
-          break;
-      }
-    }
-    if (k == p-2)
-    {
-      kase = 4;
-    }
-    else
-    {
-      int ks;
-      for (ks = p-1; ks >= k; --ks)
-      {
-        if (ks == k)
-          break;
-        Scalar t = (ks != p ? ei_abs(e[ks]) : Scalar(0)) + (ks != k+1 ? ei_abs(e[ks-1]) : Scalar(0));
-        if (ei_abs(m_sigma[ks]) <= eps*t)
-        {
-          m_sigma[ks] = 0.0;
-          break;
-        }
-      }
-      if (ks == k)
-      {
-        kase = 3;
-      }
-      else if (ks == p-1)
-      {
-        kase = 1;
-      }
-      else
-      {
-        kase = 2;
-        k = ks;
-      }
-    }
-    ++k;
-
-    // Perform the task indicated by kase.
-    switch (kase)
-    {
-
-      // Deflate negligible s(p).
-      case 1:
-      {
-        Scalar f(e[p-2]);
-        e[p-2] = 0.0;
-        for (j = p-2; j >= k; --j)
-        {
-          Scalar t(numext::hypot(m_sigma[j],f));
-          Scalar cs(m_sigma[j]/t);
-          Scalar sn(f/t);
-          m_sigma[j] = t;
-          if (j != k)
-          {
-            f = -sn*e[j-1];
-            e[j-1] = cs*e[j-1];
-          }
-          if (wantv)
-          {
-            for (i = 0; i < n; ++i)
-            {
-              t = cs*m_matV(i,j) + sn*m_matV(i,p-1);
-              m_matV(i,p-1) = -sn*m_matV(i,j) + cs*m_matV(i,p-1);
-              m_matV(i,j) = t;
-            }
-          }
-        }
-      }
-      break;
-
-      // Split at negligible s(k).
-      case 2:
-      {
-        Scalar f(e[k-1]);
-        e[k-1] = 0.0;
-        for (j = k; j < p; ++j)
-        {
-          Scalar t(numext::hypot(m_sigma[j],f));
-          Scalar cs( m_sigma[j]/t);
-          Scalar sn(f/t);
-          m_sigma[j] = t;
-          f = -sn*e[j];
-          e[j] = cs*e[j];
-          if (wantu)
-          {
-            for (i = 0; i < m; ++i)
-            {
-              t = cs*m_matU(i,j) + sn*m_matU(i,k-1);
-              m_matU(i,k-1) = -sn*m_matU(i,j) + cs*m_matU(i,k-1);
-              m_matU(i,j) = t;
-            }
-          }
-        }
-      }
-      break;
-
-      // Perform one qr step.
-      case 3:
-      {
-        // Calculate the shift.
-        Scalar scale = (std::max)((std::max)((std::max)((std::max)(
-                        ei_abs(m_sigma[p-1]),ei_abs(m_sigma[p-2])),ei_abs(e[p-2])),
-                        ei_abs(m_sigma[k])),ei_abs(e[k]));
-        Scalar sp = m_sigma[p-1]/scale;
-        Scalar spm1 = m_sigma[p-2]/scale;
-        Scalar epm1 = e[p-2]/scale;
-        Scalar sk = m_sigma[k]/scale;
-        Scalar ek = e[k]/scale;
-        Scalar b = ((spm1 + sp)*(spm1 - sp) + epm1*epm1)/Scalar(2);
-        Scalar c = (sp*epm1)*(sp*epm1);
-        Scalar shift(0);
-        if ((b != 0.0) || (c != 0.0))
-        {
-          shift = ei_sqrt(b*b + c);
-          if (b < 0.0)
-            shift = -shift;
-          shift = c/(b + shift);
-        }
-        Scalar f = (sk + sp)*(sk - sp) + shift;
-        Scalar g = sk*ek;
-
-        // Chase zeros.
-
-        for (j = k; j < p-1; ++j)
-        {
-          Scalar t = numext::hypot(f,g);
-          Scalar cs = f/t;
-          Scalar sn = g/t;
-          if (j != k)
-            e[j-1] = t;
-          f = cs*m_sigma[j] + sn*e[j];
-          e[j] = cs*e[j] - sn*m_sigma[j];
-          g = sn*m_sigma[j+1];
-          m_sigma[j+1] = cs*m_sigma[j+1];
-          if (wantv)
-          {
-            for (i = 0; i < n; ++i)
-            {
-              t = cs*m_matV(i,j) + sn*m_matV(i,j+1);
-              m_matV(i,j+1) = -sn*m_matV(i,j) + cs*m_matV(i,j+1);
-              m_matV(i,j) = t;
-            }
-          }
-          t = numext::hypot(f,g);
-          cs = f/t;
-          sn = g/t;
-          m_sigma[j] = t;
-          f = cs*e[j] + sn*m_sigma[j+1];
-          m_sigma[j+1] = -sn*e[j] + cs*m_sigma[j+1];
-          g = sn*e[j+1];
-          e[j+1] = cs*e[j+1];
-          if (wantu && (j < m-1))
-          {
-            for (i = 0; i < m; ++i)
-            {
-              t = cs*m_matU(i,j) + sn*m_matU(i,j+1);
-              m_matU(i,j+1) = -sn*m_matU(i,j) + cs*m_matU(i,j+1);
-              m_matU(i,j) = t;
-            }
-          }
-        }
-        e[p-2] = f;
-        iter = iter + 1;
-      }
-      break;
-
-      // Convergence.
-      case 4:
-      {
-        // Make the singular values positive.
-        if (m_sigma[k] <= 0.0)
-        {
-          m_sigma[k] = m_sigma[k] < Scalar(0) ? -m_sigma[k] : Scalar(0);
-          if (wantv)
-            m_matV.col(k).start(pp+1) = -m_matV.col(k).start(pp+1);
-        }
-
-        // Order the singular values.
-        while (k < pp)
-        {
-          if (m_sigma[k] >= m_sigma[k+1])
-            break;
-          Scalar t = m_sigma[k];
-          m_sigma[k] = m_sigma[k+1];
-          m_sigma[k+1] = t;
-          if (wantv && (k < n-1))
-            m_matV.col(k).swap(m_matV.col(k+1));
-          if (wantu && (k < m-1))
-            m_matU.col(k).swap(m_matU.col(k+1));
-          ++k;
-        }
-        iter = 0;
-        p--;
-      }
-      break;
-    } // end big switch
-  } // end iterations
-}
-
-template<typename MatrixType>
-SVD<MatrixType>& SVD<MatrixType>::sort()
-{
-  int mu = m_matU.rows();
-  int mv = m_matV.rows();
-  int n  = m_matU.cols();
-
-  for (int i=0; i<n; ++i)
-  {
-    int  k = i;
-    Scalar p = m_sigma.coeff(i);
-
-    for (int j=i+1; j<n; ++j)
-    {
-      if (m_sigma.coeff(j) > p)
-      {
-        k = j;
-        p = m_sigma.coeff(j);
-      }
-    }
-    if (k != i)
-    {
-      m_sigma.coeffRef(k) = m_sigma.coeff(i);  // i.e.
-      m_sigma.coeffRef(i) = p;                 // swaps the i-th and the k-th elements
-
-      int j = mu;
-      for(int s=0; j!=0; ++s, --j)
-        std::swap(m_matU.coeffRef(s,i), m_matU.coeffRef(s,k));
-
-      j = mv;
-      for (int s=0; j!=0; ++s, --j)
-        std::swap(m_matV.coeffRef(s,i), m_matV.coeffRef(s,k));
-    }
-  }
-  return *this;
-}
-
-/** \returns the solution of \f$ A x = b \f$ using the current SVD decomposition of A.
-  * The parts of the solution corresponding to zero singular values are ignored.
-  *
-  * \sa MatrixBase::svd(), LU::solve(), LLT::solve()
-  */
-template<typename MatrixType>
-template<typename OtherDerived, typename ResultType>
-bool SVD<MatrixType>::solve(const MatrixBase<OtherDerived> &b, ResultType* result) const
-{
-  ei_assert(b.rows() == m_matU.rows());
-
-  Scalar maxVal = m_sigma.cwise().abs().maxCoeff();
-  for (int j=0; j<b.cols(); ++j)
-  {
-    Matrix<Scalar,MatrixUType::RowsAtCompileTime,1> aux = m_matU.transpose() * b.col(j);
-
-    for (int i = 0; i <m_matU.cols(); ++i)
-    {
-      Scalar si = m_sigma.coeff(i);
-      if (ei_isMuchSmallerThan(ei_abs(si),maxVal))
-        aux.coeffRef(i) = 0;
-      else
-        aux.coeffRef(i) /= si;
-    }
-
-    result->col(j) = m_matV * aux;
-  }
-  return true;
-}
-
-/** Computes the polar decomposition of the matrix, as a product unitary x positive.
-  *
-  * If either pointer is zero, the corresponding computation is skipped.
-  *
-  * Only for square matrices.
-  *
-  * \sa computePositiveUnitary(), computeRotationScaling()
-  */
-template<typename MatrixType>
-template<typename UnitaryType, typename PositiveType>
-void SVD<MatrixType>::computeUnitaryPositive(UnitaryType *unitary,
-                                             PositiveType *positive) const
-{
-  ei_assert(m_matU.cols() == m_matV.cols() && "Polar decomposition is only for square matrices");
-  if(unitary) *unitary = m_matU * m_matV.adjoint();
-  if(positive) *positive = m_matV * m_sigma.asDiagonal() * m_matV.adjoint();
-}
-
-/** Computes the polar decomposition of the matrix, as a product positive x unitary.
-  *
-  * If either pointer is zero, the corresponding computation is skipped.
-  *
-  * Only for square matrices.
-  *
-  * \sa computeUnitaryPositive(), computeRotationScaling()
-  */
-template<typename MatrixType>
-template<typename UnitaryType, typename PositiveType>
-void SVD<MatrixType>::computePositiveUnitary(UnitaryType *positive,
-                                             PositiveType *unitary) const
-{
-  ei_assert(m_matU.rows() == m_matV.rows() && "Polar decomposition is only for square matrices");
-  if(unitary) *unitary = m_matU * m_matV.adjoint();
-  if(positive) *positive = m_matU * m_sigma.asDiagonal() * m_matU.adjoint();
-}
-
-/** decomposes the matrix as a product rotation x scaling, the scaling being
-  * not necessarily positive.
-  *
-  * If either pointer is zero, the corresponding computation is skipped.
-  *
-  * This method requires the Geometry module.
-  *
-  * \sa computeScalingRotation(), computeUnitaryPositive()
-  */
-template<typename MatrixType>
-template<typename RotationType, typename ScalingType>
-void SVD<MatrixType>::computeRotationScaling(RotationType *rotation, ScalingType *scaling) const
-{
-  ei_assert(m_matU.rows() == m_matV.rows() && "Polar decomposition is only for square matrices");
-  Scalar x = (m_matU * m_matV.adjoint()).determinant(); // so x has absolute value 1
-  Matrix<Scalar, MatrixType::RowsAtCompileTime, 1> sv(m_sigma);
-  sv.coeffRef(0) *= x;
-  if(scaling) scaling->lazyAssign(m_matV * sv.asDiagonal() * m_matV.adjoint());
-  if(rotation)
-  {
-    MatrixType m(m_matU);
-    m.col(0) /= x;
-    rotation->lazyAssign(m * m_matV.adjoint());
-  }
-}
-
-/** decomposes the matrix as a product scaling x rotation, the scaling being
-  * not necessarily positive.
-  *
-  * If either pointer is zero, the corresponding computation is skipped.
-  *
-  * This method requires the Geometry module.
-  *
-  * \sa computeRotationScaling(), computeUnitaryPositive()
-  */
-template<typename MatrixType>
-template<typename ScalingType, typename RotationType>
-void SVD<MatrixType>::computeScalingRotation(ScalingType *scaling, RotationType *rotation) const
-{
-  ei_assert(m_matU.rows() == m_matV.rows() && "Polar decomposition is only for square matrices");
-  Scalar x = (m_matU * m_matV.adjoint()).determinant(); // so x has absolute value 1
-  Matrix<Scalar, MatrixType::RowsAtCompileTime, 1> sv(m_sigma);
-  sv.coeffRef(0) *= x;
-  if(scaling) scaling->lazyAssign(m_matU * sv.asDiagonal() * m_matU.adjoint());
-  if(rotation)
-  {
-    MatrixType m(m_matU);
-    m.col(0) /= x;
-    rotation->lazyAssign(m * m_matV.adjoint());
-  }
-}
-
-
-/** \svd_module
-  * \returns the SVD decomposition of \c *this
-  */
-template<typename Derived>
-inline SVD<typename MatrixBase<Derived>::PlainObject>
-MatrixBase<Derived>::svd() const
-{
-  return SVD<PlainObject>(derived());
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN2_SVD_H
diff --git a/Eigen/src/Eigen2Support/TriangularSolver.h b/Eigen/src/Eigen2Support/TriangularSolver.h
deleted file mode 100644
index ebbeb3b..0000000
--- a/Eigen/src/Eigen2Support/TriangularSolver.h
+++ /dev/null
@@ -1,42 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_TRIANGULAR_SOLVER2_H
-#define EIGEN_TRIANGULAR_SOLVER2_H
-
-namespace Eigen { 
-
-const unsigned int UnitDiagBit = UnitDiag;
-const unsigned int SelfAdjointBit = SelfAdjoint;
-const unsigned int UpperTriangularBit = Upper;
-const unsigned int LowerTriangularBit = Lower;
-
-const unsigned int UpperTriangular = Upper;
-const unsigned int LowerTriangular = Lower;
-const unsigned int UnitUpperTriangular = UnitUpper;
-const unsigned int UnitLowerTriangular = UnitLower;
-
-template<typename ExpressionType, unsigned int Added, unsigned int Removed>
-template<typename OtherDerived>
-typename ExpressionType::PlainObject
-Flagged<ExpressionType,Added,Removed>::solveTriangular(const MatrixBase<OtherDerived>& other) const
-{
-  return m_matrix.template triangularView<Added>().solve(other.derived());
-}
-
-template<typename ExpressionType, unsigned int Added, unsigned int Removed>
-template<typename OtherDerived>
-void Flagged<ExpressionType,Added,Removed>::solveTriangularInPlace(const MatrixBase<OtherDerived>& other) const
-{
-  m_matrix.template triangularView<Added>().solveInPlace(other.derived());
-}
-
-} // end namespace Eigen
-    
-#endif // EIGEN_TRIANGULAR_SOLVER2_H
diff --git a/Eigen/src/Eigen2Support/VectorBlock.h b/Eigen/src/Eigen2Support/VectorBlock.h
deleted file mode 100644
index 71a8080..0000000
--- a/Eigen/src/Eigen2Support/VectorBlock.h
+++ /dev/null
@@ -1,94 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
-// Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN2_VECTORBLOCK_H
-#define EIGEN2_VECTORBLOCK_H
-
-namespace Eigen { 
-
-/** \deprecated use DenseMase::head(Index) */
-template<typename Derived>
-inline VectorBlock<Derived>
-MatrixBase<Derived>::start(Index size)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<Derived>(derived(), 0, size);
-}
-
-/** \deprecated use DenseMase::head(Index) */
-template<typename Derived>
-inline const VectorBlock<const Derived>
-MatrixBase<Derived>::start(Index size) const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<const Derived>(derived(), 0, size);
-}
-
-/** \deprecated use DenseMase::tail(Index) */
-template<typename Derived>
-inline VectorBlock<Derived>
-MatrixBase<Derived>::end(Index size)
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<Derived>(derived(), this->size() - size, size);
-}
-
-/** \deprecated use DenseMase::tail(Index) */
-template<typename Derived>
-inline const VectorBlock<const Derived>
-MatrixBase<Derived>::end(Index size) const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<const Derived>(derived(), this->size() - size, size);
-}
-
-/** \deprecated use DenseMase::head() */
-template<typename Derived>
-template<int Size>
-inline VectorBlock<Derived,Size>
-MatrixBase<Derived>::start()
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<Derived,Size>(derived(), 0);
-}
-
-/** \deprecated use DenseMase::head() */
-template<typename Derived>
-template<int Size>
-inline const VectorBlock<const Derived,Size>
-MatrixBase<Derived>::start() const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<const Derived,Size>(derived(), 0);
-}
-
-/** \deprecated use DenseMase::tail() */
-template<typename Derived>
-template<int Size>
-inline VectorBlock<Derived,Size>
-MatrixBase<Derived>::end()
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<Derived, Size>(derived(), size() - Size);
-}
-
-/** \deprecated use DenseMase::tail() */
-template<typename Derived>
-template<int Size>
-inline const VectorBlock<const Derived,Size>
-MatrixBase<Derived>::end() const
-{
-  EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
-  return VectorBlock<const Derived, Size>(derived(), size() - Size);
-}
-
-} // end namespace Eigen
-
-#endif // EIGEN2_VECTORBLOCK_H
diff --git a/Eigen/src/Eigenvalues/CMakeLists.txt b/Eigen/src/Eigenvalues/CMakeLists.txt
deleted file mode 100644
index 193e026..0000000
--- a/Eigen/src/Eigenvalues/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_EIGENVALUES_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_EIGENVALUES_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Eigenvalues COMPONENT Devel
-  )
diff --git a/Eigen/src/Eigenvalues/ComplexEigenSolver.h b/Eigen/src/Eigenvalues/ComplexEigenSolver.h
index 417c729..dc5fae0 100644
--- a/Eigen/src/Eigenvalues/ComplexEigenSolver.h
+++ b/Eigen/src/Eigenvalues/ComplexEigenSolver.h
@@ -60,7 +60,7 @@
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for #MatrixType.
       *
@@ -104,7 +104,7 @@
       * according to the specified problem \a size.
       * \sa ComplexEigenSolver()
       */
-    ComplexEigenSolver(Index size)
+    explicit ComplexEigenSolver(Index size)
             : m_eivec(size, size),
               m_eivalues(size),
               m_schur(size),
@@ -122,7 +122,8 @@
       *
       * This constructor calls compute() to compute the eigendecomposition.
       */
-      ComplexEigenSolver(const MatrixType& matrix, bool computeEigenvectors = true)
+    template<typename InputType>
+    explicit ComplexEigenSolver(const EigenBase<InputType>& matrix, bool computeEigenvectors = true)
             : m_eivec(matrix.rows(),matrix.cols()),
               m_eivalues(matrix.cols()),
               m_schur(matrix.rows()),
@@ -130,7 +131,7 @@
               m_eigenvectorsOk(false),
               m_matX(matrix.rows(),matrix.cols())
     {
-      compute(matrix, computeEigenvectors);
+      compute(matrix.derived(), computeEigenvectors);
     }
 
     /** \brief Returns the eigenvectors of given matrix.
@@ -208,7 +209,8 @@
       * Example: \include ComplexEigenSolver_compute.cpp
       * Output: \verbinclude ComplexEigenSolver_compute.out
       */
-    ComplexEigenSolver& compute(const MatrixType& matrix, bool computeEigenvectors = true);
+    template<typename InputType>
+    ComplexEigenSolver& compute(const EigenBase<InputType>& matrix, bool computeEigenvectors = true);
 
     /** \brief Reports whether previous computation was successful.
       *
@@ -248,14 +250,15 @@
     EigenvectorType m_matX;
 
   private:
-    void doComputeEigenvectors(const RealScalar& matrixnorm);
+    void doComputeEigenvectors(RealScalar matrixnorm);
     void sortEigenvalues(bool computeEigenvectors);
 };
 
 
 template<typename MatrixType>
+template<typename InputType>
 ComplexEigenSolver<MatrixType>& 
-ComplexEigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEigenvectors)
+ComplexEigenSolver<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeEigenvectors)
 {
   check_template_parameters();
   
@@ -264,13 +267,13 @@
 
   // Do a complex Schur decomposition, A = U T U^*
   // The eigenvalues are on the diagonal of T.
-  m_schur.compute(matrix, computeEigenvectors);
+  m_schur.compute(matrix.derived(), computeEigenvectors);
 
   if(m_schur.info() == Success)
   {
     m_eivalues = m_schur.matrixT().diagonal();
     if(computeEigenvectors)
-      doComputeEigenvectors(matrix.norm());
+      doComputeEigenvectors(m_schur.matrixT().norm());
     sortEigenvalues(computeEigenvectors);
   }
 
@@ -281,10 +284,12 @@
 
 
 template<typename MatrixType>
-void ComplexEigenSolver<MatrixType>::doComputeEigenvectors(const RealScalar& matrixnorm)
+void ComplexEigenSolver<MatrixType>::doComputeEigenvectors(RealScalar matrixnorm)
 {
   const Index n = m_eivalues.size();
 
+  matrixnorm = numext::maxi(matrixnorm,(std::numeric_limits<RealScalar>::min)());
+
   // Compute X such that T = X D X^(-1), where D is the diagonal of T.
   // The matrix X is unit triangular.
   m_matX = EigenvectorType::Zero(n, n);
diff --git a/Eigen/src/Eigenvalues/ComplexSchur.h b/Eigen/src/Eigenvalues/ComplexSchur.h
index 89e6cad..7f38919 100644
--- a/Eigen/src/Eigenvalues/ComplexSchur.h
+++ b/Eigen/src/Eigenvalues/ComplexSchur.h
@@ -63,7 +63,7 @@
     /** \brief Scalar type for matrices of type \p _MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for \p _MatrixType. 
       *
@@ -91,7 +91,7 @@
       *
       * \sa compute() for an example.
       */
-    ComplexSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
+    explicit ComplexSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
       : m_matT(size,size),
         m_matU(size,size),
         m_hess(size),
@@ -109,7 +109,8 @@
       *
       * \sa matrixT() and matrixU() for examples.
       */
-    ComplexSchur(const MatrixType& matrix, bool computeU = true)
+    template<typename InputType>
+    explicit ComplexSchur(const EigenBase<InputType>& matrix, bool computeU = true)
       : m_matT(matrix.rows(),matrix.cols()),
         m_matU(matrix.rows(),matrix.cols()),
         m_hess(matrix.rows()),
@@ -117,7 +118,7 @@
         m_matUisUptodate(false),
         m_maxIters(-1)
     {
-      compute(matrix, computeU);
+      compute(matrix.derived(), computeU);
     }
 
     /** \brief Returns the unitary matrix in the Schur decomposition. 
@@ -186,7 +187,8 @@
       *
       * \sa compute(const MatrixType&, bool, Index)
       */
-    ComplexSchur& compute(const MatrixType& matrix, bool computeU = true);
+    template<typename InputType>
+    ComplexSchur& compute(const EigenBase<InputType>& matrix, bool computeU = true);
     
     /** \brief Compute Schur decomposition from a given Hessenberg matrix
      *  \param[in] matrixH Matrix in Hessenberg form H
@@ -313,14 +315,15 @@
 
 
 template<typename MatrixType>
-ComplexSchur<MatrixType>& ComplexSchur<MatrixType>::compute(const MatrixType& matrix, bool computeU)
+template<typename InputType>
+ComplexSchur<MatrixType>& ComplexSchur<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeU)
 {
   m_matUisUptodate = false;
   eigen_assert(matrix.cols() == matrix.rows());
 
   if(matrix.cols() == 1)
   {
-    m_matT = matrix.template cast<ComplexScalar>();
+    m_matT = matrix.derived().template cast<ComplexScalar>();
     if(computeU)  m_matU = ComplexMatrixType::Identity(1,1);
     m_info = Success;
     m_isInitialized = true;
@@ -328,7 +331,7 @@
     return *this;
   }
 
-  internal::complex_schur_reduce_to_hessenberg<MatrixType, NumTraits<Scalar>::IsComplex>::run(*this, matrix, computeU);
+  internal::complex_schur_reduce_to_hessenberg<MatrixType, NumTraits<Scalar>::IsComplex>::run(*this, matrix.derived(), computeU);
   computeFromHessenberg(m_matT, m_matU, computeU);
   return *this;
 }
diff --git a/Eigen/src/Eigenvalues/ComplexSchur_MKL.h b/Eigen/src/Eigenvalues/ComplexSchur_LAPACKE.h
similarity index 65%
rename from Eigen/src/Eigenvalues/ComplexSchur_MKL.h
rename to Eigen/src/Eigenvalues/ComplexSchur_LAPACKE.h
index 91496ae..4980a3e 100644
--- a/Eigen/src/Eigenvalues/ComplexSchur_MKL.h
+++ b/Eigen/src/Eigenvalues/ComplexSchur_LAPACKE.h
@@ -25,27 +25,24 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to LAPACKe
  *    Complex Schur needed to complex unsymmetrical eigenvalues/eigenvectors.
  ********************************************************************************
 */
 
-#ifndef EIGEN_COMPLEX_SCHUR_MKL_H
-#define EIGEN_COMPLEX_SCHUR_MKL_H
-
-#include "Eigen/src/Core/util/MKL_support.h"
+#ifndef EIGEN_COMPLEX_SCHUR_LAPACKE_H
+#define EIGEN_COMPLEX_SCHUR_LAPACKE_H
 
 namespace Eigen { 
 
-/** \internal Specialization for the data types supported by MKL */
+/** \internal Specialization for the data types supported by LAPACKe */
 
-#define EIGEN_MKL_SCHUR_COMPLEX(EIGTYPE, MKLTYPE, MKLPREFIX, MKLPREFIX_U, EIGCOLROW, MKLCOLROW) \
-template<> inline \
+#define EIGEN_LAPACKE_SCHUR_COMPLEX(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX, LAPACKE_PREFIX_U, EIGCOLROW, LAPACKE_COLROW) \
+template<> template<typename InputType> inline \
 ComplexSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >& \
-ComplexSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW>& matrix, bool computeU) \
+ComplexSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const EigenBase<InputType>& matrix, bool computeU) \
 { \
   typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> MatrixType; \
-  typedef MatrixType::Scalar Scalar; \
   typedef MatrixType::RealScalar RealScalar; \
   typedef std::complex<RealScalar> ComplexScalar; \
 \
@@ -54,25 +51,25 @@
   m_matUisUptodate = false; \
   if(matrix.cols() == 1) \
   { \
-    m_matT = matrix.cast<ComplexScalar>(); \
+    m_matT = matrix.derived().template cast<ComplexScalar>(); \
     if(computeU)  m_matU = ComplexMatrixType::Identity(1,1); \
       m_info = Success; \
       m_isInitialized = true; \
       m_matUisUptodate = computeU; \
       return *this; \
   } \
-  lapack_int n = matrix.cols(), sdim, info; \
-  lapack_int lda = matrix.outerStride(); \
-  lapack_int matrix_order = MKLCOLROW; \
+  lapack_int n = internal::convert_index<lapack_int>(matrix.cols()), sdim, info; \
+  lapack_int matrix_order = LAPACKE_COLROW; \
   char jobvs, sort='N'; \
-  LAPACK_##MKLPREFIX_U##_SELECT1 select = 0; \
+  LAPACK_##LAPACKE_PREFIX_U##_SELECT1 select = 0; \
   jobvs = (computeU) ? 'V' : 'N'; \
   m_matU.resize(n, n); \
-  lapack_int ldvs  = m_matU.outerStride(); \
+  lapack_int ldvs  = internal::convert_index<lapack_int>(m_matU.outerStride()); \
   m_matT = matrix; \
+  lapack_int lda = internal::convert_index<lapack_int>(m_matT.outerStride()); \
   Matrix<EIGTYPE, Dynamic, Dynamic> w; \
   w.resize(n, 1);\
-  info = LAPACKE_##MKLPREFIX##gees( matrix_order, jobvs, sort, select, n, (MKLTYPE*)m_matT.data(), lda, &sdim, (MKLTYPE*)w.data(), (MKLTYPE*)m_matU.data(), ldvs ); \
+  info = LAPACKE_##LAPACKE_PREFIX##gees( matrix_order, jobvs, sort, select, n, (LAPACKE_TYPE*)m_matT.data(), lda, &sdim, (LAPACKE_TYPE*)w.data(), (LAPACKE_TYPE*)m_matU.data(), ldvs ); \
   if(info == 0) \
     m_info = Success; \
   else \
@@ -84,11 +81,11 @@
 \
 }
 
-EIGEN_MKL_SCHUR_COMPLEX(dcomplex, MKL_Complex16, z, Z, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_SCHUR_COMPLEX(scomplex, MKL_Complex8,  c, C, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_SCHUR_COMPLEX(dcomplex, MKL_Complex16, z, Z, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_SCHUR_COMPLEX(scomplex, MKL_Complex8,  c, C, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SCHUR_COMPLEX(dcomplex, lapack_complex_double, z, Z, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_COMPLEX(scomplex, lapack_complex_float,  c, C, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_COMPLEX(dcomplex, lapack_complex_double, z, Z, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SCHUR_COMPLEX(scomplex, lapack_complex_float,  c, C, RowMajor, LAPACK_ROW_MAJOR)
 
 } // end namespace Eigen
 
-#endif // EIGEN_COMPLEX_SCHUR_MKL_H
+#endif // EIGEN_COMPLEX_SCHUR_LAPACKE_H
diff --git a/Eigen/src/Eigenvalues/EigenSolver.h b/Eigen/src/Eigenvalues/EigenSolver.h
index 20c59a7..f205b18 100644
--- a/Eigen/src/Eigenvalues/EigenSolver.h
+++ b/Eigen/src/Eigenvalues/EigenSolver.h
@@ -79,7 +79,7 @@
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for #MatrixType. 
       *
@@ -110,7 +110,7 @@
       *
       * \sa compute() for an example.
       */
- EigenSolver() : m_eivec(), m_eivalues(), m_isInitialized(false), m_realSchur(), m_matT(), m_tmp() {}
+    EigenSolver() : m_eivec(), m_eivalues(), m_isInitialized(false), m_realSchur(), m_matT(), m_tmp() {}
 
     /** \brief Default constructor with memory preallocation
       *
@@ -118,7 +118,7 @@
       * according to the specified problem \a size.
       * \sa EigenSolver()
       */
-    EigenSolver(Index size)
+    explicit EigenSolver(Index size)
       : m_eivec(size, size),
         m_eivalues(size),
         m_isInitialized(false),
@@ -143,7 +143,8 @@
       *
       * \sa compute()
       */
-    EigenSolver(const MatrixType& matrix, bool computeEigenvectors = true)
+    template<typename InputType>
+    explicit EigenSolver(const EigenBase<InputType>& matrix, bool computeEigenvectors = true)
       : m_eivec(matrix.rows(), matrix.cols()),
         m_eivalues(matrix.cols()),
         m_isInitialized(false),
@@ -152,7 +153,7 @@
         m_matT(matrix.rows(), matrix.cols()), 
         m_tmp(matrix.cols())
     {
-      compute(matrix, computeEigenvectors);
+      compute(matrix.derived(), computeEigenvectors);
     }
 
     /** \brief Returns the eigenvectors of given matrix. 
@@ -273,12 +274,14 @@
       * Example: \include EigenSolver_compute.cpp
       * Output: \verbinclude EigenSolver_compute.out
       */
-    EigenSolver& compute(const MatrixType& matrix, bool computeEigenvectors = true);
+    template<typename InputType>
+    EigenSolver& compute(const EigenBase<InputType>& matrix, bool computeEigenvectors = true);
 
+    /** \returns NumericalIssue if the input contains INF or NaN values or overflow occured. Returns Success otherwise. */
     ComputationInfo info() const
     {
       eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
-      return m_realSchur.info();
+      return m_info;
     }
 
     /** \brief Sets the maximum number of iterations allowed. */
@@ -309,6 +312,7 @@
     EigenvalueType m_eivalues;
     bool m_isInitialized;
     bool m_eigenvectorsOk;
+    ComputationInfo m_info;
     RealSchur<MatrixType> m_realSchur;
     MatrixType m_matT;
 
@@ -320,11 +324,12 @@
 MatrixType EigenSolver<MatrixType>::pseudoEigenvalueMatrix() const
 {
   eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
+  const RealScalar precision = RealScalar(2)*NumTraits<RealScalar>::epsilon();
   Index n = m_eivalues.rows();
   MatrixType matD = MatrixType::Zero(n,n);
   for (Index i=0; i<n; ++i)
   {
-    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(i)), numext::real(m_eivalues.coeff(i))))
+    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(i)), numext::real(m_eivalues.coeff(i)), precision))
       matD.coeffRef(i,i) = numext::real(m_eivalues.coeff(i));
     else
     {
@@ -341,11 +346,12 @@
 {
   eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
   eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
+  const RealScalar precision = RealScalar(2)*NumTraits<RealScalar>::epsilon();
   Index n = m_eivec.cols();
   EigenvectorsType matV(n,n);
   for (Index j=0; j<n; ++j)
   {
-    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(j)), numext::real(m_eivalues.coeff(j))) || j+1==n)
+    if (internal::isMuchSmallerThan(numext::imag(m_eivalues.coeff(j)), numext::real(m_eivalues.coeff(j)), precision) || j+1==n)
     {
       // we have a real eigen value
       matV.col(j) = m_eivec.col(j).template cast<ComplexScalar>();
@@ -368,19 +374,23 @@
 }
 
 template<typename MatrixType>
+template<typename InputType>
 EigenSolver<MatrixType>& 
-EigenSolver<MatrixType>::compute(const MatrixType& matrix, bool computeEigenvectors)
+EigenSolver<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeEigenvectors)
 {
   check_template_parameters();
   
   using std::sqrt;
   using std::abs;
+  using numext::isfinite;
   eigen_assert(matrix.cols() == matrix.rows());
 
   // Reduce to real Schur form.
-  m_realSchur.compute(matrix, computeEigenvectors);
+  m_realSchur.compute(matrix.derived(), computeEigenvectors);
+  
+  m_info = m_realSchur.info();
 
-  if (m_realSchur.info() == Success)
+  if (m_info == Success)
   {
     m_matT = m_realSchur.matrixT();
     if (computeEigenvectors)
@@ -394,14 +404,40 @@
       if (i == matrix.cols() - 1 || m_matT.coeff(i+1, i) == Scalar(0)) 
       {
         m_eivalues.coeffRef(i) = m_matT.coeff(i, i);
+        if(!(isfinite)(m_eivalues.coeffRef(i)))
+        {
+          m_isInitialized = true;
+          m_eigenvectorsOk = false;
+          m_info = NumericalIssue;
+          return *this;
+        }
         ++i;
       }
       else
       {
         Scalar p = Scalar(0.5) * (m_matT.coeff(i, i) - m_matT.coeff(i+1, i+1));
-        Scalar z = sqrt(abs(p * p + m_matT.coeff(i+1, i) * m_matT.coeff(i, i+1)));
+        Scalar z;
+        // Compute z = sqrt(abs(p * p + m_matT.coeff(i+1, i) * m_matT.coeff(i, i+1)));
+        // without overflow
+        {
+          Scalar t0 = m_matT.coeff(i+1, i);
+          Scalar t1 = m_matT.coeff(i, i+1);
+          Scalar maxval = numext::maxi<Scalar>(abs(p),numext::maxi<Scalar>(abs(t0),abs(t1)));
+          t0 /= maxval;
+          t1 /= maxval;
+          Scalar p0 = p/maxval;
+          z = maxval * sqrt(abs(p0 * p0 + t0 * t1));
+        }
+        
         m_eivalues.coeffRef(i)   = ComplexScalar(m_matT.coeff(i+1, i+1) + p, z);
         m_eivalues.coeffRef(i+1) = ComplexScalar(m_matT.coeff(i+1, i+1) + p, -z);
+        if(!((isfinite)(m_eivalues.coeffRef(i)) && (isfinite)(m_eivalues.coeffRef(i+1))))
+        {
+          m_isInitialized = true;
+          m_eigenvectorsOk = false;
+          m_info = NumericalIssue;
+          return *this;
+        }
         i += 2;
       }
     }
@@ -417,26 +453,6 @@
   return *this;
 }
 
-// Complex scalar division.
-template<typename Scalar>
-std::complex<Scalar> cdiv(const Scalar& xr, const Scalar& xi, const Scalar& yr, const Scalar& yi)
-{
-  using std::abs;
-  Scalar r,d;
-  if (abs(yr) > abs(yi))
-  {
-      r = yi/yr;
-      d = yr + r*yi;
-      return std::complex<Scalar>((xr + r*xi)/d, (xi - r*xr)/d);
-  }
-  else
-  {
-      r = yr/yi;
-      d = yi + r*yr;
-      return std::complex<Scalar>((r*xr + xi)/d, (r*xi - xr)/d);
-  }
-}
-
 
 template<typename MatrixType>
 void EigenSolver<MatrixType>::doComputeEigenvectors()
@@ -453,7 +469,7 @@
   }
   
   // Backsubstitute to find vectors of upper triangular form
-  if (norm == 0.0)
+  if (norm == Scalar(0))
   {
     return;
   }
@@ -469,13 +485,13 @@
       Scalar lastr(0), lastw(0);
       Index l = n;
 
-      m_matT.coeffRef(n,n) = 1.0;
+      m_matT.coeffRef(n,n) = Scalar(1);
       for (Index i = n-1; i >= 0; i--)
       {
         Scalar w = m_matT.coeff(i,i) - p;
         Scalar r = m_matT.row(i).segment(l,n-l+1).dot(m_matT.col(n).segment(l, n-l+1));
 
-        if (m_eivalues.coeff(i).imag() < 0.0)
+        if (m_eivalues.coeff(i).imag() < Scalar(0))
         {
           lastw = w;
           lastr = r;
@@ -483,9 +499,9 @@
         else
         {
           l = i;
-          if (m_eivalues.coeff(i).imag() == 0.0)
+          if (m_eivalues.coeff(i).imag() == Scalar(0))
           {
-            if (w != 0.0)
+            if (w != Scalar(0))
               m_matT.coeffRef(i,n) = -r / w;
             else
               m_matT.coeffRef(i,n) = -r / (eps * norm);
@@ -523,19 +539,19 @@
       }
       else
       {
-        std::complex<Scalar> cc = cdiv<Scalar>(0.0,-m_matT.coeff(n-1,n),m_matT.coeff(n-1,n-1)-p,q);
+        ComplexScalar cc = ComplexScalar(Scalar(0),-m_matT.coeff(n-1,n)) / ComplexScalar(m_matT.coeff(n-1,n-1)-p,q);
         m_matT.coeffRef(n-1,n-1) = numext::real(cc);
         m_matT.coeffRef(n-1,n) = numext::imag(cc);
       }
-      m_matT.coeffRef(n,n-1) = 0.0;
-      m_matT.coeffRef(n,n) = 1.0;
+      m_matT.coeffRef(n,n-1) = Scalar(0);
+      m_matT.coeffRef(n,n) = Scalar(1);
       for (Index i = n-2; i >= 0; i--)
       {
         Scalar ra = m_matT.row(i).segment(l, n-l+1).dot(m_matT.col(n-1).segment(l, n-l+1));
         Scalar sa = m_matT.row(i).segment(l, n-l+1).dot(m_matT.col(n).segment(l, n-l+1));
         Scalar w = m_matT.coeff(i,i) - p;
 
-        if (m_eivalues.coeff(i).imag() < 0.0)
+        if (m_eivalues.coeff(i).imag() < Scalar(0))
         {
           lastw = w;
           lastra = ra;
@@ -546,7 +562,7 @@
           l = i;
           if (m_eivalues.coeff(i).imag() == RealScalar(0))
           {
-            std::complex<Scalar> cc = cdiv(-ra,-sa,w,q);
+            ComplexScalar cc = ComplexScalar(-ra,-sa) / ComplexScalar(w,q);
             m_matT.coeffRef(i,n-1) = numext::real(cc);
             m_matT.coeffRef(i,n) = numext::imag(cc);
           }
@@ -557,10 +573,10 @@
             Scalar y = m_matT.coeff(i+1,i);
             Scalar vr = (m_eivalues.coeff(i).real() - p) * (m_eivalues.coeff(i).real() - p) + m_eivalues.coeff(i).imag() * m_eivalues.coeff(i).imag() - q * q;
             Scalar vi = (m_eivalues.coeff(i).real() - p) * Scalar(2) * q;
-            if ((vr == 0.0) && (vi == 0.0))
+            if ((vr == Scalar(0)) && (vi == Scalar(0)))
               vr = eps * norm * (abs(w) + abs(q) + abs(x) + abs(y) + abs(lastw));
 
-            std::complex<Scalar> cc = cdiv(x*lastra-lastw*ra+q*sa,x*lastsa-lastw*sa-q*ra,vr,vi);
+            ComplexScalar cc = ComplexScalar(x*lastra-lastw*ra+q*sa,x*lastsa-lastw*sa-q*ra) / ComplexScalar(vr,vi);
             m_matT.coeffRef(i,n-1) = numext::real(cc);
             m_matT.coeffRef(i,n) = numext::imag(cc);
             if (abs(x) > (abs(lastw) + abs(q)))
@@ -570,15 +586,14 @@
             }
             else
             {
-              cc = cdiv(-lastra-y*m_matT.coeff(i,n-1),-lastsa-y*m_matT.coeff(i,n),lastw,q);
+              cc = ComplexScalar(-lastra-y*m_matT.coeff(i,n-1),-lastsa-y*m_matT.coeff(i,n)) / ComplexScalar(lastw,q);
               m_matT.coeffRef(i+1,n-1) = numext::real(cc);
               m_matT.coeffRef(i+1,n) = numext::imag(cc);
             }
           }
 
           // Overflow control
-          using std::max;
-          Scalar t = (max)(abs(m_matT.coeff(i,n-1)),abs(m_matT.coeff(i,n)));
+          Scalar t = numext::maxi<Scalar>(abs(m_matT.coeff(i,n-1)),abs(m_matT.coeff(i,n)));
           if ((eps * t) * t > Scalar(1))
             m_matT.block(i, n-1, size-i, 2) /= t;
 
@@ -590,7 +605,7 @@
     }
     else
     {
-      eigen_assert(0 && "Internal bug in EigenSolver"); // this should not happen
+      eigen_assert(0 && "Internal bug in EigenSolver (INF or NaN has not been detected)"); // this should not happen
     }
   }
 
diff --git a/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h b/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h
index 956e80d..87d789b 100644
--- a/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h
+++ b/Eigen/src/Eigenvalues/GeneralizedEigenSolver.h
@@ -1,8 +1,9 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2012 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2012-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2010,2012 Jitse Niesen <jitse@maths.leeds.ac.uk>
+// Copyright (C) 2016 Tobias Wood <tobias@spinicist.org.uk>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -72,7 +73,7 @@
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Complex scalar type for #MatrixType. 
       *
@@ -89,7 +90,7 @@
       */
     typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> VectorType;
 
-    /** \brief Type for vector of complex scalar values eigenvalues as returned by betas().
+    /** \brief Type for vector of complex scalar values eigenvalues as returned by alphas().
       *
       * This is a column vector with entries of type #ComplexScalar.
       * The length of the vector is the size of #MatrixType.
@@ -114,7 +115,14 @@
       *
       * \sa compute() for an example.
       */
-    GeneralizedEigenSolver() : m_eivec(), m_alphas(), m_betas(), m_isInitialized(false), m_realQZ(), m_matS(), m_tmp() {}
+    GeneralizedEigenSolver()
+      : m_eivec(),
+        m_alphas(),
+        m_betas(),
+        m_valuesOkay(false),
+        m_vectorsOkay(false),
+        m_realQZ()
+    {}
 
     /** \brief Default constructor with memory preallocation
       *
@@ -122,14 +130,13 @@
       * according to the specified problem \a size.
       * \sa GeneralizedEigenSolver()
       */
-    GeneralizedEigenSolver(Index size)
+    explicit GeneralizedEigenSolver(Index size)
       : m_eivec(size, size),
         m_alphas(size),
         m_betas(size),
-        m_isInitialized(false),
-        m_eigenvectorsOk(false),
+        m_valuesOkay(false),
+        m_vectorsOkay(false),
         m_realQZ(size),
-        m_matS(size, size),
         m_tmp(size)
     {}
 
@@ -149,10 +156,9 @@
       : m_eivec(A.rows(), A.cols()),
         m_alphas(A.cols()),
         m_betas(A.cols()),
-        m_isInitialized(false),
-        m_eigenvectorsOk(false),
+        m_valuesOkay(false),
+        m_vectorsOkay(false),
         m_realQZ(A.cols()),
-        m_matS(A.rows(), A.cols()),
         m_tmp(A.cols())
     {
       compute(A, B, computeEigenvectors);
@@ -160,22 +166,20 @@
 
     /* \brief Returns the computed generalized eigenvectors.
       *
-      * \returns  %Matrix whose columns are the (possibly complex) eigenvectors.
+      * \returns  %Matrix whose columns are the (possibly complex) right eigenvectors.
+      * i.e. the eigenvectors that solve (A - l*B)x = 0. The ordering matches the eigenvalues.
       *
       * \pre Either the constructor 
       * GeneralizedEigenSolver(const MatrixType&,const MatrixType&, bool) or the member function
       * compute(const MatrixType&, const MatrixType& bool) has been called before, and
       * \p computeEigenvectors was set to true (the default).
       *
-      * Column \f$ k \f$ of the returned matrix is an eigenvector corresponding
-      * to eigenvalue number \f$ k \f$ as returned by eigenvalues().  The
-      * eigenvectors are normalized to have (Euclidean) norm equal to one. The
-      * matrix returned by this function is the matrix \f$ V \f$ in the
-      * generalized eigendecomposition \f$ A = B V D V^{-1} \f$, if it exists.
-      *
       * \sa eigenvalues()
       */
-//    EigenvectorsType eigenvectors() const;
+    EigenvectorsType eigenvectors() const {
+      eigen_assert(m_vectorsOkay && "Eigenvectors for GeneralizedEigenSolver were not calculated.");
+      return m_eivec;
+    }
 
     /** \brief Returns an expression of the computed generalized eigenvalues.
       *
@@ -197,7 +201,7 @@
       */
     EigenvalueType eigenvalues() const
     {
-      eigen_assert(m_isInitialized && "GeneralizedEigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "GeneralizedEigenSolver is not initialized.");
       return EigenvalueType(m_alphas,m_betas);
     }
 
@@ -208,7 +212,7 @@
       * \sa betas(), eigenvalues() */
     ComplexVectorType alphas() const
     {
-      eigen_assert(m_isInitialized && "GeneralizedEigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "GeneralizedEigenSolver is not initialized.");
       return m_alphas;
     }
 
@@ -219,7 +223,7 @@
       * \sa alphas(), eigenvalues() */
     VectorType betas() const
     {
-      eigen_assert(m_isInitialized && "GeneralizedEigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "GeneralizedEigenSolver is not initialized.");
       return m_betas;
     }
 
@@ -250,7 +254,7 @@
 
     ComputationInfo info() const
     {
-      eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
+      eigen_assert(m_valuesOkay && "EigenSolver is not initialized.");
       return m_realQZ.info();
     }
 
@@ -270,29 +274,14 @@
       EIGEN_STATIC_ASSERT(!NumTraits<Scalar>::IsComplex, NUMERIC_TYPE_MUST_BE_REAL);
     }
     
-    MatrixType m_eivec;
+    EigenvectorsType m_eivec;
     ComplexVectorType m_alphas;
     VectorType m_betas;
-    bool m_isInitialized;
-    bool m_eigenvectorsOk;
+    bool m_valuesOkay, m_vectorsOkay;
     RealQZ<MatrixType> m_realQZ;
-    MatrixType m_matS;
-
-    typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> ColumnVectorType;
-    ColumnVectorType m_tmp;
+    ComplexVectorType m_tmp;
 };
 
-//template<typename MatrixType>
-//typename GeneralizedEigenSolver<MatrixType>::EigenvectorsType GeneralizedEigenSolver<MatrixType>::eigenvectors() const
-//{
-//  eigen_assert(m_isInitialized && "EigenSolver is not initialized.");
-//  eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
-//  Index n = m_eivec.cols();
-//  EigenvectorsType matV(n,n);
-//  // TODO
-//  return matV;
-//}
-
 template<typename MatrixType>
 GeneralizedEigenSolver<MatrixType>&
 GeneralizedEigenSolver<MatrixType>::compute(const MatrixType& A, const MatrixType& B, bool computeEigenvectors)
@@ -302,46 +291,125 @@
   using std::sqrt;
   using std::abs;
   eigen_assert(A.cols() == A.rows() && B.cols() == A.rows() && B.cols() == B.rows());
-
+  Index size = A.cols();
+  m_valuesOkay = false;
+  m_vectorsOkay = false;
   // Reduce to generalized real Schur form:
   // A = Q S Z and B = Q T Z
   m_realQZ.compute(A, B, computeEigenvectors);
-
   if (m_realQZ.info() == Success)
   {
-    m_matS = m_realQZ.matrixS();
+    // Resize storage
+    m_alphas.resize(size);
+    m_betas.resize(size);
     if (computeEigenvectors)
-      m_eivec = m_realQZ.matrixZ().transpose();
-  
-    // Compute eigenvalues from matS
-    m_alphas.resize(A.cols());
-    m_betas.resize(A.cols());
-    Index i = 0;
-    while (i < A.cols())
     {
-      if (i == A.cols() - 1 || m_matS.coeff(i+1, i) == Scalar(0))
+      m_eivec.resize(size,size);
+      m_tmp.resize(size);
+    }
+
+    // Aliases:
+    Map<VectorType> v(reinterpret_cast<Scalar*>(m_tmp.data()), size);
+    ComplexVectorType &cv = m_tmp;
+    const MatrixType &mS = m_realQZ.matrixS();
+    const MatrixType &mT = m_realQZ.matrixT();
+
+    Index i = 0;
+    while (i < size)
+    {
+      if (i == size - 1 || mS.coeff(i+1, i) == Scalar(0))
       {
-        m_alphas.coeffRef(i) = m_matS.coeff(i, i);
-        m_betas.coeffRef(i)  = m_realQZ.matrixT().coeff(i,i);
+        // Real eigenvalue
+        m_alphas.coeffRef(i) = mS.diagonal().coeff(i);
+        m_betas.coeffRef(i)  = mT.diagonal().coeff(i);
+        if (computeEigenvectors)
+        {
+          v.setConstant(Scalar(0.0));
+          v.coeffRef(i) = Scalar(1.0);
+          // For singular eigenvalues do nothing more
+          if(abs(m_betas.coeffRef(i)) >= (std::numeric_limits<RealScalar>::min)())
+          {
+            // Non-singular eigenvalue
+            const Scalar alpha = real(m_alphas.coeffRef(i));
+            const Scalar beta = m_betas.coeffRef(i);
+            for (Index j = i-1; j >= 0; j--)
+            {
+              const Index st = j+1;
+              const Index sz = i-j;
+              if (j > 0 && mS.coeff(j, j-1) != Scalar(0))
+              {
+                // 2x2 block
+                Matrix<Scalar, 2, 1> rhs = (alpha*mT.template block<2,Dynamic>(j-1,st,2,sz) - beta*mS.template block<2,Dynamic>(j-1,st,2,sz)) .lazyProduct( v.segment(st,sz) );
+                Matrix<Scalar, 2, 2> lhs = beta * mS.template block<2,2>(j-1,j-1) - alpha * mT.template block<2,2>(j-1,j-1);
+                v.template segment<2>(j-1) = lhs.partialPivLu().solve(rhs);
+                j--;
+              }
+              else
+              {
+                v.coeffRef(j) = -v.segment(st,sz).transpose().cwiseProduct(beta*mS.block(j,st,1,sz) - alpha*mT.block(j,st,1,sz)).sum() / (beta*mS.coeffRef(j,j) - alpha*mT.coeffRef(j,j));
+              }
+            }
+          }
+          m_eivec.col(i).real().noalias() = m_realQZ.matrixZ().transpose() * v;
+          m_eivec.col(i).real().normalize();
+          m_eivec.col(i).imag().setConstant(0);
+        }
         ++i;
       }
       else
       {
-        Scalar p = Scalar(0.5) * (m_matS.coeff(i, i) - m_matS.coeff(i+1, i+1));
-        Scalar z = sqrt(abs(p * p + m_matS.coeff(i+1, i) * m_matS.coeff(i, i+1)));
-        m_alphas.coeffRef(i)   = ComplexScalar(m_matS.coeff(i+1, i+1) + p, z);
-        m_alphas.coeffRef(i+1) = ComplexScalar(m_matS.coeff(i+1, i+1) + p, -z);
+        // We need to extract the generalized eigenvalues of the pair of a general 2x2 block S and a positive diagonal 2x2 block T
+        // Then taking beta=T_00*T_11, we can avoid any division, and alpha is the eigenvalues of A = (U^-1 * S * U) * diag(T_11,T_00):
 
-        m_betas.coeffRef(i)   = m_realQZ.matrixT().coeff(i,i);
-        m_betas.coeffRef(i+1) = m_realQZ.matrixT().coeff(i,i);
+        // T =  [a 0]
+        //      [0 b]
+        RealScalar a = mT.diagonal().coeff(i),
+                   b = mT.diagonal().coeff(i+1);
+        const RealScalar beta = m_betas.coeffRef(i) = m_betas.coeffRef(i+1) = a*b;
+
+        // ^^ NOTE: using diagonal()(i) instead of coeff(i,i) workarounds a MSVC bug.
+        Matrix<RealScalar,2,2> S2 = mS.template block<2,2>(i,i) * Matrix<Scalar,2,1>(b,a).asDiagonal();
+
+        Scalar p = Scalar(0.5) * (S2.coeff(0,0) - S2.coeff(1,1));
+        Scalar z = sqrt(abs(p * p + S2.coeff(1,0) * S2.coeff(0,1)));
+        const ComplexScalar alpha = ComplexScalar(S2.coeff(1,1) + p, (beta > 0) ? z : -z);
+        m_alphas.coeffRef(i)   = conj(alpha);
+        m_alphas.coeffRef(i+1) = alpha;
+
+        if (computeEigenvectors) {
+          // Compute eigenvector in position (i+1) and then position (i) is just the conjugate
+          cv.setZero();
+          cv.coeffRef(i+1) = Scalar(1.0);
+          // here, the "static_cast" workaound expression template issues.
+          cv.coeffRef(i) = -(static_cast<Scalar>(beta*mS.coeffRef(i,i+1)) - alpha*mT.coeffRef(i,i+1))
+                          / (static_cast<Scalar>(beta*mS.coeffRef(i,i))   - alpha*mT.coeffRef(i,i));
+          for (Index j = i-1; j >= 0; j--)
+          {
+            const Index st = j+1;
+            const Index sz = i+1-j;
+            if (j > 0 && mS.coeff(j, j-1) != Scalar(0))
+            {
+              // 2x2 block
+              Matrix<ComplexScalar, 2, 1> rhs = (alpha*mT.template block<2,Dynamic>(j-1,st,2,sz) - beta*mS.template block<2,Dynamic>(j-1,st,2,sz)) .lazyProduct( cv.segment(st,sz) );
+              Matrix<ComplexScalar, 2, 2> lhs = beta * mS.template block<2,2>(j-1,j-1) - alpha * mT.template block<2,2>(j-1,j-1);
+              cv.template segment<2>(j-1) = lhs.partialPivLu().solve(rhs);
+              j--;
+            } else {
+              cv.coeffRef(j) =  cv.segment(st,sz).transpose().cwiseProduct(beta*mS.block(j,st,1,sz) - alpha*mT.block(j,st,1,sz)).sum()
+                              / (alpha*mT.coeffRef(j,j) - static_cast<Scalar>(beta*mS.coeffRef(j,j)));
+            }
+          }
+          m_eivec.col(i+1).noalias() = (m_realQZ.matrixZ().transpose() * cv);
+          m_eivec.col(i+1).normalize();
+          m_eivec.col(i) = m_eivec.col(i+1).conjugate();
+        }
         i += 2;
       }
     }
+
+    m_valuesOkay = true;
+    m_vectorsOkay = computeEigenvectors;
   }
-
-  m_isInitialized = true;
-  m_eigenvectorsOk = false;//computeEigenvectors;
-
   return *this;
 }
 
diff --git a/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h b/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h
index 07bf1ea..5f6bb82 100644
--- a/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h
+++ b/Eigen/src/Eigenvalues/GeneralizedSelfAdjointEigenSolver.h
@@ -50,7 +50,6 @@
     typedef SelfAdjointEigenSolver<_MatrixType> Base;
   public:
 
-    typedef typename Base::Index Index;
     typedef _MatrixType MatrixType;
 
     /** \brief Default constructor for fixed-size matrices.
@@ -74,7 +73,7 @@
       *
       * \sa compute() for an example
       */
-    GeneralizedSelfAdjointEigenSolver(Index size)
+    explicit GeneralizedSelfAdjointEigenSolver(Index size)
         : Base(size)
     {}
 
diff --git a/Eigen/src/Eigenvalues/HessenbergDecomposition.h b/Eigen/src/Eigenvalues/HessenbergDecomposition.h
index 3db0c01..f647f69 100644
--- a/Eigen/src/Eigenvalues/HessenbergDecomposition.h
+++ b/Eigen/src/Eigenvalues/HessenbergDecomposition.h
@@ -71,7 +71,7 @@
 
     /** \brief Scalar type for matrices of type #MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     /** \brief Type for vector of Householder coefficients.
       *
@@ -97,7 +97,7 @@
       *
       * \sa compute() for an example.
       */
-    HessenbergDecomposition(Index size = Size==Dynamic ? 2 : Size)
+    explicit HessenbergDecomposition(Index size = Size==Dynamic ? 2 : Size)
       : m_matrix(size,size),
         m_temp(size),
         m_isInitialized(false)
@@ -115,8 +115,9 @@
       *
       * \sa matrixH() for an example.
       */
-    HessenbergDecomposition(const MatrixType& matrix)
-      : m_matrix(matrix),
+    template<typename InputType>
+    explicit HessenbergDecomposition(const EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
         m_temp(matrix.rows()),
         m_isInitialized(false)
     {
@@ -147,9 +148,10 @@
       * Example: \include HessenbergDecomposition_compute.cpp
       * Output: \verbinclude HessenbergDecomposition_compute.out
       */
-    HessenbergDecomposition& compute(const MatrixType& matrix)
+    template<typename InputType>
+    HessenbergDecomposition& compute(const EigenBase<InputType>& matrix)
     {
-      m_matrix = matrix;
+      m_matrix = matrix.derived();
       if(matrix.rows()<2)
       {
         m_isInitialized = true;
@@ -337,7 +339,6 @@
 template<typename MatrixType> struct HessenbergDecompositionMatrixHReturnType
 : public ReturnByValue<HessenbergDecompositionMatrixHReturnType<MatrixType> >
 {
-    typedef typename MatrixType::Index Index;
   public:
     /** \brief Constructor.
       *
diff --git a/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h b/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h
index 4fec8af..e4e4260 100644
--- a/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h
+++ b/Eigen/src/Eigenvalues/MatrixBaseEigenvalues.h
@@ -66,7 +66,6 @@
 inline typename MatrixBase<Derived>::EigenvaluesReturnType
 MatrixBase<Derived>::eigenvalues() const
 {
-  typedef typename internal::traits<Derived>::Scalar Scalar;
   return internal::eigenvalues_selector<Derived, NumTraits<Scalar>::IsComplex>::run(derived());
 }
 
@@ -88,7 +87,6 @@
 inline typename SelfAdjointView<MatrixType, UpLo>::EigenvaluesReturnType
 SelfAdjointView<MatrixType, UpLo>::eigenvalues() const
 {
-  typedef typename SelfAdjointView<MatrixType, UpLo>::PlainObject PlainObject;
   PlainObject thisAsMatrix(*this);
   return SelfAdjointEigenSolver<PlainObject>(thisAsMatrix, false).eigenvalues();
 }
diff --git a/Eigen/src/Eigenvalues/RealQZ.h b/Eigen/src/Eigenvalues/RealQZ.h
index aa3833e..b3a910d 100644
--- a/Eigen/src/Eigenvalues/RealQZ.h
+++ b/Eigen/src/Eigenvalues/RealQZ.h
@@ -67,7 +67,7 @@
       };
       typedef typename MatrixType::Scalar Scalar;
       typedef std::complex<typename NumTraits<Scalar>::Real> ComplexScalar;
-      typedef typename MatrixType::Index Index;
+      typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
       typedef Matrix<ComplexScalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> EigenvalueType;
       typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> ColumnVectorType;
@@ -83,7 +83,7 @@
        *
        * \sa compute() for an example.
        */
-      RealQZ(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime) : 
+      explicit RealQZ(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime) :
         m_S(size, size),
         m_T(size, size),
         m_Q(size, size),
@@ -276,7 +276,7 @@
 
   /** \internal Look for single small sub-diagonal element S(res, res-1) and return res (or 0) */
   template<typename MatrixType>
-    inline typename MatrixType::Index RealQZ<MatrixType>::findSmallSubdiagEntry(Index iu)
+    inline Index RealQZ<MatrixType>::findSmallSubdiagEntry(Index iu)
     {
       using std::abs;
       Index res = iu;
@@ -294,7 +294,7 @@
 
   /** \internal Look for single small diagonal element T(res, res) for res between f and l, and return res (or f-1)  */
   template<typename MatrixType>
-    inline typename MatrixType::Index RealQZ<MatrixType>::findSmallDiagEntry(Index f, Index l)
+    inline Index RealQZ<MatrixType>::findSmallDiagEntry(Index f, Index l)
     {
       using std::abs;
       Index res = l;
@@ -315,8 +315,8 @@
       const Index dim=m_S.cols();
       if (abs(m_S.coeff(i+1,i))==Scalar(0))
         return;
-      Index z = findSmallDiagEntry(i,i+1);
-      if (z==i-1)
+      Index j = findSmallDiagEntry(i,i+1);
+      if (j==i-1)
       {
         // block of (S T^{-1})
         Matrix2s STi = m_T.template block<2,2>(i,i).template triangularView<Upper>().
@@ -352,7 +352,7 @@
       }
       else
       {
-        pushDownZero(z,i,i+1);
+        pushDownZero(j,i,i+1);
       }
     }
 
@@ -552,7 +552,6 @@
       m_T.coeffRef(l,l-1) = Scalar(0.0);
     }
 
-
   template<typename MatrixType>
     RealQZ<MatrixType>& RealQZ<MatrixType>::compute(const MatrixType& A_in, const MatrixType& B_in, bool computeQZ)
     {
@@ -616,6 +615,37 @@
       }
       // check if we converged before reaching iterations limit
       m_info = (local_iter<m_maxIters) ? Success : NoConvergence;
+
+      // For each non triangular 2x2 diagonal block of S,
+      //    reduce the respective 2x2 diagonal block of T to positive diagonal form using 2x2 SVD.
+      // This step is not mandatory for QZ, but it does help further extraction of eigenvalues/eigenvectors,
+      // and is in par with Lapack/Matlab QZ.
+      if(m_info==Success)
+      {
+        for(Index i=0; i<dim-1; ++i)
+        {
+          if(m_S.coeff(i+1, i) != Scalar(0))
+          {
+            JacobiRotation<Scalar> j_left, j_right;
+            internal::real_2x2_jacobi_svd(m_T, i, i+1, &j_left, &j_right);
+
+            // Apply resulting Jacobi rotations
+            m_S.applyOnTheLeft(i,i+1,j_left);
+            m_S.applyOnTheRight(i,i+1,j_right);
+            m_T.applyOnTheLeft(i,i+1,j_left);
+            m_T.applyOnTheRight(i,i+1,j_right);
+            m_T(i+1,i) = m_T(i,i+1) = Scalar(0);
+
+            if(m_computeQZ) {
+              m_Q.applyOnTheRight(i,i+1,j_left.transpose());
+              m_Z.applyOnTheLeft(i,i+1,j_right.transpose());
+            }
+
+            i++;
+          }
+        }
+      }
+
       return *this;
     } // end compute
 
diff --git a/Eigen/src/Eigenvalues/RealSchur.h b/Eigen/src/Eigenvalues/RealSchur.h
index 16d3875..17ea903 100644
--- a/Eigen/src/Eigenvalues/RealSchur.h
+++ b/Eigen/src/Eigenvalues/RealSchur.h
@@ -64,7 +64,7 @@
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef std::complex<typename NumTraits<Scalar>::Real> ComplexScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     typedef Matrix<ComplexScalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> EigenvalueType;
     typedef Matrix<Scalar, ColsAtCompileTime, 1, Options & ~RowMajor, MaxColsAtCompileTime, 1> ColumnVectorType;
@@ -80,7 +80,7 @@
       *
       * \sa compute() for an example.
       */
-    RealSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
+    explicit RealSchur(Index size = RowsAtCompileTime==Dynamic ? 1 : RowsAtCompileTime)
             : m_matT(size, size),
               m_matU(size, size),
               m_workspaceVector(size),
@@ -100,7 +100,8 @@
       * Example: \include RealSchur_RealSchur_MatrixType.cpp
       * Output: \verbinclude RealSchur_RealSchur_MatrixType.out
       */
-    RealSchur(const MatrixType& matrix, bool computeU = true)
+    template<typename InputType>
+    explicit RealSchur(const EigenBase<InputType>& matrix, bool computeU = true)
             : m_matT(matrix.rows(),matrix.cols()),
               m_matU(matrix.rows(),matrix.cols()),
               m_workspaceVector(matrix.rows()),
@@ -109,7 +110,7 @@
               m_matUisUptodate(false),
               m_maxIters(-1)
     {
-      compute(matrix, computeU);
+      compute(matrix.derived(), computeU);
     }
 
     /** \brief Returns the orthogonal matrix in the Schur decomposition. 
@@ -165,7 +166,8 @@
       *
       * \sa compute(const MatrixType&, bool, Index)
       */
-    RealSchur& compute(const MatrixType& matrix, bool computeU = true);
+    template<typename InputType>
+    RealSchur& compute(const EigenBase<InputType>& matrix, bool computeU = true);
 
     /** \brief Computes Schur decomposition of a Hessenberg matrix H = Z T Z^T
      *  \param[in] matrixH Matrix in Hessenberg form H
@@ -243,26 +245,45 @@
 
 
 template<typename MatrixType>
-RealSchur<MatrixType>& RealSchur<MatrixType>::compute(const MatrixType& matrix, bool computeU)
+template<typename InputType>
+RealSchur<MatrixType>& RealSchur<MatrixType>::compute(const EigenBase<InputType>& matrix, bool computeU)
 {
+  const Scalar considerAsZero = (std::numeric_limits<Scalar>::min)();
+
   eigen_assert(matrix.cols() == matrix.rows());
   Index maxIters = m_maxIters;
   if (maxIters == -1)
     maxIters = m_maxIterationsPerRow * matrix.rows();
 
+  Scalar scale = matrix.derived().cwiseAbs().maxCoeff();
+  if(scale<considerAsZero)
+  {
+    m_matT.setZero(matrix.rows(),matrix.cols());
+    if(computeU)
+      m_matU.setIdentity(matrix.rows(),matrix.cols());
+    m_info = Success;
+    m_isInitialized = true;
+    m_matUisUptodate = computeU;
+    return *this;
+  }
+
   // Step 1. Reduce to Hessenberg form
-  m_hess.compute(matrix);
+  m_hess.compute(matrix.derived()/scale);
 
   // Step 2. Reduce to real Schur form  
   computeFromHessenberg(m_hess.matrixH(), m_hess.matrixQ(), computeU);
+
+  m_matT *= scale;
   
   return *this;
 }
 template<typename MatrixType>
 template<typename HessMatrixType, typename OrthMatrixType>
 RealSchur<MatrixType>& RealSchur<MatrixType>::computeFromHessenberg(const HessMatrixType& matrixH, const OrthMatrixType& matrixQ,  bool computeU)
-{  
-  m_matT = matrixH; 
+{
+  using std::abs;
+
+  m_matT = matrixH;
   if(computeU)
     m_matU = matrixQ;
   
@@ -282,7 +303,7 @@
   Scalar exshift(0);   // sum of exceptional shifts
   Scalar norm = computeNormOfT();
 
-  if(norm!=0)
+  if(norm!=Scalar(0))
   {
     while (iu >= 0)
     {
@@ -306,7 +327,7 @@
       else // No convergence yet
       {
         // The firstHouseholderVector vector has to be initialized to something to get rid of a silly GCC warning (-O1 -Wall -DNDEBUG )
-        Vector3s firstHouseholderVector(0,0,0), shiftInfo;
+        Vector3s firstHouseholderVector = Vector3s::Zero(), shiftInfo;
         computeShift(iu, iter, exshift, shiftInfo);
         iter = iter + 1;
         totalIter = totalIter + 1;
@@ -343,7 +364,7 @@
 
 /** \internal Look for single small sub-diagonal element and returns its index */
 template<typename MatrixType>
-inline typename MatrixType::Index RealSchur<MatrixType>::findSmallSubdiagEntry(Index iu)
+inline Index RealSchur<MatrixType>::findSmallSubdiagEntry(Index iu)
 {
   using std::abs;
   Index res = iu;
diff --git a/Eigen/src/Eigenvalues/RealSchur_MKL.h b/Eigen/src/Eigenvalues/RealSchur_LAPACKE.h
similarity index 64%
rename from Eigen/src/Eigenvalues/RealSchur_MKL.h
rename to Eigen/src/Eigenvalues/RealSchur_LAPACKE.h
index ad97364..2c22517 100644
--- a/Eigen/src/Eigenvalues/RealSchur_MKL.h
+++ b/Eigen/src/Eigenvalues/RealSchur_LAPACKE.h
@@ -25,43 +25,37 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to LAPACKe
  *    Real Schur needed to real unsymmetrical eigenvalues/eigenvectors.
  ********************************************************************************
 */
 
-#ifndef EIGEN_REAL_SCHUR_MKL_H
-#define EIGEN_REAL_SCHUR_MKL_H
-
-#include "Eigen/src/Core/util/MKL_support.h"
+#ifndef EIGEN_REAL_SCHUR_LAPACKE_H
+#define EIGEN_REAL_SCHUR_LAPACKE_H
 
 namespace Eigen { 
 
-/** \internal Specialization for the data types supported by MKL */
+/** \internal Specialization for the data types supported by LAPACKe */
 
-#define EIGEN_MKL_SCHUR_REAL(EIGTYPE, MKLTYPE, MKLPREFIX, MKLPREFIX_U, EIGCOLROW, MKLCOLROW) \
-template<> inline \
+#define EIGEN_LAPACKE_SCHUR_REAL(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX, LAPACKE_PREFIX_U, EIGCOLROW, LAPACKE_COLROW) \
+template<> template<typename InputType> inline \
 RealSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >& \
-RealSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW>& matrix, bool computeU) \
+RealSchur<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const EigenBase<InputType>& matrix, bool computeU) \
 { \
-  typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> MatrixType; \
-  typedef MatrixType::Scalar Scalar; \
-  typedef MatrixType::RealScalar RealScalar; \
-\
   eigen_assert(matrix.cols() == matrix.rows()); \
 \
-  lapack_int n = matrix.cols(), sdim, info; \
-  lapack_int lda = matrix.outerStride(); \
-  lapack_int matrix_order = MKLCOLROW; \
+  lapack_int n = internal::convert_index<lapack_int>(matrix.cols()), sdim, info; \
+  lapack_int matrix_order = LAPACKE_COLROW; \
   char jobvs, sort='N'; \
-  LAPACK_##MKLPREFIX_U##_SELECT2 select = 0; \
+  LAPACK_##LAPACKE_PREFIX_U##_SELECT2 select = 0; \
   jobvs = (computeU) ? 'V' : 'N'; \
   m_matU.resize(n, n); \
-  lapack_int ldvs  = m_matU.outerStride(); \
+  lapack_int ldvs  = internal::convert_index<lapack_int>(m_matU.outerStride()); \
   m_matT = matrix; \
+  lapack_int lda = internal::convert_index<lapack_int>(m_matT.outerStride()); \
   Matrix<EIGTYPE, Dynamic, Dynamic> wr, wi; \
   wr.resize(n, 1); wi.resize(n, 1); \
-  info = LAPACKE_##MKLPREFIX##gees( matrix_order, jobvs, sort, select, n, (MKLTYPE*)m_matT.data(), lda, &sdim, (MKLTYPE*)wr.data(), (MKLTYPE*)wi.data(), (MKLTYPE*)m_matU.data(), ldvs ); \
+  info = LAPACKE_##LAPACKE_PREFIX##gees( matrix_order, jobvs, sort, select, n, (LAPACKE_TYPE*)m_matT.data(), lda, &sdim, (LAPACKE_TYPE*)wr.data(), (LAPACKE_TYPE*)wi.data(), (LAPACKE_TYPE*)m_matU.data(), ldvs ); \
   if(info == 0) \
     m_info = Success; \
   else \
@@ -73,11 +67,11 @@
 \
 }
 
-EIGEN_MKL_SCHUR_REAL(double,   double, d, D, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_SCHUR_REAL(float,    float,  s, S, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_SCHUR_REAL(double,   double, d, D, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_SCHUR_REAL(float,    float,  s, S, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SCHUR_REAL(double,   double, d, D, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_REAL(float,    float,  s, S, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SCHUR_REAL(double,   double, d, D, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SCHUR_REAL(float,    float,  s, S, RowMajor, LAPACK_ROW_MAJOR)
 
 } // end namespace Eigen
 
-#endif // EIGEN_REAL_SCHUR_MKL_H
+#endif // EIGEN_REAL_SCHUR_LAPACKE_H
diff --git a/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h b/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
index c2e76c8..9ddd553 100644
--- a/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
+++ b/Eigen/src/Eigenvalues/SelfAdjointEigenSolver.h
@@ -20,6 +20,8 @@
 
 namespace internal {
 template<typename SolverType,int Size,bool IsComplex> struct direct_selfadjoint_eigenvalues;
+template<typename MatrixType, typename DiagType, typename SubDiagType>
+ComputationInfo computeFromTridiagonal_impl(DiagType& diag, SubDiagType& subdiag, const Index maxIterations, bool computeEigenvectors, MatrixType& eivec);
 }
 
 /** \eigenvalues_module \ingroup Eigenvalues_Module
@@ -79,7 +81,9 @@
     
     /** \brief Scalar type for matrices of type \p _MatrixType. */
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+    
+    typedef Matrix<Scalar,Size,Size,ColMajor,MaxColsAtCompileTime,MaxColsAtCompileTime> EigenvectorsType;
 
     /** \brief Real scalar type for \p _MatrixType.
       *
@@ -98,6 +102,7 @@
       */
     typedef typename internal::plain_col_type<MatrixType, RealScalar>::type RealVectorType;
     typedef Tridiagonalization<MatrixType> TridiagonalizationType;
+    typedef typename TridiagonalizationType::SubDiagonalType SubDiagonalType;
 
     /** \brief Default constructor for fixed-size matrices.
       *
@@ -109,6 +114,7 @@
       * Example: \include SelfAdjointEigenSolver_SelfAdjointEigenSolver.cpp
       * Output: \verbinclude SelfAdjointEigenSolver_SelfAdjointEigenSolver.out
       */
+    EIGEN_DEVICE_FUNC
     SelfAdjointEigenSolver()
         : m_eivec(),
           m_eivalues(),
@@ -128,7 +134,8 @@
       *
       * \sa compute() for an example
       */
-    SelfAdjointEigenSolver(Index size)
+    EIGEN_DEVICE_FUNC
+    explicit SelfAdjointEigenSolver(Index size)
         : m_eivec(size, size),
           m_eivalues(size),
           m_subdiag(size > 1 ? size - 1 : 1),
@@ -150,13 +157,15 @@
       *
       * \sa compute(const MatrixType&, int)
       */
-    SelfAdjointEigenSolver(const MatrixType& matrix, int options = ComputeEigenvectors)
+    template<typename InputType>
+    EIGEN_DEVICE_FUNC
+    explicit SelfAdjointEigenSolver(const EigenBase<InputType>& matrix, int options = ComputeEigenvectors)
       : m_eivec(matrix.rows(), matrix.cols()),
         m_eivalues(matrix.cols()),
         m_subdiag(matrix.rows() > 1 ? matrix.rows() - 1 : 1),
         m_isInitialized(false)
     {
-      compute(matrix, options);
+      compute(matrix.derived(), options);
     }
 
     /** \brief Computes eigendecomposition of given matrix.
@@ -189,24 +198,45 @@
       *
       * \sa SelfAdjointEigenSolver(const MatrixType&, int)
       */
-    SelfAdjointEigenSolver& compute(const MatrixType& matrix, int options = ComputeEigenvectors);
+    template<typename InputType>
+    EIGEN_DEVICE_FUNC
+    SelfAdjointEigenSolver& compute(const EigenBase<InputType>& matrix, int options = ComputeEigenvectors);
     
-    /** \brief Computes eigendecomposition of given matrix using a direct algorithm
+    /** \brief Computes eigendecomposition of given matrix using a closed-form algorithm
       *
       * This is a variant of compute(const MatrixType&, int options) which
       * directly solves the underlying polynomial equation.
       * 
-      * Currently only 3x3 matrices for which the sizes are known at compile time are supported (e.g., Matrix3d).
+      * Currently only 2x2 and 3x3 matrices for which the sizes are known at compile time are supported (e.g., Matrix3d).
       * 
-      * This method is usually significantly faster than the QR algorithm
+      * This method is usually significantly faster than the QR iterative algorithm
       * but it might also be less accurate. It is also worth noting that
       * for 3x3 matrices it involves trigonometric operations which are
       * not necessarily available for all scalar types.
+      * 
+      * For the 3x3 case, we observed the following worst case relative error regarding the eigenvalues:
+      *   - double: 1e-8
+      *   - float:  1e-3
       *
       * \sa compute(const MatrixType&, int options)
       */
+    EIGEN_DEVICE_FUNC
     SelfAdjointEigenSolver& computeDirect(const MatrixType& matrix, int options = ComputeEigenvectors);
 
+    /**
+      *\brief Computes the eigen decomposition from a tridiagonal symmetric matrix
+      *
+      * \param[in] diag The vector containing the diagonal of the matrix.
+      * \param[in] subdiag The subdiagonal of the matrix.
+      * \param[in] options Can be #ComputeEigenvectors (default) or #EigenvaluesOnly.
+      * \returns Reference to \c *this
+      *
+      * This function assumes that the matrix has been reduced to tridiagonal form.
+      *
+      * \sa compute(const MatrixType&, int) for more information
+      */
+    SelfAdjointEigenSolver& computeFromTridiagonal(const RealVectorType& diag, const SubDiagonalType& subdiag , int options=ComputeEigenvectors);
+
     /** \brief Returns the eigenvectors of given matrix.
       *
       * \returns  A const reference to the matrix whose columns are the eigenvectors.
@@ -225,7 +255,8 @@
       *
       * \sa eigenvalues()
       */
-    const MatrixType& eigenvectors() const
+    EIGEN_DEVICE_FUNC
+    const EigenvectorsType& eigenvectors() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
       eigen_assert(m_eigenvectorsOk && "The eigenvectors have not been computed together with the eigenvalues.");
@@ -247,6 +278,7 @@
       *
       * \sa eigenvectors(), MatrixBase::eigenvalues()
       */
+    EIGEN_DEVICE_FUNC
     const RealVectorType& eigenvalues() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -268,9 +300,9 @@
       * Example: \include SelfAdjointEigenSolver_operatorSqrt.cpp
       * Output: \verbinclude SelfAdjointEigenSolver_operatorSqrt.out
       *
-      * \sa operatorInverseSqrt(),
-      *     \ref MatrixFunctions_Module "MatrixFunctions Module"
+      * \sa operatorInverseSqrt(), <a href="unsupported/group__MatrixFunctions__Module.html">MatrixFunctions Module</a>
       */
+    EIGEN_DEVICE_FUNC
     MatrixType operatorSqrt() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -293,9 +325,9 @@
       * Example: \include SelfAdjointEigenSolver_operatorInverseSqrt.cpp
       * Output: \verbinclude SelfAdjointEigenSolver_operatorInverseSqrt.out
       *
-      * \sa operatorSqrt(), MatrixBase::inverse(),
-      *     \ref MatrixFunctions_Module "MatrixFunctions Module"
+      * \sa operatorSqrt(), MatrixBase::inverse(), <a href="unsupported/group__MatrixFunctions__Module.html">MatrixFunctions Module</a>
       */
+    EIGEN_DEVICE_FUNC
     MatrixType operatorInverseSqrt() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -307,6 +339,7 @@
       *
       * \returns \c Success if computation was succesful, \c NoConvergence otherwise.
       */
+    EIGEN_DEVICE_FUNC
     ComputationInfo info() const
     {
       eigen_assert(m_isInitialized && "SelfAdjointEigenSolver is not initialized.");
@@ -320,43 +353,13 @@
       */
     static const int m_maxIterations = 30;
 
-    #ifdef EIGEN2_SUPPORT
-    SelfAdjointEigenSolver(const MatrixType& matrix, bool computeEigenvectors)
-      : m_eivec(matrix.rows(), matrix.cols()),
-        m_eivalues(matrix.cols()),
-        m_subdiag(matrix.rows() > 1 ? matrix.rows() - 1 : 1),
-        m_isInitialized(false)
-    {
-      compute(matrix, computeEigenvectors);
-    }
-    
-    SelfAdjointEigenSolver(const MatrixType& matA, const MatrixType& matB, bool computeEigenvectors = true)
-        : m_eivec(matA.cols(), matA.cols()),
-          m_eivalues(matA.cols()),
-          m_subdiag(matA.cols() > 1 ? matA.cols() - 1 : 1),
-          m_isInitialized(false)
-    {
-      static_cast<GeneralizedSelfAdjointEigenSolver<MatrixType>*>(this)->compute(matA, matB, computeEigenvectors ? ComputeEigenvectors : EigenvaluesOnly);
-    }
-    
-    void compute(const MatrixType& matrix, bool computeEigenvectors)
-    {
-      compute(matrix, computeEigenvectors ? ComputeEigenvectors : EigenvaluesOnly);
-    }
-
-    void compute(const MatrixType& matA, const MatrixType& matB, bool computeEigenvectors = true)
-    {
-      compute(matA, matB, computeEigenvectors ? ComputeEigenvectors : EigenvaluesOnly);
-    }
-    #endif // EIGEN2_SUPPORT
-
   protected:
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
     
-    MatrixType m_eivec;
+    EigenvectorsType m_eivec;
     RealVectorType m_eivalues;
     typename TridiagonalizationType::SubDiagonalType m_subdiag;
     ComputationInfo m_info;
@@ -364,6 +367,7 @@
     bool m_eigenvectorsOk;
 };
 
+namespace internal {
 /** \internal
   *
   * \eigenvalues_module \ingroup Eigenvalues_Module
@@ -371,8 +375,12 @@
   * Performs a QR step on a tridiagonal symmetric matrix represented as a
   * pair of two vectors \a diag and \a subdiag.
   *
-  * \param matA the input selfadjoint matrix
-  * \param hCoeffs returned Householder coefficients
+  * \param diag the diagonal part of the input selfadjoint tridiagonal matrix
+  * \param subdiag the sub-diagonal part of the input selfadjoint tridiagonal matrix
+  * \param start starting index of the submatrix to work on
+  * \param end last+1 index of the submatrix to work on
+  * \param matrixQ pointer to the column-major matrix holding the eigenvectors, can be 0
+  * \param n size of the input matrix
   *
   * For compilation efficiency reasons, this procedure does not use eigen expression
   * for its arguments.
@@ -380,17 +388,21 @@
   * Implemented from Golub's "Matrix Computations", algorithm 8.3.2:
   * "implicit symmetric QR step with Wilkinson shift"
   */
-namespace internal {
 template<int StorageOrder,typename RealScalar, typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC
 static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index start, Index end, Scalar* matrixQ, Index n);
 }
 
 template<typename MatrixType>
+template<typename InputType>
+EIGEN_DEVICE_FUNC
 SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
-::compute(const MatrixType& matrix, int options)
+::compute(const EigenBase<InputType>& a_matrix, int options)
 {
   check_template_parameters();
   
+  const InputType &matrix(a_matrix.derived());
+  
   using std::abs;
   eigen_assert(matrix.cols() == matrix.rows());
   eigen_assert((options&~(EigVecMask|GenEigMask))==0
@@ -402,7 +414,8 @@
 
   if(n==1)
   {
-    m_eivalues.coeffRef(0,0) = numext::real(matrix.coeff(0,0));
+    m_eivec = matrix;
+    m_eivalues.coeffRef(0,0) = numext::real(m_eivec.coeff(0,0));
     if(computeEigenvectors)
       m_eivec.setOnes(n,n);
     m_info = Success;
@@ -413,7 +426,7 @@
 
   // declare some aliases
   RealVectorType& diag = m_eivalues;
-  MatrixType& mat = m_eivec;
+  EigenvectorsType& mat = m_eivec;
 
   // map the matrix coefficients to [-1:1] to avoid over- and underflow.
   mat = matrix.template triangularView<Lower>();
@@ -422,58 +435,8 @@
   mat.template triangularView<Lower>() /= scale;
   m_subdiag.resize(n-1);
   internal::tridiagonalization_inplace(mat, diag, m_subdiag, computeEigenvectors);
-  
-  Index end = n-1;
-  Index start = 0;
-  Index iter = 0; // total number of iterations
 
-  while (end>0)
-  {
-    for (Index i = start; i<end; ++i)
-      if (internal::isMuchSmallerThan(abs(m_subdiag[i]),(abs(diag[i])+abs(diag[i+1]))))
-        m_subdiag[i] = 0;
-
-    // find the largest unreduced block
-    while (end>0 && m_subdiag[end-1]==0)
-    {
-      end--;
-    }
-    if (end<=0)
-      break;
-
-    // if we spent too many iterations, we give up
-    iter++;
-    if(iter > m_maxIterations * n) break;
-
-    start = end - 1;
-    while (start>0 && m_subdiag[start-1]!=0)
-      start--;
-
-    internal::tridiagonal_qr_step<MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor>(diag.data(), m_subdiag.data(), start, end, computeEigenvectors ? m_eivec.data() : (Scalar*)0, n);
-  }
-
-  if (iter <= m_maxIterations * n)
-    m_info = Success;
-  else
-    m_info = NoConvergence;
-
-  // Sort eigenvalues and corresponding vectors.
-  // TODO make the sort optional ?
-  // TODO use a better sort algorithm !!
-  if (m_info == Success)
-  {
-    for (Index i = 0; i < n-1; ++i)
-    {
-      Index k;
-      m_eivalues.segment(i,n-i).minCoeff(&k);
-      if (k > 0)
-      {
-        std::swap(m_eivalues[i], m_eivalues[k+i]);
-        if(computeEigenvectors)
-          m_eivec.col(i).swap(m_eivec.col(k+i));
-      }
-    }
-  }
+  m_info = internal::computeFromTridiagonal_impl(diag, m_subdiag, m_maxIterations, computeEigenvectors, m_eivec);
   
   // scale back the eigen values
   m_eivalues *= scale;
@@ -483,11 +446,107 @@
   return *this;
 }
 
+template<typename MatrixType>
+SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
+::computeFromTridiagonal(const RealVectorType& diag, const SubDiagonalType& subdiag , int options)
+{
+  //TODO : Add an option to scale the values beforehand
+  bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors;
+
+  m_eivalues = diag;
+  m_subdiag = subdiag;
+  if (computeEigenvectors)
+  {
+    m_eivec.setIdentity(diag.size(), diag.size());
+  }
+  m_info = internal::computeFromTridiagonal_impl(m_eivalues, m_subdiag, m_maxIterations, computeEigenvectors, m_eivec);
+
+  m_isInitialized = true;
+  m_eigenvectorsOk = computeEigenvectors;
+  return *this;
+}
 
 namespace internal {
+/**
+  * \internal
+  * \brief Compute the eigendecomposition from a tridiagonal matrix
+  *
+  * \param[in,out] diag : On input, the diagonal of the matrix, on output the eigenvalues
+  * \param[in,out] subdiag : The subdiagonal part of the matrix (entries are modified during the decomposition)
+  * \param[in] maxIterations : the maximum number of iterations
+  * \param[in] computeEigenvectors : whether the eigenvectors have to be computed or not
+  * \param[out] eivec : The matrix to store the eigenvectors if computeEigenvectors==true. Must be allocated on input.
+  * \returns \c Success or \c NoConvergence
+  */
+template<typename MatrixType, typename DiagType, typename SubDiagType>
+ComputationInfo computeFromTridiagonal_impl(DiagType& diag, SubDiagType& subdiag, const Index maxIterations, bool computeEigenvectors, MatrixType& eivec)
+{
+  using std::abs;
+
+  ComputationInfo info;
+  typedef typename MatrixType::Scalar Scalar;
+
+  Index n = diag.size();
+  Index end = n-1;
+  Index start = 0;
+  Index iter = 0; // total number of iterations
+  
+  typedef typename DiagType::RealScalar RealScalar;
+  const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
+  const RealScalar precision = RealScalar(2)*NumTraits<RealScalar>::epsilon();
+  
+  while (end>0)
+  {
+    for (Index i = start; i<end; ++i)
+      if (internal::isMuchSmallerThan(abs(subdiag[i]),(abs(diag[i])+abs(diag[i+1])),precision) || abs(subdiag[i]) <= considerAsZero)
+        subdiag[i] = 0;
+
+    // find the largest unreduced block
+    while (end>0 && subdiag[end-1]==RealScalar(0))
+    {
+      end--;
+    }
+    if (end<=0)
+      break;
+
+    // if we spent too many iterations, we give up
+    iter++;
+    if(iter > maxIterations * n) break;
+
+    start = end - 1;
+    while (start>0 && subdiag[start-1]!=0)
+      start--;
+
+    internal::tridiagonal_qr_step<MatrixType::Flags&RowMajorBit ? RowMajor : ColMajor>(diag.data(), subdiag.data(), start, end, computeEigenvectors ? eivec.data() : (Scalar*)0, n);
+  }
+  if (iter <= maxIterations * n)
+    info = Success;
+  else
+    info = NoConvergence;
+
+  // Sort eigenvalues and corresponding vectors.
+  // TODO make the sort optional ?
+  // TODO use a better sort algorithm !!
+  if (info == Success)
+  {
+    for (Index i = 0; i < n-1; ++i)
+    {
+      Index k;
+      diag.segment(i,n-i).minCoeff(&k);
+      if (k > 0)
+      {
+        std::swap(diag[i], diag[k+i]);
+        if(computeEigenvectors)
+          eivec.col(i).swap(eivec.col(k+i));
+      }
+    }
+  }
+  return info;
+}
   
 template<typename SolverType,int Size,bool IsComplex> struct direct_selfadjoint_eigenvalues
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(SolverType& eig, const typename SolverType::MatrixType& A, int options)
   { eig.compute(A,options); }
 };
@@ -497,20 +556,22 @@
   typedef typename SolverType::MatrixType MatrixType;
   typedef typename SolverType::RealVectorType VectorType;
   typedef typename SolverType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
+  typedef typename SolverType::EigenvectorsType EigenvectorsType;
   
+
   /** \internal
    * Computes the roots of the characteristic polynomial of \a m.
    * For numerical stability m.trace() should be near zero and to avoid over- or underflow m should be normalized.
    */
+  EIGEN_DEVICE_FUNC
   static inline void computeRoots(const MatrixType& m, VectorType& roots)
   {
-    using std::sqrt;
-    using std::atan2;
-    using std::cos;
-    using std::sin;
-    const Scalar s_inv3 = Scalar(1.0)/Scalar(3.0);
-    const Scalar s_sqrt3 = sqrt(Scalar(3.0));
+    EIGEN_USING_STD_MATH(sqrt)
+    EIGEN_USING_STD_MATH(atan2)
+    EIGEN_USING_STD_MATH(cos)
+    EIGEN_USING_STD_MATH(sin)
+    const Scalar s_inv3 = Scalar(1)/Scalar(3);
+    const Scalar s_sqrt3 = sqrt(Scalar(3));
 
     // The characteristic equation is x^3 - c2*x^2 + c1*x - c0 = 0.  The
     // eigenvalues are the roots to this equation, all guaranteed to be
@@ -523,14 +584,12 @@
     // and in solving the equation for the roots in closed form.
     Scalar c2_over_3 = c2*s_inv3;
     Scalar a_over_3 = (c2*c2_over_3 - c1)*s_inv3;
-    if(a_over_3<Scalar(0))
-      a_over_3 = Scalar(0);
+    a_over_3 = numext::maxi(a_over_3, Scalar(0));
 
     Scalar half_b = Scalar(0.5)*(c0 + c2_over_3*(Scalar(2)*c2_over_3*c2_over_3 - c1));
 
     Scalar q = a_over_3*a_over_3*a_over_3 - half_b*half_b;
-    if(q<Scalar(0))
-      q = Scalar(0);
+    q = numext::maxi(q, Scalar(0));
 
     // Compute the eigenvalues by solving for the roots of the polynomial.
     Scalar rho = sqrt(a_over_3);
@@ -543,6 +602,7 @@
     roots(2) = c2_over_3 + Scalar(2)*rho*cos_theta;
   }
 
+  EIGEN_DEVICE_FUNC
   static inline bool extract_kernel(MatrixType& mat, Ref<VectorType> res, Ref<VectorType> representative)
   {
     using std::abs;
@@ -562,6 +622,7 @@
     return true;
   }
 
+  EIGEN_DEVICE_FUNC
   static inline void run(SolverType& solver, const MatrixType& mat, int options)
   {
     eigen_assert(mat.cols() == 3 && mat.cols() == mat.rows());
@@ -570,7 +631,7 @@
             && "invalid option parameter");
     bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors;
     
-    MatrixType& eivecs = solver.m_eivec;
+    EigenvectorsType& eivecs = solver.m_eivec;
     VectorType& eivals = solver.m_eivalues;
   
     // Shift the matrix to the mean eigenvalue and map the matrix coefficients to [-1:1] to avoid over- and underflow.
@@ -603,7 +664,7 @@
         Index k(0), l(2);
         if(d0 > d1)
         {
-          std::swap(k,l);
+          numext::swap(k,l);
           d0 = d1;
         }
 
@@ -647,12 +708,15 @@
 };
 
 // 2x2 direct eigenvalues decomposition, code from Hauke Heibel
-template<typename SolverType> struct direct_selfadjoint_eigenvalues<SolverType,2,false>
+template<typename SolverType> 
+struct direct_selfadjoint_eigenvalues<SolverType,2,false>
 {
   typedef typename SolverType::MatrixType MatrixType;
   typedef typename SolverType::RealVectorType VectorType;
   typedef typename SolverType::Scalar Scalar;
+  typedef typename SolverType::EigenvectorsType EigenvectorsType;
   
+  EIGEN_DEVICE_FUNC
   static inline void computeRoots(const MatrixType& m, VectorType& roots)
   {
     using std::sqrt;
@@ -662,28 +726,33 @@
     roots(1) = t1 + t0;
   }
   
+  EIGEN_DEVICE_FUNC
   static inline void run(SolverType& solver, const MatrixType& mat, int options)
   {
-    using std::sqrt;
-    using std::abs;
-
+    EIGEN_USING_STD_MATH(sqrt);
+    EIGEN_USING_STD_MATH(abs);
+    
     eigen_assert(mat.cols() == 2 && mat.cols() == mat.rows());
     eigen_assert((options&~(EigVecMask|GenEigMask))==0
             && (options&EigVecMask)!=EigVecMask
             && "invalid option parameter");
     bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors;
     
-    MatrixType& eivecs = solver.m_eivec;
+    EigenvectorsType& eivecs = solver.m_eivec;
     VectorType& eivals = solver.m_eivalues;
   
-    // map the matrix coefficients to [-1:1] to avoid over- and underflow.
-    Scalar scale = mat.cwiseAbs().maxCoeff();
-    scale = (std::max)(scale,Scalar(1));
-    MatrixType scaledMat = mat / scale;
-    
+    // Shift the matrix to the mean eigenvalue and map the matrix coefficients to [-1:1] to avoid over- and underflow.
+    Scalar shift = mat.trace() / Scalar(2);
+    MatrixType scaledMat = mat;
+    scaledMat.coeffRef(0,1) = mat.coeff(1,0);
+    scaledMat.diagonal().array() -= shift;
+    Scalar scale = scaledMat.cwiseAbs().maxCoeff();
+    if(scale > Scalar(0))
+      scaledMat /= scale;
+
     // Compute the eigenvalues
     computeRoots(scaledMat,eivals);
-    
+
     // compute the eigen vectors
     if(computeEigenvectors)
     {
@@ -711,10 +780,11 @@
         eivecs.col(0) << eivecs.col(1).unitOrthogonal();
       }
     }
-    
+
     // Rescale back to the original size.
     eivals *= scale;
-    
+    eivals.array() += shift;
+
     solver.m_info = Success;
     solver.m_isInitialized = true;
     solver.m_eigenvectorsOk = computeEigenvectors;
@@ -724,6 +794,7 @@
 }
 
 template<typename MatrixType>
+EIGEN_DEVICE_FUNC
 SelfAdjointEigenSolver<MatrixType>& SelfAdjointEigenSolver<MatrixType>
 ::computeDirect(const MatrixType& matrix, int options)
 {
@@ -733,6 +804,7 @@
 
 namespace internal {
 template<int StorageOrder,typename RealScalar, typename Scalar, typename Index>
+EIGEN_DEVICE_FUNC
 static void tridiagonal_qr_step(RealScalar* diag, RealScalar* subdiag, Index start, Index end, Scalar* matrixQ, Index n)
 {
   using std::abs;
@@ -744,14 +816,14 @@
 //   RealScalar mu = diag[end] - e2 / (td + (td>0 ? 1 : -1) * sqrt(td*td + e2));
   // This explain the following, somewhat more complicated, version:
   RealScalar mu = diag[end];
-  if(td==0)
+  if(td==RealScalar(0))
     mu -= abs(e);
   else
   {
     RealScalar e2 = numext::abs2(subdiag[end-1]);
     RealScalar h = numext::hypot(td,e);
-    if(e2==0)  mu -= (e / (td + (td>0 ? 1 : -1))) * (e / h);
-    else       mu -= e2 / (td + (td>0 ? h : -h));
+    if(e2==RealScalar(0)) mu -= (e / (td + (td>RealScalar(0) ? RealScalar(1) : RealScalar(-1)))) * (e / h);
+    else                  mu -= e2 / (td + (td>RealScalar(0) ? h : -h));
   }
   
   RealScalar x = diag[start] - mu;
diff --git a/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_MKL.h b/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h
similarity index 65%
rename from Eigen/src/Eigenvalues/SelfAdjointEigenSolver_MKL.h
rename to Eigen/src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h
index 17c0dad..b0c947d 100644
--- a/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_MKL.h
+++ b/Eigen/src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h
@@ -25,38 +25,36 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to LAPACKe
  *    Self-adjoint eigenvalues/eigenvectors.
  ********************************************************************************
 */
 
-#ifndef EIGEN_SAEIGENSOLVER_MKL_H
-#define EIGEN_SAEIGENSOLVER_MKL_H
-
-#include "Eigen/src/Core/util/MKL_support.h"
+#ifndef EIGEN_SAEIGENSOLVER_LAPACKE_H
+#define EIGEN_SAEIGENSOLVER_LAPACKE_H
 
 namespace Eigen { 
 
-/** \internal Specialization for the data types supported by MKL */
+/** \internal Specialization for the data types supported by LAPACKe */
 
-#define EIGEN_MKL_EIG_SELFADJ(EIGTYPE, MKLTYPE, MKLRTYPE, MKLNAME, EIGCOLROW, MKLCOLROW ) \
-template<> inline \
+#define EIGEN_LAPACKE_EIG_SELFADJ_2(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME, EIGCOLROW ) \
+template<> template<typename InputType> inline \
 SelfAdjointEigenSolver<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >& \
-SelfAdjointEigenSolver<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW>& matrix, int options) \
+SelfAdjointEigenSolver<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW> >::compute(const EigenBase<InputType>& matrix, int options) \
 { \
   eigen_assert(matrix.cols() == matrix.rows()); \
   eigen_assert((options&~(EigVecMask|GenEigMask))==0 \
           && (options&EigVecMask)!=EigVecMask \
           && "invalid option parameter"); \
   bool computeEigenvectors = (options&ComputeEigenvectors)==ComputeEigenvectors; \
-  lapack_int n = matrix.cols(), lda, matrix_order, info; \
+  lapack_int n = internal::convert_index<lapack_int>(matrix.cols()), lda, info; \
   m_eivalues.resize(n,1); \
   m_subdiag.resize(n-1); \
   m_eivec = matrix; \
 \
   if(n==1) \
   { \
-    m_eivalues.coeffRef(0,0) = numext::real(matrix.coeff(0,0)); \
+    m_eivalues.coeffRef(0,0) = numext::real(m_eivec.coeff(0,0)); \
     if(computeEigenvectors) m_eivec.setOnes(n,n); \
     m_info = Success; \
     m_isInitialized = true; \
@@ -64,28 +62,25 @@
     return *this; \
   } \
 \
-  lda = matrix.outerStride(); \
-  matrix_order=MKLCOLROW; \
+  lda = internal::convert_index<lapack_int>(m_eivec.outerStride()); \
   char jobz, uplo='L'/*, range='A'*/; \
   jobz = computeEigenvectors ? 'V' : 'N'; \
 \
-  info = LAPACKE_##MKLNAME( matrix_order, jobz, uplo, n, (MKLTYPE*)m_eivec.data(), lda, (MKLRTYPE*)m_eivalues.data() ); \
+  info = LAPACKE_##LAPACKE_NAME( LAPACK_COL_MAJOR, jobz, uplo, n, (LAPACKE_TYPE*)m_eivec.data(), lda, (LAPACKE_RTYPE*)m_eivalues.data() ); \
   m_info = (info==0) ? Success : NoConvergence; \
   m_isInitialized = true; \
   m_eigenvectorsOk = computeEigenvectors; \
   return *this; \
 }
 
+#define EIGEN_LAPACKE_EIG_SELFADJ(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME )              \
+        EIGEN_LAPACKE_EIG_SELFADJ_2(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME, ColMajor )  \
+        EIGEN_LAPACKE_EIG_SELFADJ_2(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_NAME, RowMajor ) 
 
-EIGEN_MKL_EIG_SELFADJ(double,   double,        double, dsyev, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_EIG_SELFADJ(float,    float,         float,  ssyev, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_EIG_SELFADJ(dcomplex, MKL_Complex16, double, zheev, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_EIG_SELFADJ(scomplex, MKL_Complex8,  float,  cheev, ColMajor, LAPACK_COL_MAJOR)
-
-EIGEN_MKL_EIG_SELFADJ(double,   double,        double, dsyev, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_EIG_SELFADJ(float,    float,         float,  ssyev, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_EIG_SELFADJ(dcomplex, MKL_Complex16, double, zheev, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_EIG_SELFADJ(scomplex, MKL_Complex8,  float,  cheev, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_EIG_SELFADJ(double,   double,                double, dsyev)
+EIGEN_LAPACKE_EIG_SELFADJ(float,    float,                 float,  ssyev)
+EIGEN_LAPACKE_EIG_SELFADJ(dcomplex, lapack_complex_double, double, zheev)
+EIGEN_LAPACKE_EIG_SELFADJ(scomplex, lapack_complex_float,  float,  cheev)
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/Eigenvalues/Tridiagonalization.h b/Eigen/src/Eigenvalues/Tridiagonalization.h
index 192278d..1d102c1 100644
--- a/Eigen/src/Eigenvalues/Tridiagonalization.h
+++ b/Eigen/src/Eigenvalues/Tridiagonalization.h
@@ -18,8 +18,10 @@
 template<typename MatrixType> struct TridiagonalizationMatrixTReturnType;
 template<typename MatrixType>
 struct traits<TridiagonalizationMatrixTReturnType<MatrixType> >
+  : public traits<typename MatrixType::PlainObject>
 {
-  typedef typename MatrixType::PlainObject ReturnType;
+  typedef typename MatrixType::PlainObject ReturnType; // FIXME shall it be a BandMatrix?
+  enum { Flags = 0 };
 };
 
 template<typename MatrixType, typename CoeffVectorType>
@@ -67,7 +69,7 @@
 
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
 
     enum {
       Size = MatrixType::RowsAtCompileTime,
@@ -89,10 +91,8 @@
             >::type DiagonalReturnType;
 
     typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
-              typename internal::add_const_on_value_type<typename Diagonal<
-                Block<const MatrixType,SizeMinusOne,SizeMinusOne> >::RealReturnType>::type,
-              const Diagonal<
-                Block<const MatrixType,SizeMinusOne,SizeMinusOne> >
+              typename internal::add_const_on_value_type<typename Diagonal<const MatrixType, -1>::RealReturnType>::type,
+              const Diagonal<const MatrixType, -1>
             >::type SubDiagonalReturnType;
 
     /** \brief Return type of matrixQ() */
@@ -110,7 +110,7 @@
       *
       * \sa compute() for an example.
       */
-    Tridiagonalization(Index size = Size==Dynamic ? 2 : Size)
+    explicit Tridiagonalization(Index size = Size==Dynamic ? 2 : Size)
       : m_matrix(size,size),
         m_hCoeffs(size > 1 ? size-1 : 1),
         m_isInitialized(false)
@@ -126,8 +126,9 @@
       * Example: \include Tridiagonalization_Tridiagonalization_MatrixType.cpp
       * Output: \verbinclude Tridiagonalization_Tridiagonalization_MatrixType.out
       */
-    Tridiagonalization(const MatrixType& matrix)
-      : m_matrix(matrix),
+    template<typename InputType>
+    explicit Tridiagonalization(const EigenBase<InputType>& matrix)
+      : m_matrix(matrix.derived()),
         m_hCoeffs(matrix.cols() > 1 ? matrix.cols()-1 : 1),
         m_isInitialized(false)
     {
@@ -152,9 +153,10 @@
       * Example: \include Tridiagonalization_compute.cpp
       * Output: \verbinclude Tridiagonalization_compute.out
       */
-    Tridiagonalization& compute(const MatrixType& matrix)
+    template<typename InputType>
+    Tridiagonalization& compute(const EigenBase<InputType>& matrix)
     {
-      m_matrix = matrix;
+      m_matrix = matrix.derived();
       m_hCoeffs.resize(matrix.rows()-1, 1);
       internal::tridiagonalization_inplace(m_matrix, m_hCoeffs);
       m_isInitialized = true;
@@ -305,7 +307,7 @@
 Tridiagonalization<MatrixType>::diagonal() const
 {
   eigen_assert(m_isInitialized && "Tridiagonalization is not initialized.");
-  return m_matrix.diagonal();
+  return m_matrix.diagonal().real();
 }
 
 template<typename MatrixType>
@@ -313,8 +315,7 @@
 Tridiagonalization<MatrixType>::subDiagonal() const
 {
   eigen_assert(m_isInitialized && "Tridiagonalization is not initialized.");
-  Index n = m_matrix.rows();
-  return Block<const MatrixType,SizeMinusOne,SizeMinusOne>(m_matrix, 1, 0, n-1,n-1).diagonal();
+  return m_matrix.template diagonal<-1>().real();
 }
 
 namespace internal {
@@ -346,7 +347,6 @@
 void tridiagonalization_inplace(MatrixType& matA, CoeffVectorType& hCoeffs)
 {
   using numext::conj;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename MatrixType::RealScalar RealScalar;
   Index n = matA.rows();
@@ -367,10 +367,10 @@
     hCoeffs.tail(n-i-1).noalias() = (matA.bottomRightCorner(remainingSize,remainingSize).template selfadjointView<Lower>()
                                   * (conj(h) * matA.col(i).tail(remainingSize)));
 
-    hCoeffs.tail(n-i-1) += (conj(h)*Scalar(-0.5)*(hCoeffs.tail(remainingSize).dot(matA.col(i).tail(remainingSize)))) * matA.col(i).tail(n-i-1);
+    hCoeffs.tail(n-i-1) += (conj(h)*RealScalar(-0.5)*(hCoeffs.tail(remainingSize).dot(matA.col(i).tail(remainingSize)))) * matA.col(i).tail(n-i-1);
 
     matA.bottomRightCorner(remainingSize, remainingSize).template selfadjointView<Lower>()
-      .rankUpdate(matA.col(i).tail(remainingSize), hCoeffs.tail(remainingSize), -1);
+      .rankUpdate(matA.col(i).tail(remainingSize), hCoeffs.tail(remainingSize), Scalar(-1));
 
     matA.col(i).coeffRef(i+1) = beta;
     hCoeffs.coeffRef(i) = h;
@@ -438,7 +438,6 @@
 {
   typedef typename Tridiagonalization<MatrixType>::CoeffVectorType CoeffVectorType;
   typedef typename Tridiagonalization<MatrixType>::HouseholderSequenceType HouseholderSequenceType;
-  typedef typename MatrixType::Index Index;
   template<typename DiagonalType, typename SubDiagonalType>
   static void run(MatrixType& mat, DiagonalType& diag, SubDiagonalType& subdiag, bool extractQ)
   {
@@ -467,9 +466,10 @@
   static void run(MatrixType& mat, DiagonalType& diag, SubDiagonalType& subdiag, bool extractQ)
   {
     using std::sqrt;
+    const RealScalar tol = (std::numeric_limits<RealScalar>::min)();
     diag[0] = mat(0,0);
     RealScalar v1norm2 = numext::abs2(mat(2,0));
-    if(v1norm2 == RealScalar(0))
+    if(v1norm2 <= tol)
     {
       diag[1] = mat(1,1);
       diag[2] = mat(2,2);
@@ -526,7 +526,6 @@
 template<typename MatrixType> struct TridiagonalizationMatrixTReturnType
 : public ReturnByValue<TridiagonalizationMatrixTReturnType<MatrixType> >
 {
-    typedef typename MatrixType::Index Index;
   public:
     /** \brief Constructor.
       *
diff --git a/Eigen/src/Geometry/AlignedBox.h b/Eigen/src/Geometry/AlignedBox.h
index b226336..066eae4 100644
--- a/Eigen/src/Geometry/AlignedBox.h
+++ b/Eigen/src/Geometry/AlignedBox.h
@@ -34,10 +34,11 @@
   enum { AmbientDimAtCompileTime = _AmbientDim };
   typedef _Scalar                                   Scalar;
   typedef NumTraits<Scalar>                         ScalarTraits;
-  typedef DenseIndex                                Index;
+  typedef Eigen::Index                              Index; ///< \deprecated since Eigen 3.3
   typedef typename ScalarTraits::Real               RealScalar;
-  typedef typename ScalarTraits::NonInteger      NonInteger;
+  typedef typename ScalarTraits::NonInteger         NonInteger;
   typedef Matrix<Scalar,AmbientDimAtCompileTime,1>  VectorType;
+  typedef CwiseBinaryOp<internal::scalar_sum_op<Scalar>, const VectorType, const VectorType> VectorTypeSum;
 
   /** Define constants to name the corners of a 1D, 2D or 3D axis aligned bounding box */
   enum CornerType
@@ -61,77 +62,76 @@
 
 
   /** Default constructor initializing a null box. */
-  inline AlignedBox()
+  EIGEN_DEVICE_FUNC inline AlignedBox()
   { if (AmbientDimAtCompileTime!=Dynamic) setEmpty(); }
 
   /** Constructs a null box with \a _dim the dimension of the ambient space. */
-  inline explicit AlignedBox(Index _dim) : m_min(_dim), m_max(_dim)
+  EIGEN_DEVICE_FUNC inline explicit AlignedBox(Index _dim) : m_min(_dim), m_max(_dim)
   { setEmpty(); }
 
   /** Constructs a box with extremities \a _min and \a _max.
    * \warning If either component of \a _min is larger than the same component of \a _max, the constructed box is empty. */
   template<typename OtherVectorType1, typename OtherVectorType2>
-  inline AlignedBox(const OtherVectorType1& _min, const OtherVectorType2& _max) : m_min(_min), m_max(_max) {}
+  EIGEN_DEVICE_FUNC inline AlignedBox(const OtherVectorType1& _min, const OtherVectorType2& _max) : m_min(_min), m_max(_max) {}
 
   /** Constructs a box containing a single point \a p. */
   template<typename Derived>
-  inline explicit AlignedBox(const MatrixBase<Derived>& p) : m_min(p), m_max(m_min)
+  EIGEN_DEVICE_FUNC inline explicit AlignedBox(const MatrixBase<Derived>& p) : m_min(p), m_max(m_min)
   { }
 
-  ~AlignedBox() {}
+  EIGEN_DEVICE_FUNC ~AlignedBox() {}
 
   /** \returns the dimension in which the box holds */
-  inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_min.size() : Index(AmbientDimAtCompileTime); }
+  EIGEN_DEVICE_FUNC inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_min.size() : Index(AmbientDimAtCompileTime); }
 
   /** \deprecated use isEmpty() */
-  inline bool isNull() const { return isEmpty(); }
+  EIGEN_DEVICE_FUNC inline bool isNull() const { return isEmpty(); }
 
   /** \deprecated use setEmpty() */
-  inline void setNull() { setEmpty(); }
+  EIGEN_DEVICE_FUNC inline void setNull() { setEmpty(); }
 
   /** \returns true if the box is empty.
    * \sa setEmpty */
-  inline bool isEmpty() const { return (m_min.array() > m_max.array()).any(); }
+  EIGEN_DEVICE_FUNC inline bool isEmpty() const { return (m_min.array() > m_max.array()).any(); }
 
   /** Makes \c *this an empty box.
    * \sa isEmpty */
-  inline void setEmpty()
+  EIGEN_DEVICE_FUNC inline void setEmpty()
   {
     m_min.setConstant( ScalarTraits::highest() );
     m_max.setConstant( ScalarTraits::lowest() );
   }
 
   /** \returns the minimal corner */
-  inline const VectorType& (min)() const { return m_min; }
+  EIGEN_DEVICE_FUNC inline const VectorType& (min)() const { return m_min; }
   /** \returns a non const reference to the minimal corner */
-  inline VectorType& (min)() { return m_min; }
+  EIGEN_DEVICE_FUNC inline VectorType& (min)() { return m_min; }
   /** \returns the maximal corner */
-  inline const VectorType& (max)() const { return m_max; }
+  EIGEN_DEVICE_FUNC inline const VectorType& (max)() const { return m_max; }
   /** \returns a non const reference to the maximal corner */
-  inline VectorType& (max)() { return m_max; }
+  EIGEN_DEVICE_FUNC inline VectorType& (max)() { return m_max; }
 
   /** \returns the center of the box */
-  inline const CwiseUnaryOp<internal::scalar_quotient1_op<Scalar>,
-                            const CwiseBinaryOp<internal::scalar_sum_op<Scalar>, const VectorType, const VectorType> >
+  EIGEN_DEVICE_FUNC inline const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(VectorTypeSum, RealScalar, quotient)
   center() const
-  { return (m_min+m_max)/2; }
+  { return (m_min+m_max)/RealScalar(2); }
 
   /** \returns the lengths of the sides of the bounding box.
     * Note that this function does not get the same
     * result for integral or floating scalar types: see
     */
-  inline const CwiseBinaryOp< internal::scalar_difference_op<Scalar>, const VectorType, const VectorType> sizes() const
+  EIGEN_DEVICE_FUNC inline const CwiseBinaryOp< internal::scalar_difference_op<Scalar,Scalar>, const VectorType, const VectorType> sizes() const
   { return m_max - m_min; }
 
   /** \returns the volume of the bounding box */
-  inline Scalar volume() const
+  EIGEN_DEVICE_FUNC inline Scalar volume() const
   { return sizes().prod(); }
 
   /** \returns an expression for the bounding box diagonal vector
     * if the length of the diagonal is needed: diagonal().norm()
     * will provide it.
     */
-  inline CwiseBinaryOp< internal::scalar_difference_op<Scalar>, const VectorType, const VectorType> diagonal() const
+  EIGEN_DEVICE_FUNC inline CwiseBinaryOp< internal::scalar_difference_op<Scalar,Scalar>, const VectorType, const VectorType> diagonal() const
   { return sizes(); }
 
   /** \returns the vertex of the bounding box at the corner defined by
@@ -143,7 +143,7 @@
     * For 3D bounding boxes, the following names are added:
     * BottomLeftCeil, BottomRightCeil, TopLeftCeil, TopRightCeil.
     */
-  inline VectorType corner(CornerType corner) const
+  EIGEN_DEVICE_FUNC inline VectorType corner(CornerType corner) const
   {
     EIGEN_STATIC_ASSERT(_AmbientDim <= 3, THIS_METHOD_IS_ONLY_FOR_VECTORS_OF_A_SPECIFIC_SIZE);
 
@@ -161,9 +161,9 @@
 
   /** \returns a random point inside the bounding box sampled with
    * a uniform distribution */
-  inline VectorType sample() const
+  EIGEN_DEVICE_FUNC inline VectorType sample() const
   {
-    VectorType r;
+    VectorType r(dim());
     for(Index d=0; d<dim(); ++d)
     {
       if(!ScalarTraits::IsInteger)
@@ -179,27 +179,27 @@
 
   /** \returns true if the point \a p is inside the box \c *this. */
   template<typename Derived>
-  inline bool contains(const MatrixBase<Derived>& p) const
+  EIGEN_DEVICE_FUNC inline bool contains(const MatrixBase<Derived>& p) const
   {
-    typename internal::nested<Derived,2>::type p_n(p.derived());
+    typename internal::nested_eval<Derived,2>::type p_n(p.derived());
     return (m_min.array()<=p_n.array()).all() && (p_n.array()<=m_max.array()).all();
   }
 
   /** \returns true if the box \a b is entirely inside the box \c *this. */
-  inline bool contains(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline bool contains(const AlignedBox& b) const
   { return (m_min.array()<=(b.min)().array()).all() && ((b.max)().array()<=m_max.array()).all(); }
 
   /** \returns true if the box \a b is intersecting the box \c *this.
    * \sa intersection, clamp */
-  inline bool intersects(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline bool intersects(const AlignedBox& b) const
   { return (m_min.array()<=(b.max)().array()).all() && ((b.min)().array()<=m_max.array()).all(); }
 
   /** Extends \c *this such that it contains the point \a p and returns a reference to \c *this.
    * \sa extend(const AlignedBox&) */
   template<typename Derived>
-  inline AlignedBox& extend(const MatrixBase<Derived>& p)
+  EIGEN_DEVICE_FUNC inline AlignedBox& extend(const MatrixBase<Derived>& p)
   {
-    typename internal::nested<Derived,2>::type p_n(p.derived());
+    typename internal::nested_eval<Derived,2>::type p_n(p.derived());
     m_min = m_min.cwiseMin(p_n);
     m_max = m_max.cwiseMax(p_n);
     return *this;
@@ -207,7 +207,7 @@
 
   /** Extends \c *this such that it contains the box \a b and returns a reference to \c *this.
    * \sa merged, extend(const MatrixBase&) */
-  inline AlignedBox& extend(const AlignedBox& b)
+  EIGEN_DEVICE_FUNC inline AlignedBox& extend(const AlignedBox& b)
   {
     m_min = m_min.cwiseMin(b.m_min);
     m_max = m_max.cwiseMax(b.m_max);
@@ -217,7 +217,7 @@
   /** Clamps \c *this by the box \a b and returns a reference to \c *this.
    * \note If the boxes don't intersect, the resulting box is empty.
    * \sa intersection(), intersects() */
-  inline AlignedBox& clamp(const AlignedBox& b)
+  EIGEN_DEVICE_FUNC inline AlignedBox& clamp(const AlignedBox& b)
   {
     m_min = m_min.cwiseMax(b.m_min);
     m_max = m_max.cwiseMin(b.m_max);
@@ -227,20 +227,20 @@
   /** Returns an AlignedBox that is the intersection of \a b and \c *this
    * \note If the boxes don't intersect, the resulting box is empty.
    * \sa intersects(), clamp, contains()  */
-  inline AlignedBox intersection(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline AlignedBox intersection(const AlignedBox& b) const
   {return AlignedBox(m_min.cwiseMax(b.m_min), m_max.cwiseMin(b.m_max)); }
 
   /** Returns an AlignedBox that is the union of \a b and \c *this.
    * \note Merging with an empty box may result in a box bigger than \c *this. 
    * \sa extend(const AlignedBox&) */
-  inline AlignedBox merged(const AlignedBox& b) const
+  EIGEN_DEVICE_FUNC inline AlignedBox merged(const AlignedBox& b) const
   { return AlignedBox(m_min.cwiseMin(b.m_min), m_max.cwiseMax(b.m_max)); }
 
   /** Translate \c *this by the vector \a t and returns a reference to \c *this. */
   template<typename Derived>
-  inline AlignedBox& translate(const MatrixBase<Derived>& a_t)
+  EIGEN_DEVICE_FUNC inline AlignedBox& translate(const MatrixBase<Derived>& a_t)
   {
-    const typename internal::nested<Derived,2>::type t(a_t.derived());
+    const typename internal::nested_eval<Derived,2>::type t(a_t.derived());
     m_min += t;
     m_max += t;
     return *this;
@@ -251,28 +251,28 @@
     * \sa exteriorDistance(const MatrixBase&), squaredExteriorDistance(const AlignedBox&)
     */
   template<typename Derived>
-  inline Scalar squaredExteriorDistance(const MatrixBase<Derived>& p) const;
+  EIGEN_DEVICE_FUNC inline Scalar squaredExteriorDistance(const MatrixBase<Derived>& p) const;
 
   /** \returns the squared distance between the boxes \a b and \c *this,
     * and zero if the boxes intersect.
     * \sa exteriorDistance(const AlignedBox&), squaredExteriorDistance(const MatrixBase&)
     */
-  inline Scalar squaredExteriorDistance(const AlignedBox& b) const;
+  EIGEN_DEVICE_FUNC inline Scalar squaredExteriorDistance(const AlignedBox& b) const;
 
   /** \returns the distance between the point \a p and the box \c *this,
     * and zero if \a p is inside the box.
     * \sa squaredExteriorDistance(const MatrixBase&), exteriorDistance(const AlignedBox&)
     */
   template<typename Derived>
-  inline NonInteger exteriorDistance(const MatrixBase<Derived>& p) const
-  { using std::sqrt; return sqrt(NonInteger(squaredExteriorDistance(p))); }
+  EIGEN_DEVICE_FUNC inline NonInteger exteriorDistance(const MatrixBase<Derived>& p) const
+  { EIGEN_USING_STD_MATH(sqrt) return sqrt(NonInteger(squaredExteriorDistance(p))); }
 
   /** \returns the distance between the boxes \a b and \c *this,
     * and zero if the boxes intersect.
     * \sa squaredExteriorDistance(const AlignedBox&), exteriorDistance(const MatrixBase&)
     */
-  inline NonInteger exteriorDistance(const AlignedBox& b) const
-  { using std::sqrt; return sqrt(NonInteger(squaredExteriorDistance(b))); }
+  EIGEN_DEVICE_FUNC inline NonInteger exteriorDistance(const AlignedBox& b) const
+  { EIGEN_USING_STD_MATH(sqrt) return sqrt(NonInteger(squaredExteriorDistance(b))); }
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -280,7 +280,7 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<AlignedBox,
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<AlignedBox,
            AlignedBox<NewScalarType,AmbientDimAtCompileTime> >::type cast() const
   {
     return typename internal::cast_return_type<AlignedBox,
@@ -289,7 +289,7 @@
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit AlignedBox(const AlignedBox<OtherScalarType,AmbientDimAtCompileTime>& other)
+  EIGEN_DEVICE_FUNC inline explicit AlignedBox(const AlignedBox<OtherScalarType,AmbientDimAtCompileTime>& other)
   {
     m_min = (other.min)().template cast<Scalar>();
     m_max = (other.max)().template cast<Scalar>();
@@ -299,7 +299,7 @@
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const AlignedBox& other, const RealScalar& prec = ScalarTraits::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const AlignedBox& other, const RealScalar& prec = ScalarTraits::dummy_precision()) const
   { return m_min.isApprox(other.m_min, prec) && m_max.isApprox(other.m_max, prec); }
 
 protected:
@@ -311,9 +311,9 @@
 
 template<typename Scalar,int AmbientDim>
 template<typename Derived>
-inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const MatrixBase<Derived>& a_p) const
+EIGEN_DEVICE_FUNC inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const MatrixBase<Derived>& a_p) const
 {
-  typename internal::nested<Derived,2*AmbientDim>::type p(a_p.derived());
+  typename internal::nested_eval<Derived,2*AmbientDim>::type p(a_p.derived());
   Scalar dist2(0);
   Scalar aux;
   for (Index k=0; k<dim(); ++k)
@@ -333,7 +333,7 @@
 }
 
 template<typename Scalar,int AmbientDim>
-inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const AlignedBox& b) const
+EIGEN_DEVICE_FUNC inline Scalar AlignedBox<Scalar,AmbientDim>::squaredExteriorDistance(const AlignedBox& b) const
 {
   Scalar dist2(0);
   Scalar aux;
diff --git a/Eigen/src/Geometry/AngleAxis.h b/Eigen/src/Geometry/AngleAxis.h
index 553d38c..83ee1be 100644
--- a/Eigen/src/Geometry/AngleAxis.h
+++ b/Eigen/src/Geometry/AngleAxis.h
@@ -69,50 +69,61 @@
 public:
 
   /** Default constructor without initialization. */
-  AngleAxis() {}
+  EIGEN_DEVICE_FUNC AngleAxis() {}
   /** Constructs and initialize the angle-axis rotation from an \a angle in radian
     * and an \a axis which \b must \b be \b normalized.
     *
     * \warning If the \a axis vector is not normalized, then the angle-axis object
     *          represents an invalid rotation. */
   template<typename Derived>
+  EIGEN_DEVICE_FUNC 
   inline AngleAxis(const Scalar& angle, const MatrixBase<Derived>& axis) : m_axis(axis), m_angle(angle) {}
-  /** Constructs and initialize the angle-axis rotation from a quaternion \a q. */
-  template<typename QuatDerived> inline explicit AngleAxis(const QuaternionBase<QuatDerived>& q) { *this = q; }
+  /** Constructs and initialize the angle-axis rotation from a quaternion \a q.
+    * This function implicitly normalizes the quaternion \a q.
+    */
+  template<typename QuatDerived> 
+  EIGEN_DEVICE_FUNC inline explicit AngleAxis(const QuaternionBase<QuatDerived>& q) { *this = q; }
   /** Constructs and initialize the angle-axis rotation from a 3x3 rotation matrix. */
   template<typename Derived>
-  inline explicit AngleAxis(const MatrixBase<Derived>& m) { *this = m; }
+  EIGEN_DEVICE_FUNC inline explicit AngleAxis(const MatrixBase<Derived>& m) { *this = m; }
 
-  Scalar angle() const { return m_angle; }
-  Scalar& angle() { return m_angle; }
+  /** \returns the value of the rotation angle in radian */
+  EIGEN_DEVICE_FUNC Scalar angle() const { return m_angle; }
+  /** \returns a read-write reference to the stored angle in radian */
+  EIGEN_DEVICE_FUNC Scalar& angle() { return m_angle; }
 
-  const Vector3& axis() const { return m_axis; }
-  Vector3& axis() { return m_axis; }
+  /** \returns the rotation axis */
+  EIGEN_DEVICE_FUNC const Vector3& axis() const { return m_axis; }
+  /** \returns a read-write reference to the stored rotation axis.
+    *
+    * \warning The rotation axis must remain a \b unit vector.
+    */
+  EIGEN_DEVICE_FUNC Vector3& axis() { return m_axis; }
 
   /** Concatenates two rotations */
-  inline QuaternionType operator* (const AngleAxis& other) const
+  EIGEN_DEVICE_FUNC inline QuaternionType operator* (const AngleAxis& other) const
   { return QuaternionType(*this) * QuaternionType(other); }
 
   /** Concatenates two rotations */
-  inline QuaternionType operator* (const QuaternionType& other) const
+  EIGEN_DEVICE_FUNC inline QuaternionType operator* (const QuaternionType& other) const
   { return QuaternionType(*this) * other; }
 
   /** Concatenates two rotations */
-  friend inline QuaternionType operator* (const QuaternionType& a, const AngleAxis& b)
+  friend EIGEN_DEVICE_FUNC inline QuaternionType operator* (const QuaternionType& a, const AngleAxis& b)
   { return a * QuaternionType(b); }
 
   /** \returns the inverse rotation, i.e., an angle-axis with opposite rotation angle */
-  AngleAxis inverse() const
+  EIGEN_DEVICE_FUNC AngleAxis inverse() const
   { return AngleAxis(-m_angle, m_axis); }
 
   template<class QuatDerived>
-  AngleAxis& operator=(const QuaternionBase<QuatDerived>& q);
+  EIGEN_DEVICE_FUNC AngleAxis& operator=(const QuaternionBase<QuatDerived>& q);
   template<typename Derived>
-  AngleAxis& operator=(const MatrixBase<Derived>& m);
+  EIGEN_DEVICE_FUNC AngleAxis& operator=(const MatrixBase<Derived>& m);
 
   template<typename Derived>
-  AngleAxis& fromRotationMatrix(const MatrixBase<Derived>& m);
-  Matrix3 toRotationMatrix(void) const;
+  EIGEN_DEVICE_FUNC AngleAxis& fromRotationMatrix(const MatrixBase<Derived>& m);
+  EIGEN_DEVICE_FUNC Matrix3 toRotationMatrix(void) const;
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -120,24 +131,24 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type cast() const
   { return typename internal::cast_return_type<AngleAxis,AngleAxis<NewScalarType> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit AngleAxis(const AngleAxis<OtherScalarType>& other)
+  EIGEN_DEVICE_FUNC inline explicit AngleAxis(const AngleAxis<OtherScalarType>& other)
   {
     m_axis = other.axis().template cast<Scalar>();
     m_angle = Scalar(other.angle());
   }
 
-  static inline const AngleAxis Identity() { return AngleAxis(0, Vector3::UnitX()); }
+  EIGEN_DEVICE_FUNC static inline const AngleAxis Identity() { return AngleAxis(Scalar(0), Vector3::UnitX()); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const AngleAxis& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const AngleAxis& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_axis.isApprox(other.m_axis, prec) && internal::isApprox(m_angle,other.m_angle, prec); }
 };
 
@@ -149,29 +160,32 @@
 typedef AngleAxis<double> AngleAxisd;
 
 /** Set \c *this from a \b unit quaternion.
-  * The axis is normalized.
+  *
+  * The resulting axis is normalized, and the computed angle is in the [0,pi] range.
   * 
-  * \warning As any other method dealing with quaternion, if the input quaternion
-  *          is not normalized then the result is undefined.
+  * This function implicitly normalizes the quaternion \a q.
   */
 template<typename Scalar>
 template<typename QuatDerived>
-AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const QuaternionBase<QuatDerived>& q)
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const QuaternionBase<QuatDerived>& q)
 {
-  using std::acos;
-  using std::min;
-  using std::max;
-  using std::sqrt;
-  Scalar n2 = q.vec().squaredNorm();
-  if (n2 < NumTraits<Scalar>::dummy_precision()*NumTraits<Scalar>::dummy_precision())
+  EIGEN_USING_STD_MATH(atan2)
+  EIGEN_USING_STD_MATH(abs)
+  Scalar n = q.vec().norm();
+  if(n<NumTraits<Scalar>::epsilon())
+    n = q.vec().stableNorm();
+
+  if (n != Scalar(0))
   {
-    m_angle = 0;
-    m_axis << 1, 0, 0;
+    m_angle = Scalar(2)*atan2(n, abs(q.w()));
+    if(q.w() < Scalar(0))
+      n = -n;
+    m_axis  = q.vec() / n;
   }
   else
   {
-    m_angle = Scalar(2)*acos((min)((max)(Scalar(-1),q.w()),Scalar(1)));
-    m_axis = q.vec() / sqrt(n2);
+    m_angle = Scalar(0);
+    m_axis << Scalar(1), Scalar(0), Scalar(0);
   }
   return *this;
 }
@@ -180,7 +194,7 @@
   */
 template<typename Scalar>
 template<typename Derived>
-AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const MatrixBase<Derived>& mat)
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>& AngleAxis<Scalar>::operator=(const MatrixBase<Derived>& mat)
 {
   // Since a direct conversion would not be really faster,
   // let's use the robust Quaternion implementation:
@@ -192,7 +206,7 @@
 **/
 template<typename Scalar>
 template<typename Derived>
-AngleAxis<Scalar>& AngleAxis<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>& AngleAxis<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
 {
   return *this = QuaternionType(mat);
 }
@@ -201,10 +215,10 @@
   */
 template<typename Scalar>
 typename AngleAxis<Scalar>::Matrix3
-AngleAxis<Scalar>::toRotationMatrix(void) const
+EIGEN_DEVICE_FUNC AngleAxis<Scalar>::toRotationMatrix(void) const
 {
-  using std::sin;
-  using std::cos;
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
   Matrix3 res;
   Vector3 sin_axis  = sin(m_angle) * m_axis;
   Scalar c = cos(m_angle);
diff --git a/Eigen/src/Geometry/CMakeLists.txt b/Eigen/src/Geometry/CMakeLists.txt
deleted file mode 100644
index f8f728b..0000000
--- a/Eigen/src/Geometry/CMakeLists.txt
+++ /dev/null
@@ -1,8 +0,0 @@
-FILE(GLOB Eigen_Geometry_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Geometry_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Geometry COMPONENT Devel
-  )
-
-ADD_SUBDIRECTORY(arch)
diff --git a/Eigen/src/Geometry/EulerAngles.h b/Eigen/src/Geometry/EulerAngles.h
index 82802fb..c633268 100644
--- a/Eigen/src/Geometry/EulerAngles.h
+++ b/Eigen/src/Geometry/EulerAngles.h
@@ -33,12 +33,12 @@
   * \sa class AngleAxis
   */
 template<typename Derived>
-inline Matrix<typename MatrixBase<Derived>::Scalar,3,1>
+EIGEN_DEVICE_FUNC inline Matrix<typename MatrixBase<Derived>::Scalar,3,1>
 MatrixBase<Derived>::eulerAngles(Index a0, Index a1, Index a2) const
 {
-  using std::atan2;
-  using std::sin;
-  using std::cos;
+  EIGEN_USING_STD_MATH(atan2)
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
   /* Implemented from Graphics Gems IV */
   EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(Derived,3,3)
 
@@ -55,7 +55,12 @@
     res[0] = atan2(coeff(j,i), coeff(k,i));
     if((odd && res[0]<Scalar(0)) || ((!odd) && res[0]>Scalar(0)))
     {
-      res[0] = (res[0] > Scalar(0)) ? res[0] - Scalar(M_PI) : res[0] + Scalar(M_PI);
+      if(res[0] > Scalar(0)) {
+        res[0] -= Scalar(EIGEN_PI);
+      }
+      else {
+        res[0] += Scalar(EIGEN_PI);
+      }
       Scalar s2 = Vector2(coeff(j,i), coeff(k,i)).norm();
       res[1] = -atan2(s2, coeff(i,i));
     }
@@ -84,7 +89,12 @@
     res[0] = atan2(coeff(j,k), coeff(k,k));
     Scalar c2 = Vector2(coeff(i,i), coeff(i,j)).norm();
     if((odd && res[0]<Scalar(0)) || ((!odd) && res[0]>Scalar(0))) {
-      res[0] = (res[0] > Scalar(0)) ? res[0] - Scalar(M_PI) : res[0] + Scalar(M_PI);
+      if(res[0] > Scalar(0)) {
+        res[0] -= Scalar(EIGEN_PI);
+      }
+      else {
+        res[0] += Scalar(EIGEN_PI);
+      }
       res[1] = atan2(-coeff(i,k), -c2);
     }
     else
diff --git a/Eigen/src/Geometry/Homogeneous.h b/Eigen/src/Geometry/Homogeneous.h
index 372e422..5f0da1a 100644
--- a/Eigen/src/Geometry/Homogeneous.h
+++ b/Eigen/src/Geometry/Homogeneous.h
@@ -34,7 +34,7 @@
  : traits<MatrixType>
 {
   typedef typename traits<MatrixType>::StorageKind StorageKind;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
   enum {
     RowsPlusOne = (MatrixType::RowsAtCompileTime != Dynamic) ?
@@ -48,8 +48,7 @@
     TmpFlags = _MatrixTypeNested::Flags & HereditaryBits,
     Flags = ColsAtCompileTime==1 ? (TmpFlags & ~RowMajorBit)
           : RowsAtCompileTime==1 ? (TmpFlags | RowMajorBit)
-          : TmpFlags,
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost
+          : TmpFlags
   };
 };
 
@@ -59,102 +58,117 @@
 } // end namespace internal
 
 template<typename MatrixType,int _Direction> class Homogeneous
-  : internal::no_assignment_operator, public MatrixBase<Homogeneous<MatrixType,_Direction> >
+  : public MatrixBase<Homogeneous<MatrixType,_Direction> >, internal::no_assignment_operator
 {
   public:
 
+    typedef MatrixType NestedExpression;
     enum { Direction = _Direction };
 
     typedef MatrixBase<Homogeneous> Base;
     EIGEN_DENSE_PUBLIC_INTERFACE(Homogeneous)
 
-    inline Homogeneous(const MatrixType& matrix)
+    EIGEN_DEVICE_FUNC explicit inline Homogeneous(const MatrixType& matrix)
       : m_matrix(matrix)
     {}
 
-    inline Index rows() const { return m_matrix.rows() + (int(Direction)==Vertical   ? 1 : 0); }
-    inline Index cols() const { return m_matrix.cols() + (int(Direction)==Horizontal ? 1 : 0); }
-
-    inline Scalar coeff(Index row, Index col) const
-    {
-      if(  (int(Direction)==Vertical   && row==m_matrix.rows())
-        || (int(Direction)==Horizontal && col==m_matrix.cols()))
-        return Scalar(1);
-      return m_matrix.coeff(row, col);
-    }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_matrix.rows() + (int(Direction)==Vertical   ? 1 : 0); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_matrix.cols() + (int(Direction)==Horizontal ? 1 : 0); }
+    
+    EIGEN_DEVICE_FUNC const NestedExpression& nestedExpression() const { return m_matrix; }
 
     template<typename Rhs>
-    inline const internal::homogeneous_right_product_impl<Homogeneous,Rhs>
+    EIGEN_DEVICE_FUNC inline const Product<Homogeneous,Rhs>
     operator* (const MatrixBase<Rhs>& rhs) const
     {
       eigen_assert(int(Direction)==Horizontal);
-      return internal::homogeneous_right_product_impl<Homogeneous,Rhs>(m_matrix,rhs.derived());
+      return Product<Homogeneous,Rhs>(*this,rhs.derived());
     }
 
     template<typename Lhs> friend
-    inline const internal::homogeneous_left_product_impl<Homogeneous,Lhs>
+    EIGEN_DEVICE_FUNC inline const Product<Lhs,Homogeneous>
     operator* (const MatrixBase<Lhs>& lhs, const Homogeneous& rhs)
     {
       eigen_assert(int(Direction)==Vertical);
-      return internal::homogeneous_left_product_impl<Homogeneous,Lhs>(lhs.derived(),rhs.m_matrix);
+      return Product<Lhs,Homogeneous>(lhs.derived(),rhs);
     }
 
     template<typename Scalar, int Dim, int Mode, int Options> friend
-    inline const internal::homogeneous_left_product_impl<Homogeneous,Transform<Scalar,Dim,Mode,Options> >
+    EIGEN_DEVICE_FUNC inline const Product<Transform<Scalar,Dim,Mode,Options>, Homogeneous >
     operator* (const Transform<Scalar,Dim,Mode,Options>& lhs, const Homogeneous& rhs)
     {
       eigen_assert(int(Direction)==Vertical);
-      return internal::homogeneous_left_product_impl<Homogeneous,Transform<Scalar,Dim,Mode,Options> >(lhs,rhs.m_matrix);
+      return Product<Transform<Scalar,Dim,Mode,Options>, Homogeneous>(lhs,rhs);
+    }
+
+    template<typename Func>
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::result_of<Func(Scalar,Scalar)>::type
+    redux(const Func& func) const
+    {
+      return func(m_matrix.redux(func), Scalar(1));
     }
 
   protected:
     typename MatrixType::Nested m_matrix;
 };
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
-  * \return an expression of the equivalent homogeneous vector
+  * \returns a vector expression that is one longer than the vector argument, with the value 1 symbolically appended as the last coefficient.
+  *
+  * This can be used to convert affine coordinates to homogeneous coordinates.
   *
   * \only_for_vectors
   *
   * Example: \include MatrixBase_homogeneous.cpp
   * Output: \verbinclude MatrixBase_homogeneous.out
   *
-  * \sa class Homogeneous
+  * \sa VectorwiseOp::homogeneous(), class Homogeneous
   */
 template<typename Derived>
-inline typename MatrixBase<Derived>::HomogeneousReturnType
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::HomogeneousReturnType
 MatrixBase<Derived>::homogeneous() const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
-  return derived();
+  return HomogeneousReturnType(derived());
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
-  * \returns a matrix expression of homogeneous column (or row) vectors
+  * \returns an expression where the value 1 is symbolically appended as the final coefficient to each column (or row) of the matrix.
+  *
+  * This can be used to convert affine coordinates to homogeneous coordinates.
   *
   * Example: \include VectorwiseOp_homogeneous.cpp
   * Output: \verbinclude VectorwiseOp_homogeneous.out
   *
-  * \sa MatrixBase::homogeneous() */
+  * \sa MatrixBase::homogeneous(), class Homogeneous */
 template<typename ExpressionType, int Direction>
-inline Homogeneous<ExpressionType,Direction>
+EIGEN_DEVICE_FUNC inline Homogeneous<ExpressionType,Direction>
 VectorwiseOp<ExpressionType,Direction>::homogeneous() const
 {
-  return _expression();
+  return HomogeneousReturnType(_expression());
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
-  * \returns an expression of the homogeneous normalized vector of \c *this
+  * \brief homogeneous normalization
+  *
+  * \returns a vector expression of the N-1 first coefficients of \c *this divided by that last coefficient.
+  *
+  * This can be used to convert homogeneous coordinates to affine coordinates.
+  *
+  * It is essentially a shortcut for:
+  * \code
+    this->head(this->size()-1)/this->coeff(this->size()-1);
+    \endcode
   *
   * Example: \include MatrixBase_hnormalized.cpp
   * Output: \verbinclude MatrixBase_hnormalized.out
   *
   * \sa VectorwiseOp::hnormalized() */
 template<typename Derived>
-inline const typename MatrixBase<Derived>::HNormalizedReturnType
+EIGEN_DEVICE_FUNC inline const typename MatrixBase<Derived>::HNormalizedReturnType
 MatrixBase<Derived>::hnormalized() const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
@@ -163,16 +177,22 @@
     ColsAtCompileTime==1?1:size()-1) / coeff(size()-1);
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
-  * \returns an expression of the homogeneous normalized vector of \c *this
+  * \brief column or row-wise homogeneous normalization
+  *
+  * \returns an expression of the first N-1 coefficients of each column (or row) of \c *this divided by the last coefficient of each column (or row).
+  *
+  * This can be used to convert homogeneous coordinates to affine coordinates.
+  *
+  * It is conceptually equivalent to calling MatrixBase::hnormalized() to each column (or row) of \c *this.
   *
   * Example: \include DirectionWise_hnormalized.cpp
   * Output: \verbinclude DirectionWise_hnormalized.out
   *
   * \sa MatrixBase::hnormalized() */
 template<typename ExpressionType, int Direction>
-inline const typename VectorwiseOp<ExpressionType,Direction>::HNormalizedReturnType
+EIGEN_DEVICE_FUNC inline const typename VectorwiseOp<ExpressionType,Direction>::HNormalizedReturnType
 VectorwiseOp<ExpressionType,Direction>::hnormalized() const
 {
   return HNormalized_Block(_expression(),0,0,
@@ -196,7 +216,7 @@
 struct take_matrix_for_product
 {
   typedef MatrixOrTransformType type;
-  static const type& run(const type &x) { return x; }
+  EIGEN_DEVICE_FUNC static const type& run(const type &x) { return x; }
 };
 
 template<typename Scalar, int Dim, int Mode,int Options>
@@ -204,7 +224,7 @@
 {
   typedef Transform<Scalar, Dim, Mode, Options> TransformType;
   typedef typename internal::add_const<typename TransformType::ConstAffinePart>::type type;
-  static type run (const TransformType& x) { return x.affine(); }
+  EIGEN_DEVICE_FUNC static type run (const TransformType& x) { return x.affine(); }
 };
 
 template<typename Scalar, int Dim, int Options>
@@ -212,7 +232,7 @@
 {
   typedef Transform<Scalar, Dim, Projective, Options> TransformType;
   typedef typename TransformType::MatrixType type;
-  static const type& run (const TransformType& x) { return x.matrix(); }
+  EIGEN_DEVICE_FUNC static const type& run (const TransformType& x) { return x.matrix(); }
 };
 
 template<typename MatrixType,typename Lhs>
@@ -237,16 +257,15 @@
   typedef typename traits<homogeneous_left_product_impl>::LhsMatrixType LhsMatrixType;
   typedef typename remove_all<LhsMatrixType>::type LhsMatrixTypeCleaned;
   typedef typename remove_all<typename LhsMatrixTypeCleaned::Nested>::type LhsMatrixTypeNested;
-  typedef typename MatrixType::Index Index;
-  homogeneous_left_product_impl(const Lhs& lhs, const MatrixType& rhs)
+  EIGEN_DEVICE_FUNC homogeneous_left_product_impl(const Lhs& lhs, const MatrixType& rhs)
     : m_lhs(take_matrix_for_product<Lhs>::run(lhs)),
       m_rhs(rhs)
   {}
 
-  inline Index rows() const { return m_lhs.rows(); }
-  inline Index cols() const { return m_rhs.cols(); }
+  EIGEN_DEVICE_FUNC inline Index rows() const { return m_lhs.rows(); }
+  EIGEN_DEVICE_FUNC inline Index cols() const { return m_rhs.cols(); }
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  template<typename Dest> EIGEN_DEVICE_FUNC void evalTo(Dest& dst) const
   {
     // FIXME investigate how to allow lazy evaluation of this product when possible
     dst = Block<const LhsMatrixTypeNested,
@@ -277,15 +296,14 @@
   : public ReturnByValue<homogeneous_right_product_impl<Homogeneous<MatrixType,Horizontal>,Rhs> >
 {
   typedef typename remove_all<typename Rhs::Nested>::type RhsNested;
-  typedef typename MatrixType::Index Index;
-  homogeneous_right_product_impl(const MatrixType& lhs, const Rhs& rhs)
+  EIGEN_DEVICE_FUNC homogeneous_right_product_impl(const MatrixType& lhs, const Rhs& rhs)
     : m_lhs(lhs), m_rhs(rhs)
   {}
 
-  inline Index rows() const { return m_lhs.rows(); }
-  inline Index cols() const { return m_rhs.cols(); }
+  EIGEN_DEVICE_FUNC inline Index rows() const { return m_lhs.rows(); }
+  EIGEN_DEVICE_FUNC inline Index cols() const { return m_rhs.cols(); }
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  template<typename Dest> EIGEN_DEVICE_FUNC void evalTo(Dest& dst) const
   {
     // FIXME investigate how to allow lazy evaluation of this product when possible
     dst = m_lhs * Block<const RhsNested,
@@ -300,6 +318,178 @@
   typename Rhs::Nested m_rhs;
 };
 
+template<typename ArgType,int Direction>
+struct evaluator_traits<Homogeneous<ArgType,Direction> >
+{
+  typedef typename storage_kind_to_evaluator_kind<typename ArgType::StorageKind>::Kind Kind;
+  typedef HomogeneousShape Shape;  
+};
+
+template<> struct AssignmentKind<DenseShape,HomogeneousShape> { typedef Dense2Dense Kind; };
+
+
+template<typename ArgType,int Direction>
+struct unary_evaluator<Homogeneous<ArgType,Direction>, IndexBased>
+  : evaluator<typename Homogeneous<ArgType,Direction>::PlainObject >
+{
+  typedef Homogeneous<ArgType,Direction> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op)
+    : Base(), m_temp(op)
+  {
+    ::new (static_cast<Base*>(this)) Base(m_temp);
+  }
+
+protected:
+  PlainObject m_temp;
+};
+
+// dense = homogeneous
+template< typename DstXprType, typename ArgType, typename Scalar>
+struct Assignment<DstXprType, Homogeneous<ArgType,Vertical>, internal::assign_op<Scalar,typename ArgType::Scalar>, Dense2Dense>
+{
+  typedef Homogeneous<ArgType,Vertical> SrcXprType;
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename ArgType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst.template topRows<ArgType::RowsAtCompileTime>(src.nestedExpression().rows()) = src.nestedExpression();
+    dst.row(dst.rows()-1).setOnes();
+  }
+};
+
+// dense = homogeneous
+template< typename DstXprType, typename ArgType, typename Scalar>
+struct Assignment<DstXprType, Homogeneous<ArgType,Horizontal>, internal::assign_op<Scalar,typename ArgType::Scalar>, Dense2Dense>
+{
+  typedef Homogeneous<ArgType,Horizontal> SrcXprType;
+  EIGEN_DEVICE_FUNC static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename ArgType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst.template leftCols<ArgType::ColsAtCompileTime>(src.nestedExpression().cols()) = src.nestedExpression();
+    dst.col(dst.cols()-1).setOnes();
+  }
+};
+
+template<typename LhsArg, typename Rhs, int ProductTag>
+struct generic_product_impl<Homogeneous<LhsArg,Horizontal>, Rhs, HomogeneousShape, DenseShape, ProductTag>
+{
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC static void evalTo(Dest& dst, const Homogeneous<LhsArg,Horizontal>& lhs, const Rhs& rhs)
+  {
+    homogeneous_right_product_impl<Homogeneous<LhsArg,Horizontal>, Rhs>(lhs.nestedExpression(), rhs).evalTo(dst);
+  }
+};
+
+template<typename Lhs,typename Rhs>
+struct homogeneous_right_product_refactoring_helper
+{
+  enum {
+    Dim  = Lhs::ColsAtCompileTime,
+    Rows = Lhs::RowsAtCompileTime
+  };
+  typedef typename Rhs::template ConstNRowsBlockXpr<Dim>::Type          LinearBlockConst;
+  typedef typename remove_const<LinearBlockConst>::type                 LinearBlock;
+  typedef typename Rhs::ConstRowXpr                                     ConstantColumn;
+  typedef Replicate<const ConstantColumn,Rows,1>                        ConstantBlock;
+  typedef Product<Lhs,LinearBlock,LazyProduct>                          LinearProduct;
+  typedef CwiseBinaryOp<internal::scalar_sum_op<typename Lhs::Scalar,typename Rhs::Scalar>, const LinearProduct, const ConstantBlock> Xpr;
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, HomogeneousShape, DenseShape>
+ : public evaluator<typename homogeneous_right_product_refactoring_helper<typename Lhs::NestedExpression,Rhs>::Xpr>
+{
+  typedef Product<Lhs, Rhs, LazyProduct> XprType;
+  typedef homogeneous_right_product_refactoring_helper<typename Lhs::NestedExpression,Rhs> helper;
+  typedef typename helper::ConstantBlock ConstantBlock;
+  typedef typename helper::Xpr RefactoredXpr;
+  typedef evaluator<RefactoredXpr> Base;
+  
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(  xpr.lhs().nestedExpression() .lazyProduct(  xpr.rhs().template topRows<helper::Dim>(xpr.lhs().nestedExpression().cols()) )
+            + ConstantBlock(xpr.rhs().row(xpr.rhs().rows()-1),xpr.lhs().rows(), 1) )
+  {}
+};
+
+template<typename Lhs, typename RhsArg, int ProductTag>
+struct generic_product_impl<Lhs, Homogeneous<RhsArg,Vertical>, DenseShape, HomogeneousShape, ProductTag>
+{
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC static void evalTo(Dest& dst, const Lhs& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
+  {
+    homogeneous_left_product_impl<Homogeneous<RhsArg,Vertical>, Lhs>(lhs, rhs.nestedExpression()).evalTo(dst);
+  }
+};
+
+// TODO: the following specialization is to address a regression from 3.2 to 3.3
+// In the future, this path should be optimized.
+template<typename Lhs, typename RhsArg, int ProductTag>
+struct generic_product_impl<Lhs, Homogeneous<RhsArg,Vertical>, TriangularShape, HomogeneousShape, ProductTag>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
+  {
+    dst.noalias() = lhs * rhs.eval();
+  }
+};
+
+template<typename Lhs,typename Rhs>
+struct homogeneous_left_product_refactoring_helper
+{
+  enum {
+    Dim = Rhs::RowsAtCompileTime,
+    Cols = Rhs::ColsAtCompileTime
+  };
+  typedef typename Lhs::template ConstNColsBlockXpr<Dim>::Type          LinearBlockConst;
+  typedef typename remove_const<LinearBlockConst>::type                 LinearBlock;
+  typedef typename Lhs::ConstColXpr                                     ConstantColumn;
+  typedef Replicate<const ConstantColumn,1,Cols>                        ConstantBlock;
+  typedef Product<LinearBlock,Rhs,LazyProduct>                          LinearProduct;
+  typedef CwiseBinaryOp<internal::scalar_sum_op<typename Lhs::Scalar,typename Rhs::Scalar>, const LinearProduct, const ConstantBlock> Xpr;
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, LazyProduct>, ProductTag, DenseShape, HomogeneousShape>
+ : public evaluator<typename homogeneous_left_product_refactoring_helper<Lhs,typename Rhs::NestedExpression>::Xpr>
+{
+  typedef Product<Lhs, Rhs, LazyProduct> XprType;
+  typedef homogeneous_left_product_refactoring_helper<Lhs,typename Rhs::NestedExpression> helper;
+  typedef typename helper::ConstantBlock ConstantBlock;
+  typedef typename helper::Xpr RefactoredXpr;
+  typedef evaluator<RefactoredXpr> Base;
+  
+  EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr)
+    : Base(   xpr.lhs().template leftCols<helper::Dim>(xpr.rhs().nestedExpression().rows()) .lazyProduct( xpr.rhs().nestedExpression() )
+            + ConstantBlock(xpr.lhs().col(xpr.lhs().cols()-1),1,xpr.rhs().cols()) )
+  {}
+};
+
+template<typename Scalar, int Dim, int Mode,int Options, typename RhsArg, int ProductTag>
+struct generic_product_impl<Transform<Scalar,Dim,Mode,Options>, Homogeneous<RhsArg,Vertical>, DenseShape, HomogeneousShape, ProductTag>
+{
+  typedef Transform<Scalar,Dim,Mode,Options> TransformType;
+  template<typename Dest>
+  EIGEN_DEVICE_FUNC static void evalTo(Dest& dst, const TransformType& lhs, const Homogeneous<RhsArg,Vertical>& rhs)
+  {
+    homogeneous_left_product_impl<Homogeneous<RhsArg,Vertical>, TransformType>(lhs, rhs.nestedExpression()).evalTo(dst);
+  }
+};
+
+template<typename ExpressionType, int Side, bool Transposed>
+struct permutation_matrix_product<ExpressionType, Side, Transposed, HomogeneousShape>
+  : public permutation_matrix_product<ExpressionType, Side, Transposed, DenseShape>
+{};
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Geometry/Hyperplane.h b/Eigen/src/Geometry/Hyperplane.h
index 00b7c43..05929b2 100644
--- a/Eigen/src/Geometry/Hyperplane.h
+++ b/Eigen/src/Geometry/Hyperplane.h
@@ -22,8 +22,8 @@
   * A hyperplane is an affine subspace of dimension n-1 in a space of dimension n.
   * For example, a hyperplane in a plane is a line; a hyperplane in 3-space is a plane.
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  * \param _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients
+  * \tparam _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
   *             Notice that the dimension of the hyperplane is _AmbientDim-1.
   *
   * This class represents an hyperplane as the zero set of the implicit equation
@@ -41,7 +41,7 @@
   };
   typedef _Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef DenseIndex Index;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
   typedef Matrix<Scalar,AmbientDimAtCompileTime,1> VectorType;
   typedef Matrix<Scalar,Index(AmbientDimAtCompileTime)==Dynamic
                         ? Dynamic
@@ -50,21 +50,21 @@
   typedef const Block<const Coefficients,AmbientDimAtCompileTime,1> ConstNormalReturnType;
 
   /** Default constructor without initialization */
-  inline Hyperplane() {}
+  EIGEN_DEVICE_FUNC inline Hyperplane() {}
   
   template<int OtherOptions>
-  Hyperplane(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC Hyperplane(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
    : m_coeffs(other.coeffs())
   {}
 
   /** Constructs a dynamic-size hyperplane with \a _dim the dimension
     * of the ambient space */
-  inline explicit Hyperplane(Index _dim) : m_coeffs(_dim+1) {}
+  EIGEN_DEVICE_FUNC inline explicit Hyperplane(Index _dim) : m_coeffs(_dim+1) {}
 
   /** Construct a plane from its normal \a n and a point \a e onto the plane.
     * \warning the vector normal is assumed to be normalized.
     */
-  inline Hyperplane(const VectorType& n, const VectorType& e)
+  EIGEN_DEVICE_FUNC inline Hyperplane(const VectorType& n, const VectorType& e)
     : m_coeffs(n.size()+1)
   {
     normal() = n;
@@ -75,7 +75,7 @@
     * such that the algebraic equation of the plane is \f$ n \cdot x + d = 0 \f$.
     * \warning the vector normal is assumed to be normalized.
     */
-  inline Hyperplane(const VectorType& n, const Scalar& d)
+  EIGEN_DEVICE_FUNC inline Hyperplane(const VectorType& n, const Scalar& d)
     : m_coeffs(n.size()+1)
   {
     normal() = n;
@@ -85,7 +85,7 @@
   /** Constructs a hyperplane passing through the two points. If the dimension of the ambient space
     * is greater than 2, then there isn't uniqueness, so an arbitrary choice is made.
     */
-  static inline Hyperplane Through(const VectorType& p0, const VectorType& p1)
+  EIGEN_DEVICE_FUNC static inline Hyperplane Through(const VectorType& p0, const VectorType& p1)
   {
     Hyperplane result(p0.size());
     result.normal() = (p1 - p0).unitOrthogonal();
@@ -96,7 +96,7 @@
   /** Constructs a hyperplane passing through the three points. The dimension of the ambient space
     * is required to be exactly 3.
     */
-  static inline Hyperplane Through(const VectorType& p0, const VectorType& p1, const VectorType& p2)
+  EIGEN_DEVICE_FUNC static inline Hyperplane Through(const VectorType& p0, const VectorType& p1, const VectorType& p2)
   {
     EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 3)
     Hyperplane result(p0.size());
@@ -120,19 +120,19 @@
     * so an arbitrary choice is made.
     */
   // FIXME to be consitent with the rest this could be implemented as a static Through function ??
-  explicit Hyperplane(const ParametrizedLine<Scalar, AmbientDimAtCompileTime>& parametrized)
+  EIGEN_DEVICE_FUNC explicit Hyperplane(const ParametrizedLine<Scalar, AmbientDimAtCompileTime>& parametrized)
   {
     normal() = parametrized.direction().unitOrthogonal();
     offset() = -parametrized.origin().dot(normal());
   }
 
-  ~Hyperplane() {}
+  EIGEN_DEVICE_FUNC ~Hyperplane() {}
 
   /** \returns the dimension in which the plane holds */
-  inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_coeffs.size()-1 : Index(AmbientDimAtCompileTime); }
+  EIGEN_DEVICE_FUNC inline Index dim() const { return AmbientDimAtCompileTime==Dynamic ? m_coeffs.size()-1 : Index(AmbientDimAtCompileTime); }
 
   /** normalizes \c *this */
-  void normalize(void)
+  EIGEN_DEVICE_FUNC void normalize(void)
   {
     m_coeffs /= normal().norm();
   }
@@ -140,45 +140,45 @@
   /** \returns the signed distance between the plane \c *this and a point \a p.
     * \sa absDistance()
     */
-  inline Scalar signedDistance(const VectorType& p) const { return normal().dot(p) + offset(); }
+  EIGEN_DEVICE_FUNC inline Scalar signedDistance(const VectorType& p) const { return normal().dot(p) + offset(); }
 
   /** \returns the absolute distance between the plane \c *this and a point \a p.
     * \sa signedDistance()
     */
-  inline Scalar absDistance(const VectorType& p) const { using std::abs; return abs(signedDistance(p)); }
+  EIGEN_DEVICE_FUNC inline Scalar absDistance(const VectorType& p) const { return numext::abs(signedDistance(p)); }
 
   /** \returns the projection of a point \a p onto the plane \c *this.
     */
-  inline VectorType projection(const VectorType& p) const { return p - signedDistance(p) * normal(); }
+  EIGEN_DEVICE_FUNC inline VectorType projection(const VectorType& p) const { return p - signedDistance(p) * normal(); }
 
   /** \returns a constant reference to the unit normal vector of the plane, which corresponds
     * to the linear part of the implicit equation.
     */
-  inline ConstNormalReturnType normal() const { return ConstNormalReturnType(m_coeffs,0,0,dim(),1); }
+  EIGEN_DEVICE_FUNC inline ConstNormalReturnType normal() const { return ConstNormalReturnType(m_coeffs,0,0,dim(),1); }
 
   /** \returns a non-constant reference to the unit normal vector of the plane, which corresponds
     * to the linear part of the implicit equation.
     */
-  inline NormalReturnType normal() { return NormalReturnType(m_coeffs,0,0,dim(),1); }
+  EIGEN_DEVICE_FUNC inline NormalReturnType normal() { return NormalReturnType(m_coeffs,0,0,dim(),1); }
 
   /** \returns the distance to the origin, which is also the "constant term" of the implicit equation
     * \warning the vector normal is assumed to be normalized.
     */
-  inline const Scalar& offset() const { return m_coeffs.coeff(dim()); }
+  EIGEN_DEVICE_FUNC inline const Scalar& offset() const { return m_coeffs.coeff(dim()); }
 
   /** \returns a non-constant reference to the distance to the origin, which is also the constant part
     * of the implicit equation */
-  inline Scalar& offset() { return m_coeffs(dim()); }
+  EIGEN_DEVICE_FUNC inline Scalar& offset() { return m_coeffs(dim()); }
 
   /** \returns a constant reference to the coefficients c_i of the plane equation:
     * \f$ c_0*x_0 + ... + c_{d-1}*x_{d-1} + c_d = 0 \f$
     */
-  inline const Coefficients& coeffs() const { return m_coeffs; }
+  EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs; }
 
   /** \returns a non-constant reference to the coefficients c_i of the plane equation:
     * \f$ c_0*x_0 + ... + c_{d-1}*x_{d-1} + c_d = 0 \f$
     */
-  inline Coefficients& coeffs() { return m_coeffs; }
+  EIGEN_DEVICE_FUNC inline Coefficients& coeffs() { return m_coeffs; }
 
   /** \returns the intersection of *this with \a other.
     *
@@ -186,16 +186,15 @@
     *
     * \note If \a other is approximately parallel to *this, this method will return any point on *this.
     */
-  VectorType intersection(const Hyperplane& other) const
+  EIGEN_DEVICE_FUNC VectorType intersection(const Hyperplane& other) const
   {
-    using std::abs;
     EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 2)
     Scalar det = coeffs().coeff(0) * other.coeffs().coeff(1) - coeffs().coeff(1) * other.coeffs().coeff(0);
     // since the line equations ax+by=c are normalized with a^2+b^2=1, the following tests
     // whether the two lines are approximately parallel.
     if(internal::isMuchSmallerThan(det, Scalar(1)))
     {   // special case where the two lines are approximately parallel. Pick any point on the first line.
-        if(abs(coeffs().coeff(1))>abs(coeffs().coeff(0)))
+        if(numext::abs(coeffs().coeff(1))>numext::abs(coeffs().coeff(0)))
             return VectorType(coeffs().coeff(1), -coeffs().coeff(2)/coeffs().coeff(1)-coeffs().coeff(0));
         else
             return VectorType(-coeffs().coeff(2)/coeffs().coeff(0)-coeffs().coeff(1), coeffs().coeff(0));
@@ -215,10 +214,13 @@
     *               or a more generic #Affine transformation. The default is #Affine.
     */
   template<typename XprType>
-  inline Hyperplane& transform(const MatrixBase<XprType>& mat, TransformTraits traits = Affine)
+  EIGEN_DEVICE_FUNC inline Hyperplane& transform(const MatrixBase<XprType>& mat, TransformTraits traits = Affine)
   {
     if (traits==Affine)
+    {
       normal() = mat.inverse().transpose() * normal();
+      m_coeffs /= normal().norm();
+    }
     else if (traits==Isometry)
       normal() = mat * normal();
     else
@@ -236,7 +238,7 @@
     *               Other kind of transformations are not supported.
     */
   template<int TrOptions>
-  inline Hyperplane& transform(const Transform<Scalar,AmbientDimAtCompileTime,Affine,TrOptions>& t,
+  EIGEN_DEVICE_FUNC inline Hyperplane& transform(const Transform<Scalar,AmbientDimAtCompileTime,Affine,TrOptions>& t,
                                 TransformTraits traits = Affine)
   {
     transform(t.linear(), traits);
@@ -250,7 +252,7 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Hyperplane,
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Hyperplane,
            Hyperplane<NewScalarType,AmbientDimAtCompileTime,Options> >::type cast() const
   {
     return typename internal::cast_return_type<Hyperplane,
@@ -259,7 +261,7 @@
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType,int OtherOptions>
-  inline explicit Hyperplane(const Hyperplane<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline explicit Hyperplane(const Hyperplane<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
   { m_coeffs = other.coeffs().template cast<Scalar>(); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
@@ -267,7 +269,7 @@
     *
     * \sa MatrixBase::isApprox() */
   template<int OtherOptions>
-  bool isApprox(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Hyperplane<Scalar,AmbientDimAtCompileTime,OtherOptions>& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_coeffs.isApprox(other.m_coeffs, prec); }
 
 protected:
diff --git a/Eigen/src/Geometry/OrthoMethods.h b/Eigen/src/Geometry/OrthoMethods.h
index 556bc81..a035e63 100644
--- a/Eigen/src/Geometry/OrthoMethods.h
+++ b/Eigen/src/Geometry/OrthoMethods.h
@@ -13,16 +13,24 @@
 
 namespace Eigen { 
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
   * \returns the cross product of \c *this and \a other
   *
   * Here is a very good explanation of cross-product: http://xkcd.com/199/
+  * 
+  * With complex numbers, the cross product is implemented as
+  * \f$ (\mathbf{a}+i\mathbf{b}) \times (\mathbf{c}+i\mathbf{d}) = (\mathbf{a} \times \mathbf{c} - \mathbf{b} \times \mathbf{d}) - i(\mathbf{a} \times \mathbf{d} - \mathbf{b} \times \mathbf{c})\f$
+  * 
   * \sa MatrixBase::cross3()
   */
 template<typename Derived>
 template<typename OtherDerived>
-inline typename MatrixBase<Derived>::template cross_product_return_type<OtherDerived>::type
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::template cross_product_return_type<OtherDerived>::type
+#else
+inline typename MatrixBase<Derived>::PlainObject
+#endif
 MatrixBase<Derived>::cross(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Derived,3)
@@ -30,8 +38,8 @@
 
   // Note that there is no need for an expression here since the compiler
   // optimize such a small temporary very well (even within a complex expression)
-  typename internal::nested<Derived,2>::type lhs(derived());
-  typename internal::nested<OtherDerived,2>::type rhs(other.derived());
+  typename internal::nested_eval<Derived,2>::type lhs(derived());
+  typename internal::nested_eval<OtherDerived,2>::type rhs(other.derived());
   return typename cross_product_return_type<OtherDerived>::type(
     numext::conj(lhs.coeff(1) * rhs.coeff(2) - lhs.coeff(2) * rhs.coeff(1)),
     numext::conj(lhs.coeff(2) * rhs.coeff(0) - lhs.coeff(0) * rhs.coeff(2)),
@@ -45,7 +53,7 @@
           typename Scalar = typename VectorLhs::Scalar,
           bool Vectorizable = bool((VectorLhs::Flags&VectorRhs::Flags)&PacketAccessBit)>
 struct cross3_impl {
-  static inline typename internal::plain_matrix_type<VectorLhs>::type
+  EIGEN_DEVICE_FUNC static inline typename internal::plain_matrix_type<VectorLhs>::type
   run(const VectorLhs& lhs, const VectorRhs& rhs)
   {
     return typename internal::plain_matrix_type<VectorLhs>::type(
@@ -59,7 +67,7 @@
 
 }
 
-/** \geometry_module
+/** \geometry_module \ingroup Geometry_Module
   *
   * \returns the cross product of \c *this and \a other using only the x, y, and z coefficients
   *
@@ -70,14 +78,14 @@
   */
 template<typename Derived>
 template<typename OtherDerived>
-inline typename MatrixBase<Derived>::PlainObject
+EIGEN_DEVICE_FUNC inline typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::cross3(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(Derived,4)
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,4)
 
-  typedef typename internal::nested<Derived,2>::type DerivedNested;
-  typedef typename internal::nested<OtherDerived,2>::type OtherDerivedNested;
+  typedef typename internal::nested_eval<Derived,2>::type DerivedNested;
+  typedef typename internal::nested_eval<OtherDerived,2>::type OtherDerivedNested;
   DerivedNested lhs(derived());
   OtherDerivedNested rhs(other.derived());
 
@@ -86,38 +94,42 @@
                         typename internal::remove_all<OtherDerivedNested>::type>::run(lhs,rhs);
 }
 
-/** \returns a matrix expression of the cross product of each column or row
+/** \geometry_module \ingroup Geometry_Module
+  *
+  * \returns a matrix expression of the cross product of each column or row
   * of the referenced expression with the \a other vector.
   *
   * The referenced matrix must have one dimension equal to 3.
   * The result matrix has the same dimensions than the referenced one.
   *
-  * \geometry_module
-  *
   * \sa MatrixBase::cross() */
 template<typename ExpressionType, int Direction>
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC 
 const typename VectorwiseOp<ExpressionType,Direction>::CrossReturnType
 VectorwiseOp<ExpressionType,Direction>::cross(const MatrixBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,3)
   EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
     YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
+  
+  typename internal::nested_eval<ExpressionType,2>::type mat(_expression());
+  typename internal::nested_eval<OtherDerived,2>::type vec(other.derived());
 
   CrossReturnType res(_expression().rows(),_expression().cols());
   if(Direction==Vertical)
   {
     eigen_assert(CrossReturnType::RowsAtCompileTime==3 && "the matrix must have exactly 3 rows");
-    res.row(0) = (_expression().row(1) * other.coeff(2) - _expression().row(2) * other.coeff(1)).conjugate();
-    res.row(1) = (_expression().row(2) * other.coeff(0) - _expression().row(0) * other.coeff(2)).conjugate();
-    res.row(2) = (_expression().row(0) * other.coeff(1) - _expression().row(1) * other.coeff(0)).conjugate();
+    res.row(0) = (mat.row(1) * vec.coeff(2) - mat.row(2) * vec.coeff(1)).conjugate();
+    res.row(1) = (mat.row(2) * vec.coeff(0) - mat.row(0) * vec.coeff(2)).conjugate();
+    res.row(2) = (mat.row(0) * vec.coeff(1) - mat.row(1) * vec.coeff(0)).conjugate();
   }
   else
   {
     eigen_assert(CrossReturnType::ColsAtCompileTime==3 && "the matrix must have exactly 3 columns");
-    res.col(0) = (_expression().col(1) * other.coeff(2) - _expression().col(2) * other.coeff(1)).conjugate();
-    res.col(1) = (_expression().col(2) * other.coeff(0) - _expression().col(0) * other.coeff(2)).conjugate();
-    res.col(2) = (_expression().col(0) * other.coeff(1) - _expression().col(1) * other.coeff(0)).conjugate();
+    res.col(0) = (mat.col(1) * vec.coeff(2) - mat.col(2) * vec.coeff(1)).conjugate();
+    res.col(1) = (mat.col(2) * vec.coeff(0) - mat.col(0) * vec.coeff(2)).conjugate();
+    res.col(2) = (mat.col(0) * vec.coeff(1) - mat.col(1) * vec.coeff(0)).conjugate();
   }
   return res;
 }
@@ -130,8 +142,8 @@
   typedef typename plain_matrix_type<Derived>::type VectorType;
   typedef typename traits<Derived>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef typename Derived::Index Index;
   typedef Matrix<Scalar,2,1> Vector2;
+  EIGEN_DEVICE_FUNC
   static inline VectorType run(const Derived& src)
   {
     VectorType perp = VectorType::Zero(src.size());
@@ -154,6 +166,7 @@
   typedef typename plain_matrix_type<Derived>::type VectorType;
   typedef typename traits<Derived>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
+  EIGEN_DEVICE_FUNC
   static inline VectorType run(const Derived& src)
   {
     VectorType perp;
@@ -192,13 +205,16 @@
 struct unitOrthogonal_selector<Derived,2>
 {
   typedef typename plain_matrix_type<Derived>::type VectorType;
+  EIGEN_DEVICE_FUNC
   static inline VectorType run(const Derived& src)
   { return VectorType(-numext::conj(src.y()), numext::conj(src.x())).normalized(); }
 };
 
 } // end namespace internal
 
-/** \returns a unit vector which is orthogonal to \c *this
+/** \geometry_module \ingroup Geometry_Module
+  *
+  * \returns a unit vector which is orthogonal to \c *this
   *
   * The size of \c *this must be at least 2. If the size is exactly 2,
   * then the returned vector is a counter clock wise rotation of \c *this, i.e., (-y,x).normalized().
@@ -206,7 +222,7 @@
   * \sa cross()
   */
 template<typename Derived>
-typename MatrixBase<Derived>::PlainObject
+EIGEN_DEVICE_FUNC typename MatrixBase<Derived>::PlainObject
 MatrixBase<Derived>::unitOrthogonal() const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
diff --git a/Eigen/src/Geometry/ParametrizedLine.h b/Eigen/src/Geometry/ParametrizedLine.h
index 77fa228..1e985d8 100644
--- a/Eigen/src/Geometry/ParametrizedLine.h
+++ b/Eigen/src/Geometry/ParametrizedLine.h
@@ -23,8 +23,8 @@
   * direction vector \f$ \mathbf{d} \f$ such that the line corresponds to
   * the set \f$ l(t) = \mathbf{o} + t \mathbf{d} \f$, \f$ t \in \mathbf{R} \f$.
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
-  * \param _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients
+  * \tparam _AmbientDim the dimension of the ambient space, can be a compile time value or Dynamic.
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 class ParametrizedLine
@@ -37,49 +37,49 @@
   };
   typedef _Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef DenseIndex Index;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
   typedef Matrix<Scalar,AmbientDimAtCompileTime,1,Options> VectorType;
 
   /** Default constructor without initialization */
-  inline ParametrizedLine() {}
+  EIGEN_DEVICE_FUNC inline ParametrizedLine() {}
   
   template<int OtherOptions>
-  ParametrizedLine(const ParametrizedLine<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC ParametrizedLine(const ParametrizedLine<Scalar,AmbientDimAtCompileTime,OtherOptions>& other)
    : m_origin(other.origin()), m_direction(other.direction())
   {}
 
   /** Constructs a dynamic-size line with \a _dim the dimension
     * of the ambient space */
-  inline explicit ParametrizedLine(Index _dim) : m_origin(_dim), m_direction(_dim) {}
+  EIGEN_DEVICE_FUNC inline explicit ParametrizedLine(Index _dim) : m_origin(_dim), m_direction(_dim) {}
 
   /** Initializes a parametrized line of direction \a direction and origin \a origin.
     * \warning the vector direction is assumed to be normalized.
     */
-  ParametrizedLine(const VectorType& origin, const VectorType& direction)
+  EIGEN_DEVICE_FUNC ParametrizedLine(const VectorType& origin, const VectorType& direction)
     : m_origin(origin), m_direction(direction) {}
 
   template <int OtherOptions>
-  explicit ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane);
+  EIGEN_DEVICE_FUNC explicit ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane);
 
   /** Constructs a parametrized line going from \a p0 to \a p1. */
-  static inline ParametrizedLine Through(const VectorType& p0, const VectorType& p1)
+  EIGEN_DEVICE_FUNC static inline ParametrizedLine Through(const VectorType& p0, const VectorType& p1)
   { return ParametrizedLine(p0, (p1-p0).normalized()); }
 
-  ~ParametrizedLine() {}
+  EIGEN_DEVICE_FUNC ~ParametrizedLine() {}
 
   /** \returns the dimension in which the line holds */
-  inline Index dim() const { return m_direction.size(); }
+  EIGEN_DEVICE_FUNC inline Index dim() const { return m_direction.size(); }
 
-  const VectorType& origin() const { return m_origin; }
-  VectorType& origin() { return m_origin; }
+  EIGEN_DEVICE_FUNC const VectorType& origin() const { return m_origin; }
+  EIGEN_DEVICE_FUNC VectorType& origin() { return m_origin; }
 
-  const VectorType& direction() const { return m_direction; }
-  VectorType& direction() { return m_direction; }
+  EIGEN_DEVICE_FUNC const VectorType& direction() const { return m_direction; }
+  EIGEN_DEVICE_FUNC VectorType& direction() { return m_direction; }
 
   /** \returns the squared distance of a point \a p to its projection onto the line \c *this.
     * \sa distance()
     */
-  RealScalar squaredDistance(const VectorType& p) const
+  EIGEN_DEVICE_FUNC RealScalar squaredDistance(const VectorType& p) const
   {
     VectorType diff = p - origin();
     return (diff - direction().dot(diff) * direction()).squaredNorm();
@@ -87,22 +87,22 @@
   /** \returns the distance of a point \a p to its projection onto the line \c *this.
     * \sa squaredDistance()
     */
-  RealScalar distance(const VectorType& p) const { using std::sqrt; return sqrt(squaredDistance(p)); }
+  EIGEN_DEVICE_FUNC RealScalar distance(const VectorType& p) const { EIGEN_USING_STD_MATH(sqrt) return sqrt(squaredDistance(p)); }
 
   /** \returns the projection of a point \a p onto the line \c *this. */
-  VectorType projection(const VectorType& p) const
+  EIGEN_DEVICE_FUNC VectorType projection(const VectorType& p) const
   { return origin() + direction().dot(p-origin()) * direction(); }
 
-  VectorType pointAt(const Scalar& t) const;
+  EIGEN_DEVICE_FUNC VectorType pointAt(const Scalar& t) const;
   
   template <int OtherOptions>
-  Scalar intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
+  EIGEN_DEVICE_FUNC Scalar intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
  
   template <int OtherOptions>
-  Scalar intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
+  EIGEN_DEVICE_FUNC Scalar intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
   
   template <int OtherOptions>
-  VectorType intersectionPoint(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
+  EIGEN_DEVICE_FUNC VectorType intersectionPoint(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const;
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -110,7 +110,7 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<ParametrizedLine,
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<ParametrizedLine,
            ParametrizedLine<NewScalarType,AmbientDimAtCompileTime,Options> >::type cast() const
   {
     return typename internal::cast_return_type<ParametrizedLine,
@@ -119,7 +119,7 @@
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType,int OtherOptions>
-  inline explicit ParametrizedLine(const ParametrizedLine<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline explicit ParametrizedLine(const ParametrizedLine<OtherScalarType,AmbientDimAtCompileTime,OtherOptions>& other)
   {
     m_origin = other.origin().template cast<Scalar>();
     m_direction = other.direction().template cast<Scalar>();
@@ -129,7 +129,7 @@
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const ParametrizedLine& other, typename NumTraits<Scalar>::Real prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const ParametrizedLine& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_origin.isApprox(other.m_origin, prec) && m_direction.isApprox(other.m_direction, prec); }
 
 protected:
@@ -143,7 +143,7 @@
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline ParametrizedLine<_Scalar, _AmbientDim,_Options>::ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim,OtherOptions>& hyperplane)
+EIGEN_DEVICE_FUNC inline ParametrizedLine<_Scalar, _AmbientDim,_Options>::ParametrizedLine(const Hyperplane<_Scalar, _AmbientDim,OtherOptions>& hyperplane)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(VectorType, 2)
   direction() = hyperplane.normal().unitOrthogonal();
@@ -153,7 +153,7 @@
 /** \returns the point at \a t along this line
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
-inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
+EIGEN_DEVICE_FUNC inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
 ParametrizedLine<_Scalar, _AmbientDim,_Options>::pointAt(const _Scalar& t) const
 {
   return origin() + (direction()*t); 
@@ -163,7 +163,7 @@
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
+EIGEN_DEVICE_FUNC inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersectionParameter(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
 {
   return -(hyperplane.offset()+hyperplane.normal().dot(origin()))
           / hyperplane.normal().dot(direction());
@@ -175,7 +175,7 @@
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
+EIGEN_DEVICE_FUNC inline _Scalar ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersection(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
 {
   return intersectionParameter(hyperplane);
 }
@@ -184,7 +184,7 @@
   */
 template <typename _Scalar, int _AmbientDim, int _Options>
 template <int OtherOptions>
-inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
+EIGEN_DEVICE_FUNC inline typename ParametrizedLine<_Scalar, _AmbientDim,_Options>::VectorType
 ParametrizedLine<_Scalar, _AmbientDim,_Options>::intersectionPoint(const Hyperplane<_Scalar, _AmbientDim, OtherOptions>& hyperplane) const
 {
   return pointAt(intersectionParameter(hyperplane));
diff --git a/Eigen/src/Geometry/Quaternion.h b/Eigen/src/Geometry/Quaternion.h
index 93056f6..c3fd8c3 100644
--- a/Eigen/src/Geometry/Quaternion.h
+++ b/Eigen/src/Geometry/Quaternion.h
@@ -34,14 +34,20 @@
 template<class Derived>
 class QuaternionBase : public RotationBase<Derived, 3>
 {
+ public:
   typedef RotationBase<Derived, 3> Base;
-public:
+
   using Base::operator*;
   using Base::derived;
 
   typedef typename internal::traits<Derived>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
   typedef typename internal::traits<Derived>::Coefficients Coefficients;
+  typedef typename Coefficients::CoeffReturnType CoeffReturnType;
+  typedef typename internal::conditional<bool(internal::traits<Derived>::Flags&LvalueBit),
+                                        Scalar&, CoeffReturnType>::type NonConstCoeffReturnType;
+
+
   enum {
     Flags = Eigen::internal::traits<Derived>::Flags
   };
@@ -57,37 +63,37 @@
 
 
   /** \returns the \c x coefficient */
-  inline Scalar x() const { return this->derived().coeffs().coeff(0); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType x() const { return this->derived().coeffs().coeff(0); }
   /** \returns the \c y coefficient */
-  inline Scalar y() const { return this->derived().coeffs().coeff(1); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType y() const { return this->derived().coeffs().coeff(1); }
   /** \returns the \c z coefficient */
-  inline Scalar z() const { return this->derived().coeffs().coeff(2); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType z() const { return this->derived().coeffs().coeff(2); }
   /** \returns the \c w coefficient */
-  inline Scalar w() const { return this->derived().coeffs().coeff(3); }
+  EIGEN_DEVICE_FUNC inline CoeffReturnType w() const { return this->derived().coeffs().coeff(3); }
 
-  /** \returns a reference to the \c x coefficient */
-  inline Scalar& x() { return this->derived().coeffs().coeffRef(0); }
-  /** \returns a reference to the \c y coefficient */
-  inline Scalar& y() { return this->derived().coeffs().coeffRef(1); }
-  /** \returns a reference to the \c z coefficient */
-  inline Scalar& z() { return this->derived().coeffs().coeffRef(2); }
-  /** \returns a reference to the \c w coefficient */
-  inline Scalar& w() { return this->derived().coeffs().coeffRef(3); }
+  /** \returns a reference to the \c x coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType x() { return this->derived().coeffs().x(); }
+  /** \returns a reference to the \c y coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType y() { return this->derived().coeffs().y(); }
+  /** \returns a reference to the \c z coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType z() { return this->derived().coeffs().z(); }
+  /** \returns a reference to the \c w coefficient (if Derived is a non-const lvalue) */
+  EIGEN_DEVICE_FUNC inline NonConstCoeffReturnType w() { return this->derived().coeffs().w(); }
 
   /** \returns a read-only vector expression of the imaginary part (x,y,z) */
-  inline const VectorBlock<const Coefficients,3> vec() const { return coeffs().template head<3>(); }
+  EIGEN_DEVICE_FUNC inline const VectorBlock<const Coefficients,3> vec() const { return coeffs().template head<3>(); }
 
   /** \returns a vector expression of the imaginary part (x,y,z) */
-  inline VectorBlock<Coefficients,3> vec() { return coeffs().template head<3>(); }
+  EIGEN_DEVICE_FUNC inline VectorBlock<Coefficients,3> vec() { return coeffs().template head<3>(); }
 
   /** \returns a read-only vector expression of the coefficients (x,y,z,w) */
-  inline const typename internal::traits<Derived>::Coefficients& coeffs() const { return derived().coeffs(); }
+  EIGEN_DEVICE_FUNC inline const typename internal::traits<Derived>::Coefficients& coeffs() const { return derived().coeffs(); }
 
   /** \returns a vector expression of the coefficients (x,y,z,w) */
-  inline typename internal::traits<Derived>::Coefficients& coeffs() { return derived().coeffs(); }
+  EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Coefficients& coeffs() { return derived().coeffs(); }
 
-  EIGEN_STRONG_INLINE QuaternionBase<Derived>& operator=(const QuaternionBase<Derived>& other);
-  template<class OtherDerived> EIGEN_STRONG_INLINE Derived& operator=(const QuaternionBase<OtherDerived>& other);
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE QuaternionBase<Derived>& operator=(const QuaternionBase<Derived>& other);
+  template<class OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const QuaternionBase<OtherDerived>& other);
 
 // disabled this copy operator as it is giving very strange compilation errors when compiling
 // test_stdvector with GCC 4.4.2. This looks like a GCC bug though, so feel free to re-enable it if it's
@@ -96,72 +102,72 @@
 //  Derived& operator=(const QuaternionBase& other)
 //  { return operator=<Derived>(other); }
 
-  Derived& operator=(const AngleAxisType& aa);
-  template<class OtherDerived> Derived& operator=(const MatrixBase<OtherDerived>& m);
+  EIGEN_DEVICE_FUNC Derived& operator=(const AngleAxisType& aa);
+  template<class OtherDerived> EIGEN_DEVICE_FUNC Derived& operator=(const MatrixBase<OtherDerived>& m);
 
   /** \returns a quaternion representing an identity rotation
     * \sa MatrixBase::Identity()
     */
-  static inline Quaternion<Scalar> Identity() { return Quaternion<Scalar>(1, 0, 0, 0); }
+  EIGEN_DEVICE_FUNC static inline Quaternion<Scalar> Identity() { return Quaternion<Scalar>(Scalar(1), Scalar(0), Scalar(0), Scalar(0)); }
 
   /** \sa QuaternionBase::Identity(), MatrixBase::setIdentity()
     */
-  inline QuaternionBase& setIdentity() { coeffs() << 0, 0, 0, 1; return *this; }
+  EIGEN_DEVICE_FUNC inline QuaternionBase& setIdentity() { coeffs() << Scalar(0), Scalar(0), Scalar(0), Scalar(1); return *this; }
 
   /** \returns the squared norm of the quaternion's coefficients
     * \sa QuaternionBase::norm(), MatrixBase::squaredNorm()
     */
-  inline Scalar squaredNorm() const { return coeffs().squaredNorm(); }
+  EIGEN_DEVICE_FUNC inline Scalar squaredNorm() const { return coeffs().squaredNorm(); }
 
   /** \returns the norm of the quaternion's coefficients
     * \sa QuaternionBase::squaredNorm(), MatrixBase::norm()
     */
-  inline Scalar norm() const { return coeffs().norm(); }
+  EIGEN_DEVICE_FUNC inline Scalar norm() const { return coeffs().norm(); }
 
   /** Normalizes the quaternion \c *this
     * \sa normalized(), MatrixBase::normalize() */
-  inline void normalize() { coeffs().normalize(); }
+  EIGEN_DEVICE_FUNC inline void normalize() { coeffs().normalize(); }
   /** \returns a normalized copy of \c *this
     * \sa normalize(), MatrixBase::normalized() */
-  inline Quaternion<Scalar> normalized() const { return Quaternion<Scalar>(coeffs().normalized()); }
+  EIGEN_DEVICE_FUNC inline Quaternion<Scalar> normalized() const { return Quaternion<Scalar>(coeffs().normalized()); }
 
     /** \returns the dot product of \c *this and \a other
     * Geometrically speaking, the dot product of two unit quaternions
     * corresponds to the cosine of half the angle between the two rotations.
     * \sa angularDistance()
     */
-  template<class OtherDerived> inline Scalar dot(const QuaternionBase<OtherDerived>& other) const { return coeffs().dot(other.coeffs()); }
+  template<class OtherDerived> EIGEN_DEVICE_FUNC inline Scalar dot(const QuaternionBase<OtherDerived>& other) const { return coeffs().dot(other.coeffs()); }
 
-  template<class OtherDerived> Scalar angularDistance(const QuaternionBase<OtherDerived>& other) const;
+  template<class OtherDerived> EIGEN_DEVICE_FUNC Scalar angularDistance(const QuaternionBase<OtherDerived>& other) const;
 
   /** \returns an equivalent 3x3 rotation matrix */
-  Matrix3 toRotationMatrix() const;
+  EIGEN_DEVICE_FUNC Matrix3 toRotationMatrix() const;
 
   /** \returns the quaternion which transform \a a into \a b through a rotation */
   template<typename Derived1, typename Derived2>
-  Derived& setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
+  EIGEN_DEVICE_FUNC Derived& setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
 
-  template<class OtherDerived> EIGEN_STRONG_INLINE Quaternion<Scalar> operator* (const QuaternionBase<OtherDerived>& q) const;
-  template<class OtherDerived> EIGEN_STRONG_INLINE Derived& operator*= (const QuaternionBase<OtherDerived>& q);
+  template<class OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Quaternion<Scalar> operator* (const QuaternionBase<OtherDerived>& q) const;
+  template<class OtherDerived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator*= (const QuaternionBase<OtherDerived>& q);
 
   /** \returns the quaternion describing the inverse rotation */
-  Quaternion<Scalar> inverse() const;
+  EIGEN_DEVICE_FUNC Quaternion<Scalar> inverse() const;
 
   /** \returns the conjugated quaternion */
-  Quaternion<Scalar> conjugate() const;
+  EIGEN_DEVICE_FUNC Quaternion<Scalar> conjugate() const;
 
-  template<class OtherDerived> Quaternion<Scalar> slerp(const Scalar& t, const QuaternionBase<OtherDerived>& other) const;
+  template<class OtherDerived> EIGEN_DEVICE_FUNC Quaternion<Scalar> slerp(const Scalar& t, const QuaternionBase<OtherDerived>& other) const;
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
   template<class OtherDerived>
-  bool isApprox(const QuaternionBase<OtherDerived>& other, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const QuaternionBase<OtherDerived>& other, const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const
   { return coeffs().isApprox(other.coeffs(), prec); }
 
-	/** return the result vector of \a v through the rotation*/
-  EIGEN_STRONG_INLINE Vector3 _transformVector(const Vector3& v) const;
+  /** return the result vector of \a v through the rotation*/
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Vector3 _transformVector(const Vector3& v) const;
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -169,7 +175,7 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type cast() const
   {
     return typename internal::cast_return_type<Derived,Quaternion<NewScalarType> >::type(derived());
   }
@@ -216,8 +222,8 @@
   typedef _Scalar Scalar;
   typedef Matrix<_Scalar,4,1,_Options> Coefficients;
   enum{
-    IsAligned = internal::traits<Coefficients>::Flags & AlignedBit,
-    Flags = IsAligned ? (AlignedBit | LvalueBit) : LvalueBit
+    Alignment = internal::traits<Coefficients>::Alignment,
+    Flags = LvalueBit
   };
 };
 }
@@ -225,10 +231,10 @@
 template<typename _Scalar, int _Options>
 class Quaternion : public QuaternionBase<Quaternion<_Scalar,_Options> >
 {
-  typedef QuaternionBase<Quaternion<_Scalar,_Options> > Base;
-  enum { IsAligned = internal::traits<Quaternion>::IsAligned };
-
 public:
+  typedef QuaternionBase<Quaternion<_Scalar,_Options> > Base;
+  enum { NeedsAlignment = internal::traits<Quaternion>::Alignment>0 };
+
   typedef _Scalar Scalar;
 
   EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Quaternion)
@@ -238,7 +244,7 @@
   typedef typename Base::AngleAxisType AngleAxisType;
 
   /** Default constructor leaving the quaternion uninitialized. */
-  inline Quaternion() {}
+  EIGEN_DEVICE_FUNC inline Quaternion() {}
 
   /** Constructs and initializes the quaternion \f$ w+xi+yj+zk \f$ from
     * its four coefficients \a w, \a x, \a y and \a z.
@@ -247,36 +253,42 @@
     * while internally the coefficients are stored in the following order:
     * [\c x, \c y, \c z, \c w]
     */
-  inline Quaternion(const Scalar& w, const Scalar& x, const Scalar& y, const Scalar& z) : m_coeffs(x, y, z, w){}
+  EIGEN_DEVICE_FUNC inline Quaternion(const Scalar& w, const Scalar& x, const Scalar& y, const Scalar& z) : m_coeffs(x, y, z, w){}
 
   /** Constructs and initialize a quaternion from the array data */
-  inline Quaternion(const Scalar* data) : m_coeffs(data) {}
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const Scalar* data) : m_coeffs(data) {}
 
   /** Copy constructor */
-  template<class Derived> EIGEN_STRONG_INLINE Quaternion(const QuaternionBase<Derived>& other) { this->Base::operator=(other); }
+  template<class Derived> EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Quaternion(const QuaternionBase<Derived>& other) { this->Base::operator=(other); }
 
   /** Constructs and initializes a quaternion from the angle-axis \a aa */
-  explicit inline Quaternion(const AngleAxisType& aa) { *this = aa; }
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const AngleAxisType& aa) { *this = aa; }
 
   /** Constructs and initializes a quaternion from either:
     *  - a rotation matrix expression,
     *  - a 4D vector expression representing quaternion coefficients.
     */
   template<typename Derived>
-  explicit inline Quaternion(const MatrixBase<Derived>& other) { *this = other; }
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const MatrixBase<Derived>& other) { *this = other; }
 
   /** Explicit copy constructor with scalar conversion */
   template<typename OtherScalar, int OtherOptions>
-  explicit inline Quaternion(const Quaternion<OtherScalar, OtherOptions>& other)
+  EIGEN_DEVICE_FUNC explicit inline Quaternion(const Quaternion<OtherScalar, OtherOptions>& other)
   { m_coeffs = other.coeffs().template cast<Scalar>(); }
 
+  EIGEN_DEVICE_FUNC static Quaternion UnitRandom();
+
   template<typename Derived1, typename Derived2>
-  static Quaternion FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
+  EIGEN_DEVICE_FUNC static Quaternion FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b);
 
-  inline Coefficients& coeffs() { return m_coeffs;}
-  inline const Coefficients& coeffs() const { return m_coeffs;}
+  EIGEN_DEVICE_FUNC inline Coefficients& coeffs() { return m_coeffs;}
+  EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs;}
 
-  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(IsAligned)
+  EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(bool(NeedsAlignment))
+  
+#ifdef EIGEN_QUATERNION_PLUGIN
+# include EIGEN_QUATERNION_PLUGIN
+#endif
 
 protected:
   Coefficients m_coeffs;
@@ -336,9 +348,9 @@
 class Map<const Quaternion<_Scalar>, _Options >
   : public QuaternionBase<Map<const Quaternion<_Scalar>, _Options> >
 {
+  public:
     typedef QuaternionBase<Map<const Quaternion<_Scalar>, _Options> > Base;
 
-  public:
     typedef _Scalar Scalar;
     typedef typename internal::traits<Map>::Coefficients Coefficients;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Map)
@@ -350,9 +362,9 @@
       * \code *coeffs == {x, y, z, w} \endcode
       *
       * If the template parameter _Options is set to #Aligned, then the pointer coeffs must be aligned. */
-    EIGEN_STRONG_INLINE Map(const Scalar* coeffs) : m_coeffs(coeffs) {}
+    EIGEN_DEVICE_FUNC explicit EIGEN_STRONG_INLINE Map(const Scalar* coeffs) : m_coeffs(coeffs) {}
 
-    inline const Coefficients& coeffs() const { return m_coeffs;}
+    EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs;}
 
   protected:
     const Coefficients m_coeffs;
@@ -373,9 +385,9 @@
 class Map<Quaternion<_Scalar>, _Options >
   : public QuaternionBase<Map<Quaternion<_Scalar>, _Options> >
 {
+  public:
     typedef QuaternionBase<Map<Quaternion<_Scalar>, _Options> > Base;
 
-  public:
     typedef _Scalar Scalar;
     typedef typename internal::traits<Map>::Coefficients Coefficients;
     EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Map)
@@ -387,10 +399,10 @@
       * \code *coeffs == {x, y, z, w} \endcode
       *
       * If the template parameter _Options is set to #Aligned, then the pointer coeffs must be aligned. */
-    EIGEN_STRONG_INLINE Map(Scalar* coeffs) : m_coeffs(coeffs) {}
+    EIGEN_DEVICE_FUNC explicit EIGEN_STRONG_INLINE Map(Scalar* coeffs) : m_coeffs(coeffs) {}
 
-    inline Coefficients& coeffs() { return m_coeffs; }
-    inline const Coefficients& coeffs() const { return m_coeffs; }
+    EIGEN_DEVICE_FUNC inline Coefficients& coeffs() { return m_coeffs; }
+    EIGEN_DEVICE_FUNC inline const Coefficients& coeffs() const { return m_coeffs; }
 
   protected:
     Coefficients m_coeffs;
@@ -416,9 +428,9 @@
 // Generic Quaternion * Quaternion product
 // This product can be specialized for a given architecture via the Arch template argument.
 namespace internal {
-template<int Arch, class Derived1, class Derived2, typename Scalar, int _Options> struct quat_product
+template<int Arch, class Derived1, class Derived2, typename Scalar> struct quat_product
 {
-  static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived1>& a, const QuaternionBase<Derived2>& b){
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived1>& a, const QuaternionBase<Derived2>& b){
     return Quaternion<Scalar>
     (
       a.w() * b.w() - a.x() * b.x() - a.y() * b.y() - a.z() * b.z(),
@@ -433,20 +445,19 @@
 /** \returns the concatenation of two rotations as a quaternion-quaternion product */
 template <class Derived>
 template <class OtherDerived>
-EIGEN_STRONG_INLINE Quaternion<typename internal::traits<Derived>::Scalar>
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Quaternion<typename internal::traits<Derived>::Scalar>
 QuaternionBase<Derived>::operator* (const QuaternionBase<OtherDerived>& other) const
 {
   EIGEN_STATIC_ASSERT((internal::is_same<typename Derived::Scalar, typename OtherDerived::Scalar>::value),
    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
   return internal::quat_product<Architecture::Target, Derived, OtherDerived,
-                         typename internal::traits<Derived>::Scalar,
-                         internal::traits<Derived>::IsAligned && internal::traits<OtherDerived>::IsAligned>::run(*this, other);
+                         typename internal::traits<Derived>::Scalar>::run(*this, other);
 }
 
 /** \sa operator*(Quaternion) */
 template <class Derived>
 template <class OtherDerived>
-EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator*= (const QuaternionBase<OtherDerived>& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator*= (const QuaternionBase<OtherDerived>& other)
 {
   derived() = derived() * other.derived();
   return derived();
@@ -460,7 +471,7 @@
   *   - Via a Matrix3: 24 + 15n
   */
 template <class Derived>
-EIGEN_STRONG_INLINE typename QuaternionBase<Derived>::Vector3
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename QuaternionBase<Derived>::Vector3
 QuaternionBase<Derived>::_transformVector(const Vector3& v) const
 {
     // Note that this algorithm comes from the optimization by hand
@@ -474,7 +485,7 @@
 }
 
 template<class Derived>
-EIGEN_STRONG_INLINE QuaternionBase<Derived>& QuaternionBase<Derived>::operator=(const QuaternionBase<Derived>& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE QuaternionBase<Derived>& QuaternionBase<Derived>::operator=(const QuaternionBase<Derived>& other)
 {
   coeffs() = other.coeffs();
   return derived();
@@ -482,7 +493,7 @@
 
 template<class Derived>
 template<class OtherDerived>
-EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const QuaternionBase<OtherDerived>& other)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const QuaternionBase<OtherDerived>& other)
 {
   coeffs() = other.coeffs();
   return derived();
@@ -491,10 +502,10 @@
 /** Set \c *this from an angle-axis \a aa and returns a reference to \c *this
   */
 template<class Derived>
-EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const AngleAxisType& aa)
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& QuaternionBase<Derived>::operator=(const AngleAxisType& aa)
 {
-  using std::cos;
-  using std::sin;
+  EIGEN_USING_STD_MATH(cos)
+  EIGEN_USING_STD_MATH(sin)
   Scalar ha = Scalar(0.5)*aa.angle(); // Scalar(0.5) to suppress precision loss warnings
   this->w() = cos(ha);
   this->vec() = sin(ha) * aa.axis();
@@ -509,7 +520,7 @@
 
 template<class Derived>
 template<class MatrixDerived>
-inline Derived& QuaternionBase<Derived>::operator=(const MatrixBase<MatrixDerived>& xpr)
+EIGEN_DEVICE_FUNC inline Derived& QuaternionBase<Derived>::operator=(const MatrixBase<MatrixDerived>& xpr)
 {
   EIGEN_STATIC_ASSERT((internal::is_same<typename Derived::Scalar, typename MatrixDerived::Scalar>::value),
    YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
@@ -521,7 +532,7 @@
   * be normalized, otherwise the result is undefined.
   */
 template<class Derived>
-inline typename QuaternionBase<Derived>::Matrix3
+EIGEN_DEVICE_FUNC inline typename QuaternionBase<Derived>::Matrix3
 QuaternionBase<Derived>::toRotationMatrix(void) const
 {
   // NOTE if inlined, then gcc 4.2 and 4.4 get rid of the temporary (not gcc 4.3 !!)
@@ -568,10 +579,9 @@
   */
 template<class Derived>
 template<typename Derived1, typename Derived2>
-inline Derived& QuaternionBase<Derived>::setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
+EIGEN_DEVICE_FUNC inline Derived& QuaternionBase<Derived>::setFromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
 {
-  using std::max;
-  using std::sqrt;
+  EIGEN_USING_STD_MATH(sqrt)
   Vector3 v0 = a.normalized();
   Vector3 v1 = b.normalized();
   Scalar c = v1.dot(v0);
@@ -586,7 +596,7 @@
   //    which yields a singular value problem
   if (c < Scalar(-1)+NumTraits<Scalar>::dummy_precision())
   {
-    c = (max)(c,Scalar(-1));
+    c = numext::maxi(c,Scalar(-1));
     Matrix<Scalar,2,3> m; m << v0.transpose(), v1.transpose();
     JacobiSVD<Matrix<Scalar,2,3> > svd(m, ComputeFullV);
     Vector3 axis = svd.matrixV().col(2);
@@ -605,6 +615,24 @@
   return derived();
 }
 
+/** \returns a random unit quaternion following a uniform distribution law on SO(3)
+  *
+  * \note The implementation is based on http://planning.cs.uiuc.edu/node198.html
+  */
+template<typename Scalar, int Options>
+EIGEN_DEVICE_FUNC Quaternion<Scalar,Options> Quaternion<Scalar,Options>::UnitRandom()
+{
+  EIGEN_USING_STD_MATH(sqrt)
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
+  const Scalar u1 = internal::random<Scalar>(0, 1),
+               u2 = internal::random<Scalar>(0, 2*EIGEN_PI),
+               u3 = internal::random<Scalar>(0, 2*EIGEN_PI);
+  const Scalar a = sqrt(1 - u1),
+               b = sqrt(u1);
+  return Quaternion (a * sin(u2), a * cos(u2), b * sin(u3), b * cos(u3));
+}
+
 
 /** Returns a quaternion representing a rotation between
   * the two arbitrary vectors \a a and \a b. In other words, the built
@@ -618,7 +646,7 @@
   */
 template<typename Scalar, int Options>
 template<typename Derived1, typename Derived2>
-Quaternion<Scalar,Options> Quaternion<Scalar,Options>::FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
+EIGEN_DEVICE_FUNC Quaternion<Scalar,Options> Quaternion<Scalar,Options>::FromTwoVectors(const MatrixBase<Derived1>& a, const MatrixBase<Derived2>& b)
 {
     Quaternion quat;
     quat.setFromTwoVectors(a, b);
@@ -633,7 +661,7 @@
   * \sa QuaternionBase::conjugate()
   */
 template <class Derived>
-inline Quaternion<typename internal::traits<Derived>::Scalar> QuaternionBase<Derived>::inverse() const
+EIGEN_DEVICE_FUNC inline Quaternion<typename internal::traits<Derived>::Scalar> QuaternionBase<Derived>::inverse() const
 {
   // FIXME should this function be called multiplicativeInverse and conjugate() be called inverse() or opposite()  ??
   Scalar n2 = this->squaredNorm();
@@ -646,6 +674,16 @@
   }
 }
 
+// Generic conjugate of a Quaternion
+namespace internal {
+template<int Arch, class Derived, typename Scalar> struct quat_conj
+{
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Quaternion<Scalar> run(const QuaternionBase<Derived>& q){
+    return Quaternion<Scalar>(q.w(),-q.x(),-q.y(),-q.z());
+  }
+};
+}
+                         
 /** \returns the conjugate of the \c *this which is equal to the multiplicative inverse
   * if the quaternion is normalized.
   * The conjugate of a quaternion represents the opposite rotation.
@@ -653,10 +691,12 @@
   * \sa Quaternion2::inverse()
   */
 template <class Derived>
-inline Quaternion<typename internal::traits<Derived>::Scalar>
+EIGEN_DEVICE_FUNC inline Quaternion<typename internal::traits<Derived>::Scalar>
 QuaternionBase<Derived>::conjugate() const
 {
-  return Quaternion<Scalar>(this->w(),-this->x(),-this->y(),-this->z());
+  return internal::quat_conj<Architecture::Target, Derived,
+                         typename internal::traits<Derived>::Scalar>::run(*this);
+                         
 }
 
 /** \returns the angle (in radian) between two rotations
@@ -664,13 +704,12 @@
   */
 template <class Derived>
 template <class OtherDerived>
-inline typename internal::traits<Derived>::Scalar
+EIGEN_DEVICE_FUNC inline typename internal::traits<Derived>::Scalar
 QuaternionBase<Derived>::angularDistance(const QuaternionBase<OtherDerived>& other) const
 {
-  using std::atan2;
-  using std::abs;
+  EIGEN_USING_STD_MATH(atan2)
   Quaternion<Scalar> d = (*this) * other.conjugate();
-  return Scalar(2) * atan2( d.vec().norm(), abs(d.w()) );
+  return Scalar(2) * atan2( d.vec().norm(), numext::abs(d.w()) );
 }
 
  
@@ -683,15 +722,14 @@
   */
 template <class Derived>
 template <class OtherDerived>
-Quaternion<typename internal::traits<Derived>::Scalar>
+EIGEN_DEVICE_FUNC Quaternion<typename internal::traits<Derived>::Scalar>
 QuaternionBase<Derived>::slerp(const Scalar& t, const QuaternionBase<OtherDerived>& other) const
 {
-  using std::acos;
-  using std::sin;
-  using std::abs;
-  static const Scalar one = Scalar(1) - NumTraits<Scalar>::epsilon();
+  EIGEN_USING_STD_MATH(acos)
+  EIGEN_USING_STD_MATH(sin)
+  const Scalar one = Scalar(1) - NumTraits<Scalar>::epsilon();
   Scalar d = this->dot(other);
-  Scalar absD = abs(d);
+  Scalar absD = numext::abs(d);
 
   Scalar scale0;
   Scalar scale1;
@@ -722,10 +760,10 @@
 struct quaternionbase_assign_impl<Other,3,3>
 {
   typedef typename Other::Scalar Scalar;
-  typedef DenseIndex Index;
-  template<class Derived> static inline void run(QuaternionBase<Derived>& q, const Other& mat)
+  template<class Derived> EIGEN_DEVICE_FUNC static inline void run(QuaternionBase<Derived>& q, const Other& a_mat)
   {
-    using std::sqrt;
+    const typename internal::nested_eval<Other,2>::type mat(a_mat);
+    EIGEN_USING_STD_MATH(sqrt)
     // This algorithm comes from  "Quaternion Calculus and Fast Animation",
     // Ken Shoemake, 1987 SIGGRAPH course notes
     Scalar t = mat.trace();
@@ -740,13 +778,13 @@
     }
     else
     {
-      DenseIndex i = 0;
+      Index i = 0;
       if (mat.coeff(1,1) > mat.coeff(0,0))
         i = 1;
       if (mat.coeff(2,2) > mat.coeff(i,i))
         i = 2;
-      DenseIndex j = (i+1)%3;
-      DenseIndex k = (j+1)%3;
+      Index j = (i+1)%3;
+      Index k = (j+1)%3;
 
       t = sqrt(mat.coeff(i,i)-mat.coeff(j,j)-mat.coeff(k,k) + Scalar(1.0));
       q.coeffs().coeffRef(i) = Scalar(0.5) * t;
@@ -763,7 +801,7 @@
 struct quaternionbase_assign_impl<Other,4,1>
 {
   typedef typename Other::Scalar Scalar;
-  template<class Derived> static inline void run(QuaternionBase<Derived>& q, const Other& vec)
+  template<class Derived> EIGEN_DEVICE_FUNC static inline void run(QuaternionBase<Derived>& q, const Other& vec)
   {
     q.coeffs() = vec;
   }
diff --git a/Eigen/src/Geometry/Rotation2D.h b/Eigen/src/Geometry/Rotation2D.h
index a2d59fc..884b7d0 100644
--- a/Eigen/src/Geometry/Rotation2D.h
+++ b/Eigen/src/Geometry/Rotation2D.h
@@ -18,7 +18,7 @@
   *
   * \brief Represents a rotation/orientation in a 2 dimensional space.
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients
   *
   * This class is equivalent to a single scalar representing a counter clock wise rotation
   * as a single angle in radian. It provides some additional features such as the automatic
@@ -59,41 +59,79 @@
 public:
 
   /** Construct a 2D counter clock wise rotation from the angle \a a in radian. */
-  inline Rotation2D(const Scalar& a) : m_angle(a) {}
+  EIGEN_DEVICE_FUNC explicit inline Rotation2D(const Scalar& a) : m_angle(a) {}
   
   /** Default constructor wihtout initialization. The represented rotation is undefined. */
-  Rotation2D() {}
+  EIGEN_DEVICE_FUNC Rotation2D() {}
+
+  /** Construct a 2D rotation from a 2x2 rotation matrix \a mat.
+    *
+    * \sa fromRotationMatrix()
+    */
+  template<typename Derived>
+  EIGEN_DEVICE_FUNC explicit Rotation2D(const MatrixBase<Derived>& m)
+  {
+    fromRotationMatrix(m.derived());
+  }
 
   /** \returns the rotation angle */
-  inline Scalar angle() const { return m_angle; }
+  EIGEN_DEVICE_FUNC inline Scalar angle() const { return m_angle; }
 
   /** \returns a read-write reference to the rotation angle */
-  inline Scalar& angle() { return m_angle; }
+  EIGEN_DEVICE_FUNC inline Scalar& angle() { return m_angle; }
+  
+  /** \returns the rotation angle in [0,2pi] */
+  EIGEN_DEVICE_FUNC inline Scalar smallestPositiveAngle() const {
+    Scalar tmp = numext::fmod(m_angle,Scalar(2*EIGEN_PI));
+    return tmp<Scalar(0) ? tmp + Scalar(2*EIGEN_PI) : tmp;
+  }
+  
+  /** \returns the rotation angle in [-pi,pi] */
+  EIGEN_DEVICE_FUNC inline Scalar smallestAngle() const {
+    Scalar tmp = numext::fmod(m_angle,Scalar(2*EIGEN_PI));
+    if(tmp>Scalar(EIGEN_PI))       tmp -= Scalar(2*EIGEN_PI);
+    else if(tmp<-Scalar(EIGEN_PI)) tmp += Scalar(2*EIGEN_PI);
+    return tmp;
+  }
 
   /** \returns the inverse rotation */
-  inline Rotation2D inverse() const { return -m_angle; }
+  EIGEN_DEVICE_FUNC inline Rotation2D inverse() const { return Rotation2D(-m_angle); }
 
   /** Concatenates two rotations */
-  inline Rotation2D operator*(const Rotation2D& other) const
-  { return m_angle + other.m_angle; }
+  EIGEN_DEVICE_FUNC inline Rotation2D operator*(const Rotation2D& other) const
+  { return Rotation2D(m_angle + other.m_angle); }
 
   /** Concatenates two rotations */
-  inline Rotation2D& operator*=(const Rotation2D& other)
+  EIGEN_DEVICE_FUNC inline Rotation2D& operator*=(const Rotation2D& other)
   { m_angle += other.m_angle; return *this; }
 
   /** Applies the rotation to a 2D vector */
-  Vector2 operator* (const Vector2& vec) const
+  EIGEN_DEVICE_FUNC Vector2 operator* (const Vector2& vec) const
   { return toRotationMatrix() * vec; }
   
   template<typename Derived>
-  Rotation2D& fromRotationMatrix(const MatrixBase<Derived>& m);
-  Matrix2 toRotationMatrix() const;
+  EIGEN_DEVICE_FUNC Rotation2D& fromRotationMatrix(const MatrixBase<Derived>& m);
+  EIGEN_DEVICE_FUNC Matrix2 toRotationMatrix() const;
+
+  /** Set \c *this from a 2x2 rotation matrix \a mat.
+    * In other words, this function extract the rotation angle from the rotation matrix.
+    *
+    * This method is an alias for fromRotationMatrix()
+    *
+    * \sa fromRotationMatrix()
+    */
+  template<typename Derived>
+  EIGEN_DEVICE_FUNC Rotation2D& operator=(const  MatrixBase<Derived>& m)
+  { return fromRotationMatrix(m.derived()); }
 
   /** \returns the spherical interpolation between \c *this and \a other using
     * parameter \a t. It is in fact equivalent to a linear interpolation.
     */
-  inline Rotation2D slerp(const Scalar& t, const Rotation2D& other) const
-  { return m_angle * (1-t) + other.angle() * t; }
+  EIGEN_DEVICE_FUNC inline Rotation2D slerp(const Scalar& t, const Rotation2D& other) const
+  {
+    Scalar dist = Rotation2D(other.m_angle-m_angle).smallestAngle();
+    return Rotation2D(m_angle + dist*t);
+  }
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -101,24 +139,25 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type cast() const
   { return typename internal::cast_return_type<Rotation2D,Rotation2D<NewScalarType> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit Rotation2D(const Rotation2D<OtherScalarType>& other)
+  EIGEN_DEVICE_FUNC inline explicit Rotation2D(const Rotation2D<OtherScalarType>& other)
   {
     m_angle = Scalar(other.angle());
   }
 
-  static inline Rotation2D Identity() { return Rotation2D(0); }
+  EIGEN_DEVICE_FUNC static inline Rotation2D Identity() { return Rotation2D(0); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const Rotation2D& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Rotation2D& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return internal::isApprox(m_angle,other.m_angle, prec); }
+  
 };
 
 /** \ingroup Geometry_Module
@@ -134,9 +173,9 @@
   */
 template<typename Scalar>
 template<typename Derived>
-Rotation2D<Scalar>& Rotation2D<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
+EIGEN_DEVICE_FUNC Rotation2D<Scalar>& Rotation2D<Scalar>::fromRotationMatrix(const MatrixBase<Derived>& mat)
 {
-  using std::atan2;
+  EIGEN_USING_STD_MATH(atan2)
   EIGEN_STATIC_ASSERT(Derived::RowsAtCompileTime==2 && Derived::ColsAtCompileTime==2,YOU_MADE_A_PROGRAMMING_MISTAKE)
   m_angle = atan2(mat.coeff(1,0), mat.coeff(0,0));
   return *this;
@@ -146,10 +185,10 @@
   */
 template<typename Scalar>
 typename Rotation2D<Scalar>::Matrix2
-Rotation2D<Scalar>::toRotationMatrix(void) const
+EIGEN_DEVICE_FUNC Rotation2D<Scalar>::toRotationMatrix(void) const
 {
-  using std::sin;
-  using std::cos;
+  EIGEN_USING_STD_MATH(sin)
+  EIGEN_USING_STD_MATH(cos)
   Scalar sinA = sin(m_angle);
   Scalar cosA = cos(m_angle);
   return (Matrix2() << cosA, -sinA, sinA, cosA).finished();
diff --git a/Eigen/src/Geometry/RotationBase.h b/Eigen/src/Geometry/RotationBase.h
index b88661d..f0ee0bd 100644
--- a/Eigen/src/Geometry/RotationBase.h
+++ b/Eigen/src/Geometry/RotationBase.h
@@ -22,8 +22,8 @@
   *
   * \brief Common base class for compact rotation representations
   *
-  * \param Derived is the derived type, i.e., a rotation type
-  * \param _Dim the dimension of the space
+  * \tparam Derived is the derived type, i.e., a rotation type
+  * \tparam _Dim the dimension of the space
   */
 template<typename Derived, int _Dim>
 class RotationBase
@@ -38,26 +38,26 @@
     typedef Matrix<Scalar,Dim,1> VectorType;
 
   public:
-    inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
-    inline Derived& derived() { return *static_cast<Derived*>(this); }
+    EIGEN_DEVICE_FUNC inline const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    EIGEN_DEVICE_FUNC inline Derived& derived() { return *static_cast<Derived*>(this); }
 
     /** \returns an equivalent rotation matrix */
-    inline RotationMatrixType toRotationMatrix() const { return derived().toRotationMatrix(); }
+    EIGEN_DEVICE_FUNC inline RotationMatrixType toRotationMatrix() const { return derived().toRotationMatrix(); }
 
     /** \returns an equivalent rotation matrix 
       * This function is added to be conform with the Transform class' naming scheme.
       */
-    inline RotationMatrixType matrix() const { return derived().toRotationMatrix(); }
+    EIGEN_DEVICE_FUNC inline RotationMatrixType matrix() const { return derived().toRotationMatrix(); }
 
     /** \returns the inverse rotation */
-    inline Derived inverse() const { return derived().inverse(); }
+    EIGEN_DEVICE_FUNC inline Derived inverse() const { return derived().inverse(); }
 
     /** \returns the concatenation of the rotation \c *this with a translation \a t */
-    inline Transform<Scalar,Dim,Isometry> operator*(const Translation<Scalar,Dim>& t) const
+    EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Isometry> operator*(const Translation<Scalar,Dim>& t) const
     { return Transform<Scalar,Dim,Isometry>(*this) * t; }
 
     /** \returns the concatenation of the rotation \c *this with a uniform scaling \a s */
-    inline RotationMatrixType operator*(const UniformScaling<Scalar>& s) const
+    EIGEN_DEVICE_FUNC inline RotationMatrixType operator*(const UniformScaling<Scalar>& s) const
     { return toRotationMatrix() * s.factor(); }
 
     /** \returns the concatenation of the rotation \c *this with a generic expression \a e
@@ -67,17 +67,17 @@
       *  - a vector of size Dim
       */
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE typename internal::rotation_base_generic_product_selector<Derived,OtherDerived,OtherDerived::IsVectorAtCompileTime>::ReturnType
+    EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::rotation_base_generic_product_selector<Derived,OtherDerived,OtherDerived::IsVectorAtCompileTime>::ReturnType
     operator*(const EigenBase<OtherDerived>& e) const
     { return internal::rotation_base_generic_product_selector<Derived,OtherDerived>::run(derived(), e.derived()); }
 
     /** \returns the concatenation of a linear transformation \a l with the rotation \a r */
     template<typename OtherDerived> friend
-    inline RotationMatrixType operator*(const EigenBase<OtherDerived>& l, const Derived& r)
+    EIGEN_DEVICE_FUNC inline RotationMatrixType operator*(const EigenBase<OtherDerived>& l, const Derived& r)
     { return l.derived() * r.toRotationMatrix(); }
 
     /** \returns the concatenation of a scaling \a l with the rotation \a r */
-    friend inline Transform<Scalar,Dim,Affine> operator*(const DiagonalMatrix<Scalar,Dim>& l, const Derived& r)
+    EIGEN_DEVICE_FUNC friend inline Transform<Scalar,Dim,Affine> operator*(const DiagonalMatrix<Scalar,Dim>& l, const Derived& r)
     { 
       Transform<Scalar,Dim,Affine> res(r);
       res.linear().applyOnTheLeft(l);
@@ -86,11 +86,11 @@
 
     /** \returns the concatenation of the rotation \c *this with a transformation \a t */
     template<int Mode, int Options>
-    inline Transform<Scalar,Dim,Mode> operator*(const Transform<Scalar,Dim,Mode,Options>& t) const
+    EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode> operator*(const Transform<Scalar,Dim,Mode,Options>& t) const
     { return toRotationMatrix() * t; }
 
     template<typename OtherVectorType>
-    inline VectorType _transformVector(const OtherVectorType& v) const
+    EIGEN_DEVICE_FUNC inline VectorType _transformVector(const OtherVectorType& v) const
     { return toRotationMatrix() * v; }
 };
 
@@ -102,7 +102,7 @@
 {
   enum { Dim = RotationDerived::Dim };
   typedef Matrix<typename RotationDerived::Scalar,Dim,Dim> ReturnType;
-  static inline ReturnType run(const RotationDerived& r, const MatrixType& m)
+  EIGEN_DEVICE_FUNC static inline ReturnType run(const RotationDerived& r, const MatrixType& m)
   { return r.toRotationMatrix() * m; }
 };
 
@@ -110,7 +110,7 @@
 struct rotation_base_generic_product_selector< RotationDerived, DiagonalMatrix<Scalar,Dim,MaxDim>, false >
 {
   typedef Transform<Scalar,Dim,Affine> ReturnType;
-  static inline ReturnType run(const RotationDerived& r, const DiagonalMatrix<Scalar,Dim,MaxDim>& m)
+  EIGEN_DEVICE_FUNC static inline ReturnType run(const RotationDerived& r, const DiagonalMatrix<Scalar,Dim,MaxDim>& m)
   {
     ReturnType res(r);
     res.linear() *= m;
@@ -123,7 +123,7 @@
 {
   enum { Dim = RotationDerived::Dim };
   typedef Matrix<typename RotationDerived::Scalar,Dim,1> ReturnType;
-  static EIGEN_STRONG_INLINE ReturnType run(const RotationDerived& r, const OtherVectorType& v)
+  EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE ReturnType run(const RotationDerived& r, const OtherVectorType& v)
   {
     return r._transformVector(v);
   }
@@ -137,7 +137,7 @@
   */
 template<typename _Scalar, int _Rows, int _Cols, int _Storage, int _MaxRows, int _MaxCols>
 template<typename OtherDerived>
-Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
+EIGEN_DEVICE_FUNC Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
 ::Matrix(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
 {
   EIGEN_STATIC_ASSERT_MATRIX_SPECIFIC_SIZE(Matrix,int(OtherDerived::Dim),int(OtherDerived::Dim))
@@ -150,7 +150,7 @@
   */
 template<typename _Scalar, int _Rows, int _Cols, int _Storage, int _MaxRows, int _MaxCols>
 template<typename OtherDerived>
-Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>&
+EIGEN_DEVICE_FUNC Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>&
 Matrix<_Scalar, _Rows, _Cols, _Storage, _MaxRows, _MaxCols>
 ::operator=(const RotationBase<OtherDerived,ColsAtCompileTime>& r)
 {
@@ -164,8 +164,8 @@
   *
   * Helper function to return an arbitrary rotation object to a rotation matrix.
   *
-  * \param Scalar the numeric type of the matrix coefficients
-  * \param Dim the dimension of the current space
+  * \tparam Scalar the numeric type of the matrix coefficients
+  * \tparam Dim the dimension of the current space
   *
   * It returns a Dim x Dim fixed size matrix.
   *
@@ -179,20 +179,20 @@
   * \sa class Transform, class Rotation2D, class Quaternion, class AngleAxis
   */
 template<typename Scalar, int Dim>
-static inline Matrix<Scalar,2,2> toRotationMatrix(const Scalar& s)
+EIGEN_DEVICE_FUNC static inline Matrix<Scalar,2,2> toRotationMatrix(const Scalar& s)
 {
   EIGEN_STATIC_ASSERT(Dim==2,YOU_MADE_A_PROGRAMMING_MISTAKE)
   return Rotation2D<Scalar>(s).toRotationMatrix();
 }
 
 template<typename Scalar, int Dim, typename OtherDerived>
-static inline Matrix<Scalar,Dim,Dim> toRotationMatrix(const RotationBase<OtherDerived,Dim>& r)
+EIGEN_DEVICE_FUNC static inline Matrix<Scalar,Dim,Dim> toRotationMatrix(const RotationBase<OtherDerived,Dim>& r)
 {
   return r.toRotationMatrix();
 }
 
 template<typename Scalar, int Dim, typename OtherDerived>
-static inline const MatrixBase<OtherDerived>& toRotationMatrix(const MatrixBase<OtherDerived>& mat)
+EIGEN_DEVICE_FUNC static inline const MatrixBase<OtherDerived>& toRotationMatrix(const MatrixBase<OtherDerived>& mat)
 {
   EIGEN_STATIC_ASSERT(OtherDerived::RowsAtCompileTime==Dim && OtherDerived::ColsAtCompileTime==Dim,
     YOU_MADE_A_PROGRAMMING_MISTAKE)
diff --git a/Eigen/src/Geometry/Scaling.h b/Eigen/src/Geometry/Scaling.h
old mode 100644
new mode 100755
index 1c25f36..f58ca03
--- a/Eigen/src/Geometry/Scaling.h
+++ b/Eigen/src/Geometry/Scaling.h
@@ -18,7 +18,7 @@
   *
   * \brief Represents a generic uniform scaling transformation
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients.
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients.
   *
   * This class represent a uniform scaling transformation. It is the return
   * type of Scaling(Scalar), and most of the time this is the only way it
@@ -62,10 +62,10 @@
   template<int Dim, int Mode, int Options>
   inline Transform<Scalar,Dim,(int(Mode)==int(Isometry)?Affine:Mode)> operator* (const Transform<Scalar,Dim, Mode, Options>& t) const
   {
-   Transform<Scalar,Dim,(int(Mode)==int(Isometry)?Affine:Mode)> res = t;
-   res.prescale(factor());
-   return res;
-}
+    Transform<Scalar,Dim,(int(Mode)==int(Isometry)?Affine:Mode)> res = t;
+    res.prescale(factor());
+    return res;
+  }
 
   /** Concatenates a uniform scaling and a linear transformation matrix */
   // TODO returns an expression
@@ -104,40 +104,44 @@
 
 };
 
-/** Concatenates a linear transformation matrix and a uniform scaling */
+/** \addtogroup Geometry_Module */
+//@{
+
+/** Concatenates a linear transformation matrix and a uniform scaling
+  * \relates UniformScaling
+  */
 // NOTE this operator is defiend in MatrixBase and not as a friend function
 // of UniformScaling to fix an internal crash of Intel's ICC
-template<typename Derived> typename MatrixBase<Derived>::ScalarMultipleReturnType
-MatrixBase<Derived>::operator*(const UniformScaling<Scalar>& s) const
-{ return derived() * s.factor(); }
+template<typename Derived,typename Scalar>
+EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,Scalar,product)
+operator*(const MatrixBase<Derived>& matrix, const UniformScaling<Scalar>& s)
+{ return matrix.derived() * s.factor(); }
 
 /** Constructs a uniform scaling from scale factor \a s */
-static inline UniformScaling<float> Scaling(float s) { return UniformScaling<float>(s); }
+inline UniformScaling<float> Scaling(float s) { return UniformScaling<float>(s); }
 /** Constructs a uniform scaling from scale factor \a s */
-static inline UniformScaling<double> Scaling(double s) { return UniformScaling<double>(s); }
+inline UniformScaling<double> Scaling(double s) { return UniformScaling<double>(s); }
 /** Constructs a uniform scaling from scale factor \a s */
 template<typename RealScalar>
-static inline UniformScaling<std::complex<RealScalar> > Scaling(const std::complex<RealScalar>& s)
+inline UniformScaling<std::complex<RealScalar> > Scaling(const std::complex<RealScalar>& s)
 { return UniformScaling<std::complex<RealScalar> >(s); }
 
 /** Constructs a 2D axis aligned scaling */
 template<typename Scalar>
-static inline DiagonalMatrix<Scalar,2> Scaling(const Scalar& sx, const Scalar& sy)
+inline DiagonalMatrix<Scalar,2> Scaling(const Scalar& sx, const Scalar& sy)
 { return DiagonalMatrix<Scalar,2>(sx, sy); }
 /** Constructs a 3D axis aligned scaling */
 template<typename Scalar>
-static inline DiagonalMatrix<Scalar,3> Scaling(const Scalar& sx, const Scalar& sy, const Scalar& sz)
+inline DiagonalMatrix<Scalar,3> Scaling(const Scalar& sx, const Scalar& sy, const Scalar& sz)
 { return DiagonalMatrix<Scalar,3>(sx, sy, sz); }
 
 /** Constructs an axis aligned scaling expression from vector expression \a coeffs
   * This is an alias for coeffs.asDiagonal()
   */
 template<typename Derived>
-static inline const DiagonalWrapper<const Derived> Scaling(const MatrixBase<Derived>& coeffs)
+inline const DiagonalWrapper<const Derived> Scaling(const MatrixBase<Derived>& coeffs)
 { return coeffs.asDiagonal(); }
 
-/** \addtogroup Geometry_Module */
-//@{
 /** \deprecated */
 typedef DiagonalMatrix<float, 2> AlignedScaling2f;
 /** \deprecated */
diff --git a/Eigen/src/Geometry/Transform.h b/Eigen/src/Geometry/Transform.h
index e786e53..3f31ee4 100644
--- a/Eigen/src/Geometry/Transform.h
+++ b/Eigen/src/Geometry/Transform.h
@@ -32,7 +32,8 @@
           typename MatrixType,
           int Case = transform_traits<TransformType>::IsProjective ? 0
                    : int(MatrixType::RowsAtCompileTime) == int(transform_traits<TransformType>::HDim) ? 1
-                   : 2>
+                   : 2,
+          int RhsCols = MatrixType::ColsAtCompileTime>
 struct transform_right_product_impl;
 
 template< typename Other,
@@ -62,6 +63,22 @@
 
 template<typename TransformType> struct transform_take_affine_part;
 
+template<typename _Scalar, int _Dim, int _Mode, int _Options>
+struct traits<Transform<_Scalar,_Dim,_Mode,_Options> >
+{
+  typedef _Scalar Scalar;
+  typedef Eigen::Index StorageIndex;
+  typedef Dense StorageKind;
+  enum {
+    Dim1 = _Dim==Dynamic ? _Dim : _Dim + 1,
+    RowsAtCompileTime = _Mode==Projective ? Dim1 : _Dim,
+    ColsAtCompileTime = Dim1,
+    MaxRowsAtCompileTime = RowsAtCompileTime,
+    MaxColsAtCompileTime = ColsAtCompileTime,
+    Flags = 0
+  };
+};
+
 template<int Mode> struct transform_make_affine;
 
 } // end namespace internal
@@ -102,15 +119,15 @@
   *
   * However, unlike a plain matrix, the Transform class provides many features
   * simplifying both its assembly and usage. In particular, it can be composed
-  * with any other transformations (Transform,Translation,RotationBase,Matrix)
+  * with any other transformations (Transform,Translation,RotationBase,DiagonalMatrix)
   * and can be directly used to transform implicit homogeneous vectors. All these
   * operations are handled via the operator*. For the composition of transformations,
   * its principle consists to first convert the right/left hand sides of the product
   * to a compatible (Dim+1)^2 matrix and then perform a pure matrix product.
   * Of course, internally, operator* tries to perform the minimal number of operations
   * according to the nature of each terms. Likewise, when applying the transform
-  * to non homogeneous vectors, the latters are automatically promoted to homogeneous
-  * one before doing the matrix product. The convertions to homogeneous representations
+  * to points, the latters are automatically promoted to homogeneous vectors
+  * before doing the matrix product. The conventions to homogeneous representations
   * are performed as follow:
   *
   * \b Translation t (Dim)x(1):
@@ -124,7 +141,7 @@
   * R & 0\\
   * 0\,...\,0 & 1
   * \end{array} \right) \f$
-  *
+  *<!--
   * \b Linear \b Matrix L (Dim)x(Dim):
   * \f$ \left( \begin{array}{cc}
   * L & 0\\
@@ -136,14 +153,20 @@
   * A\\
   * 0\,...\,0\,1
   * \end{array} \right) \f$
+  *-->
+  * \b Scaling \b DiagonalMatrix S (Dim)x(Dim):
+  * \f$ \left( \begin{array}{cc}
+  * S & 0\\
+  * 0\,...\,0 & 1
+  * \end{array} \right) \f$
   *
-  * \b Column \b vector v (Dim)x(1):
+  * \b Column \b point v (Dim)x(1):
   * \f$ \left( \begin{array}{c}
   * v\\
   * 1
   * \end{array} \right) \f$
   *
-  * \b Set \b of \b column \b vectors V1...Vn (Dim)x(n):
+  * \b Set \b of \b column \b points V1...Vn (Dim)x(n):
   * \f$ \left( \begin{array}{ccc}
   * v_1 & ... & v_n\\
   * 1 & ... & 1
@@ -170,7 +193,7 @@
   * preprocessor token EIGEN_QT_SUPPORT is defined.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_TRANSFORM_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_TRANSFORM_PLUGIN.
   *
   * \sa class Matrix, class Quaternion
   */
@@ -188,7 +211,8 @@
   };
   /** the scalar type of the coefficients */
   typedef _Scalar Scalar;
-  typedef DenseIndex Index;
+  typedef Eigen::Index StorageIndex;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
   /** type of the matrix used to represent the transformation */
   typedef typename internal::make_proper_matrix_type<Scalar,Rows,HDim,Options>::type MatrixType;
   /** constified MatrixType */
@@ -210,9 +234,9 @@
   /** type of a vector */
   typedef Matrix<Scalar,Dim,1> VectorType;
   /** type of a read/write reference to the translation part of the rotation */
-  typedef Block<MatrixType,Dim,1,int(Mode)==(AffineCompact)> TranslationPart;
+  typedef Block<MatrixType,Dim,1,!(internal::traits<MatrixType>::Flags & RowMajorBit)> TranslationPart;
   /** type of a read reference to the translation part of the rotation */
-  typedef const Block<ConstMatrixType,Dim,1,int(Mode)==(AffineCompact)> ConstTranslationPart;
+  typedef const Block<ConstMatrixType,Dim,1,!(internal::traits<MatrixType>::Flags & RowMajorBit)> ConstTranslationPart;
   /** corresponding translation type */
   typedef Translation<Scalar,Dim> TranslationType;
   
@@ -229,43 +253,43 @@
 
   /** Default constructor without initialization of the meaningful coefficients.
     * If Mode==Affine, then the last row is set to [0 ... 0 1] */
-  inline Transform()
+  EIGEN_DEVICE_FUNC inline Transform()
   {
     check_template_params();
     internal::transform_make_affine<(int(Mode)==Affine) ? Affine : AffineCompact>::run(m_matrix);
   }
 
-  inline Transform(const Transform& other)
+  EIGEN_DEVICE_FUNC inline Transform(const Transform& other)
   {
     check_template_params();
     m_matrix = other.m_matrix;
   }
 
-  inline explicit Transform(const TranslationType& t)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const TranslationType& t)
   {
     check_template_params();
     *this = t;
   }
-  inline explicit Transform(const UniformScaling<Scalar>& s)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const UniformScaling<Scalar>& s)
   {
     check_template_params();
     *this = s;
   }
   template<typename Derived>
-  inline explicit Transform(const RotationBase<Derived, Dim>& r)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const RotationBase<Derived, Dim>& r)
   {
     check_template_params();
     *this = r;
   }
 
-  inline Transform& operator=(const Transform& other)
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const Transform& other)
   { m_matrix = other.m_matrix; return *this; }
 
   typedef internal::transform_take_affine_part<Transform> take_affine_part;
 
   /** Constructs and initializes a transformation from a Dim^2 or a (Dim+1)^2 matrix. */
   template<typename OtherDerived>
-  inline explicit Transform(const EigenBase<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const EigenBase<OtherDerived>& other)
   {
     EIGEN_STATIC_ASSERT((internal::is_same<Scalar,typename OtherDerived::Scalar>::value),
       YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY);
@@ -276,7 +300,7 @@
 
   /** Set \c *this from a Dim^2 or (Dim+1)^2 matrix. */
   template<typename OtherDerived>
-  inline Transform& operator=(const EigenBase<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const EigenBase<OtherDerived>& other)
   {
     EIGEN_STATIC_ASSERT((internal::is_same<Scalar,typename OtherDerived::Scalar>::value),
       YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY);
@@ -286,7 +310,7 @@
   }
   
   template<int OtherOptions>
-  inline Transform(const Transform<Scalar,Dim,Mode,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline Transform(const Transform<Scalar,Dim,Mode,OtherOptions>& other)
   {
     check_template_params();
     // only the options change, we can directly copy the matrices
@@ -294,7 +318,7 @@
   }
 
   template<int OtherMode,int OtherOptions>
-  inline Transform(const Transform<Scalar,Dim,OtherMode,OtherOptions>& other)
+  EIGEN_DEVICE_FUNC inline Transform(const Transform<Scalar,Dim,OtherMode,OtherOptions>& other)
   {
     check_template_params();
     // prevent conversions as:
@@ -335,14 +359,14 @@
   }
 
   template<typename OtherDerived>
-  Transform(const ReturnByValue<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC Transform(const ReturnByValue<OtherDerived>& other)
   {
     check_template_params();
     other.evalTo(*this);
   }
 
   template<typename OtherDerived>
-  Transform& operator=(const ReturnByValue<OtherDerived>& other)
+  EIGEN_DEVICE_FUNC Transform& operator=(const ReturnByValue<OtherDerived>& other)
   {
     other.evalTo(*this);
     return *this;
@@ -356,60 +380,76 @@
   inline Transform& operator=(const QTransform& other);
   inline QTransform toQTransform(void) const;
   #endif
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return int(Mode)==int(Projective) ? m_matrix.cols() : (m_matrix.cols()-1); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_matrix.cols(); }
 
   /** shortcut for m_matrix(row,col);
     * \sa MatrixBase::operator(Index,Index) const */
-  inline Scalar operator() (Index row, Index col) const { return m_matrix(row,col); }
+  EIGEN_DEVICE_FUNC inline Scalar operator() (Index row, Index col) const { return m_matrix(row,col); }
   /** shortcut for m_matrix(row,col);
     * \sa MatrixBase::operator(Index,Index) */
-  inline Scalar& operator() (Index row, Index col) { return m_matrix(row,col); }
+  EIGEN_DEVICE_FUNC inline Scalar& operator() (Index row, Index col) { return m_matrix(row,col); }
 
   /** \returns a read-only expression of the transformation matrix */
-  inline const MatrixType& matrix() const { return m_matrix; }
+  EIGEN_DEVICE_FUNC inline const MatrixType& matrix() const { return m_matrix; }
   /** \returns a writable expression of the transformation matrix */
-  inline MatrixType& matrix() { return m_matrix; }
+  EIGEN_DEVICE_FUNC inline MatrixType& matrix() { return m_matrix; }
 
   /** \returns a read-only expression of the linear part of the transformation */
-  inline ConstLinearPart linear() const { return ConstLinearPart(m_matrix,0,0); }
+  EIGEN_DEVICE_FUNC inline ConstLinearPart linear() const { return ConstLinearPart(m_matrix,0,0); }
   /** \returns a writable expression of the linear part of the transformation */
-  inline LinearPart linear() { return LinearPart(m_matrix,0,0); }
+  EIGEN_DEVICE_FUNC inline LinearPart linear() { return LinearPart(m_matrix,0,0); }
 
   /** \returns a read-only expression of the Dim x HDim affine part of the transformation */
-  inline ConstAffinePart affine() const { return take_affine_part::run(m_matrix); }
+  EIGEN_DEVICE_FUNC inline ConstAffinePart affine() const { return take_affine_part::run(m_matrix); }
   /** \returns a writable expression of the Dim x HDim affine part of the transformation */
-  inline AffinePart affine() { return take_affine_part::run(m_matrix); }
+  EIGEN_DEVICE_FUNC inline AffinePart affine() { return take_affine_part::run(m_matrix); }
 
   /** \returns a read-only expression of the translation vector of the transformation */
-  inline ConstTranslationPart translation() const { return ConstTranslationPart(m_matrix,0,Dim); }
+  EIGEN_DEVICE_FUNC inline ConstTranslationPart translation() const { return ConstTranslationPart(m_matrix,0,Dim); }
   /** \returns a writable expression of the translation vector of the transformation */
-  inline TranslationPart translation() { return TranslationPart(m_matrix,0,Dim); }
+  EIGEN_DEVICE_FUNC inline TranslationPart translation() { return TranslationPart(m_matrix,0,Dim); }
 
-  /** \returns an expression of the product between the transform \c *this and a matrix expression \a other
+  /** \returns an expression of the product between the transform \c *this and a matrix expression \a other.
     *
-    * The right hand side \a other might be either:
-    * \li a vector of size Dim,
+    * The right-hand-side \a other can be either:
     * \li an homogeneous vector of size Dim+1,
-    * \li a set of vectors of size Dim x Dynamic,
-    * \li a set of homogeneous vectors of size Dim+1 x Dynamic,
-    * \li a linear transformation matrix of size Dim x Dim,
-    * \li an affine transformation matrix of size Dim x Dim+1,
+    * \li a set of homogeneous vectors of size Dim+1 x N,
     * \li a transformation matrix of size Dim+1 x Dim+1.
+    *
+    * Moreover, if \c *this represents an affine transformation (i.e., Mode!=Projective), then \a other can also be:
+    * \li a point of size Dim (computes: \code this->linear() * other + this->translation()\endcode),
+    * \li a set of N points as a Dim x N matrix (computes: \code (this->linear() * other).colwise() + this->translation()\endcode),
+    *
+    * In all cases, the return type is a matrix or vector of same sizes as the right-hand-side \a other.
+    *
+    * If you want to interpret \a other as a linear or affine transformation, then first convert it to a Transform<> type,
+    * or do your own cooking.
+    *
+    * Finally, if you want to apply Affine transformations to vectors, then explicitly apply the linear part only:
+    * \code
+    * Affine3f A;
+    * Vector3f v1, v2;
+    * v2 = A.linear() * v1;
+    * \endcode
+    *
     */
   // note: this function is defined here because some compilers cannot find the respective declaration
   template<typename OtherDerived>
-  EIGEN_STRONG_INLINE const typename internal::transform_right_product_impl<Transform, OtherDerived>::ResultType
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename internal::transform_right_product_impl<Transform, OtherDerived>::ResultType
   operator * (const EigenBase<OtherDerived> &other) const
   { return internal::transform_right_product_impl<Transform, OtherDerived>::run(*this,other.derived()); }
 
   /** \returns the product expression of a transformation matrix \a a times a transform \a b
     *
-    * The left hand side \a other might be either:
+    * The left hand side \a other can be either:
     * \li a linear transformation matrix of size Dim x Dim,
     * \li an affine transformation matrix of size Dim x Dim+1,
     * \li a general transformation matrix of size Dim+1 x Dim+1.
     */
   template<typename OtherDerived> friend
-  inline const typename internal::transform_left_product_impl<OtherDerived,Mode,Options,_Dim,_Dim+1>::ResultType
+  EIGEN_DEVICE_FUNC inline const typename internal::transform_left_product_impl<OtherDerived,Mode,Options,_Dim,_Dim+1>::ResultType
     operator * (const EigenBase<OtherDerived> &a, const Transform &b)
   { return internal::transform_left_product_impl<OtherDerived,Mode,Options,Dim,HDim>::run(a.derived(),b); }
 
@@ -420,11 +460,11 @@
     * mode is no isometry. In that case, the returned transform is an affinity.
     */
   template<typename DiagonalDerived>
-  inline const TransformTimeDiagonalReturnType
+  EIGEN_DEVICE_FUNC inline const TransformTimeDiagonalReturnType
     operator * (const DiagonalBase<DiagonalDerived> &b) const
   {
     TransformTimeDiagonalReturnType res(*this);
-    res.linear() *= b;
+    res.linearExt() *= b;
     return res;
   }
 
@@ -435,7 +475,7 @@
     * mode is no isometry. In that case, the returned transform is an affinity.
     */
   template<typename DiagonalDerived>
-  friend inline TransformTimeDiagonalReturnType
+  EIGEN_DEVICE_FUNC friend inline TransformTimeDiagonalReturnType
     operator * (const DiagonalBase<DiagonalDerived> &a, const Transform &b)
   {
     TransformTimeDiagonalReturnType res;
@@ -447,15 +487,15 @@
   }
 
   template<typename OtherDerived>
-  inline Transform& operator*=(const EigenBase<OtherDerived>& other) { return *this = *this * other; }
+  EIGEN_DEVICE_FUNC inline Transform& operator*=(const EigenBase<OtherDerived>& other) { return *this = *this * other; }
 
   /** Concatenates two transformations */
-  inline const Transform operator * (const Transform& other) const
+  EIGEN_DEVICE_FUNC inline const Transform operator * (const Transform& other) const
   {
     return internal::transform_transform_product_impl<Transform,Transform>::run(*this,other);
   }
   
-  #ifdef __INTEL_COMPILER
+  #if EIGEN_COMP_ICC
 private:
   // this intermediate structure permits to workaround a bug in ICC 11:
   //   error: template instantiation resulted in unexpected function type of "Eigen::Transform<double, 3, 32, 0>
@@ -482,7 +522,7 @@
   #else
   /** Concatenates two different transformations */
   template<int OtherMode,int OtherOptions>
-  inline typename internal::transform_transform_product_impl<Transform,Transform<Scalar,Dim,OtherMode,OtherOptions> >::ResultType
+  EIGEN_DEVICE_FUNC inline typename internal::transform_transform_product_impl<Transform,Transform<Scalar,Dim,OtherMode,OtherOptions> >::ResultType
     operator * (const Transform<Scalar,Dim,OtherMode,OtherOptions>& other) const
   {
     return internal::transform_transform_product_impl<Transform,Transform<Scalar,Dim,OtherMode,OtherOptions> >::run(*this,other);
@@ -490,79 +530,98 @@
   #endif
 
   /** \sa MatrixBase::setIdentity() */
-  void setIdentity() { m_matrix.setIdentity(); }
+  EIGEN_DEVICE_FUNC void setIdentity() { m_matrix.setIdentity(); }
 
   /**
    * \brief Returns an identity transformation.
    * \todo In the future this function should be returning a Transform expression.
    */
-  static const Transform Identity()
+  EIGEN_DEVICE_FUNC static const Transform Identity()
   {
     return Transform(MatrixType::Identity());
   }
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC 
   inline Transform& scale(const MatrixBase<OtherDerived> &other);
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline Transform& prescale(const MatrixBase<OtherDerived> &other);
 
-  inline Transform& scale(const Scalar& s);
-  inline Transform& prescale(const Scalar& s);
+  EIGEN_DEVICE_FUNC inline Transform& scale(const Scalar& s);
+  EIGEN_DEVICE_FUNC inline Transform& prescale(const Scalar& s);
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline Transform& translate(const MatrixBase<OtherDerived> &other);
 
   template<typename OtherDerived>
+  EIGEN_DEVICE_FUNC
   inline Transform& pretranslate(const MatrixBase<OtherDerived> &other);
 
   template<typename RotationType>
+  EIGEN_DEVICE_FUNC
   inline Transform& rotate(const RotationType& rotation);
 
   template<typename RotationType>
+  EIGEN_DEVICE_FUNC
   inline Transform& prerotate(const RotationType& rotation);
 
-  Transform& shear(const Scalar& sx, const Scalar& sy);
-  Transform& preshear(const Scalar& sx, const Scalar& sy);
+  EIGEN_DEVICE_FUNC Transform& shear(const Scalar& sx, const Scalar& sy);
+  EIGEN_DEVICE_FUNC Transform& preshear(const Scalar& sx, const Scalar& sy);
 
-  inline Transform& operator=(const TranslationType& t);
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const TranslationType& t);
+  
+  EIGEN_DEVICE_FUNC
   inline Transform& operator*=(const TranslationType& t) { return translate(t.vector()); }
-  inline Transform operator*(const TranslationType& t) const;
+  
+  EIGEN_DEVICE_FUNC inline Transform operator*(const TranslationType& t) const;
 
+  EIGEN_DEVICE_FUNC 
   inline Transform& operator=(const UniformScaling<Scalar>& t);
+  
+  EIGEN_DEVICE_FUNC
   inline Transform& operator*=(const UniformScaling<Scalar>& s) { return scale(s.factor()); }
-  inline Transform<Scalar,Dim,(int(Mode)==int(Isometry)?int(Affine):int(Mode))> operator*(const UniformScaling<Scalar>& s) const
+  
+  EIGEN_DEVICE_FUNC
+  inline TransformTimeDiagonalReturnType operator*(const UniformScaling<Scalar>& s) const
   {
-    Transform<Scalar,Dim,(int(Mode)==int(Isometry)?int(Affine):int(Mode)),Options> res = *this;
+    TransformTimeDiagonalReturnType res = *this;
     res.scale(s.factor());
     return res;
   }
 
-  inline Transform& operator*=(const DiagonalMatrix<Scalar,Dim>& s) { linear() *= s; return *this; }
+  EIGEN_DEVICE_FUNC
+  inline Transform& operator*=(const DiagonalMatrix<Scalar,Dim>& s) { linearExt() *= s; return *this; }
 
   template<typename Derived>
-  inline Transform& operator=(const RotationBase<Derived,Dim>& r);
+  EIGEN_DEVICE_FUNC inline Transform& operator=(const RotationBase<Derived,Dim>& r);
   template<typename Derived>
-  inline Transform& operator*=(const RotationBase<Derived,Dim>& r) { return rotate(r.toRotationMatrix()); }
+  EIGEN_DEVICE_FUNC inline Transform& operator*=(const RotationBase<Derived,Dim>& r) { return rotate(r.toRotationMatrix()); }
   template<typename Derived>
-  inline Transform operator*(const RotationBase<Derived,Dim>& r) const;
+  EIGEN_DEVICE_FUNC inline Transform operator*(const RotationBase<Derived,Dim>& r) const;
 
-  const LinearMatrixType rotation() const;
+  EIGEN_DEVICE_FUNC const LinearMatrixType rotation() const;
   template<typename RotationMatrixType, typename ScalingMatrixType>
+  EIGEN_DEVICE_FUNC
   void computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const;
   template<typename ScalingMatrixType, typename RotationMatrixType>
+  EIGEN_DEVICE_FUNC
   void computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const;
 
   template<typename PositionDerived, typename OrientationType, typename ScaleDerived>
+  EIGEN_DEVICE_FUNC
   Transform& fromPositionOrientationScale(const MatrixBase<PositionDerived> &position,
     const OrientationType& orientation, const MatrixBase<ScaleDerived> &scale);
 
+  EIGEN_DEVICE_FUNC
   inline Transform inverse(TransformTraits traits = (TransformTraits)Mode) const;
 
   /** \returns a const pointer to the column major internal matrix */
-  const Scalar* data() const { return m_matrix.data(); }
+  EIGEN_DEVICE_FUNC const Scalar* data() const { return m_matrix.data(); }
   /** \returns a non-const pointer to the column major internal matrix */
-  Scalar* data() { return m_matrix.data(); }
+  EIGEN_DEVICE_FUNC Scalar* data() { return m_matrix.data(); }
 
   /** \returns \c *this with scalar type casted to \a NewScalarType
     *
@@ -570,12 +629,12 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim,Mode,Options> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim,Mode,Options> >::type cast() const
   { return typename internal::cast_return_type<Transform,Transform<NewScalarType,Dim,Mode,Options> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit Transform(const Transform<OtherScalarType,Dim,Mode,Options>& other)
+  EIGEN_DEVICE_FUNC inline explicit Transform(const Transform<OtherScalarType,Dim,Mode,Options>& other)
   {
     check_template_params();
     m_matrix = other.matrix().template cast<Scalar>();
@@ -585,12 +644,12 @@
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const Transform& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Transform& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_matrix.isApprox(other.m_matrix, prec); }
 
   /** Sets the last row to [0 ... 0 1]
     */
-  void makeAffine()
+  EIGEN_DEVICE_FUNC void makeAffine()
   {
     internal::transform_make_affine<int(Mode)>::run(m_matrix);
   }
@@ -599,26 +658,26 @@
     * \returns the Dim x Dim linear part if the transformation is affine,
     *          and the HDim x Dim part for projective transformations.
     */
-  inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt()
+  EIGEN_DEVICE_FUNC inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt()
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,Dim>(0,0); }
   /** \internal
     * \returns the Dim x Dim linear part if the transformation is affine,
     *          and the HDim x Dim part for projective transformations.
     */
-  inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt() const
+  EIGEN_DEVICE_FUNC inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,Dim> linearExt() const
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,Dim>(0,0); }
 
   /** \internal
     * \returns the translation part if the transformation is affine,
     *          and the last column for projective transformations.
     */
-  inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt()
+  EIGEN_DEVICE_FUNC inline Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt()
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,1>(0,Dim); }
   /** \internal
     * \returns the translation part if the transformation is affine,
     *          and the last column for projective transformations.
     */
-  inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt() const
+  EIGEN_DEVICE_FUNC inline const Block<MatrixType,int(Mode)==int(Projective)?HDim:Dim,1> translationExt() const
   { return m_matrix.template block<int(Mode)==int(Projective)?HDim:Dim,1>(0,Dim); }
 
 
@@ -628,7 +687,7 @@
   
 protected:
   #ifndef EIGEN_PARSED_BY_DOXYGEN
-    static EIGEN_STRONG_INLINE void check_template_params()
+    EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void check_template_params()
     {
       EIGEN_STATIC_ASSERT((Options & (DontAlign|RowMajor)) == Options, INVALID_MATRIX_TEMPLATE_PARAMETERS)
     }
@@ -696,9 +755,13 @@
 Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const QMatrix& other)
 {
   EIGEN_STATIC_ASSERT(Dim==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
-  m_matrix << other.m11(), other.m21(), other.dx(),
-              other.m12(), other.m22(), other.dy(),
-              0, 0, 1;
+  if (Mode == int(AffineCompact))
+    m_matrix << other.m11(), other.m21(), other.dx(),
+                other.m12(), other.m22(), other.dy();
+  else
+    m_matrix << other.m11(), other.m21(), other.dx(),
+                other.m12(), other.m22(), other.dy(),
+                0, 0, 1;
   return *this;
 }
 
@@ -777,7 +840,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::scale(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
@@ -791,7 +854,7 @@
   * \sa prescale(Scalar)
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::scale(const Scalar& s)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::scale(const Scalar& s)
 {
   EIGEN_STATIC_ASSERT(Mode!=int(Isometry), THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS)
   linearExt() *= s;
@@ -804,12 +867,12 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::prescale(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
   EIGEN_STATIC_ASSERT(Mode!=int(Isometry), THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS)
-  m_matrix.template block<Dim,HDim>(0,0).noalias() = (other.asDiagonal() * m_matrix.template block<Dim,HDim>(0,0));
+  affine().noalias() = (other.asDiagonal() * affine());
   return *this;
 }
 
@@ -818,7 +881,7 @@
   * \sa scale(Scalar)
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::prescale(const Scalar& s)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::prescale(const Scalar& s)
 {
   EIGEN_STATIC_ASSERT(Mode!=int(Isometry), THIS_METHOD_IS_ONLY_FOR_SPECIFIC_TRANSFORMATIONS)
   m_matrix.template topRows<Dim>() *= s;
@@ -831,7 +894,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::translate(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
@@ -845,7 +908,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename OtherDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::pretranslate(const MatrixBase<OtherDerived> &other)
 {
   EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(OtherDerived,int(Dim))
@@ -875,7 +938,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename RotationType>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::rotate(const RotationType& rotation)
 {
   linearExt() *= internal::toRotationMatrix<Scalar,Dim>(rotation);
@@ -891,7 +954,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename RotationType>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::prerotate(const RotationType& rotation)
 {
   m_matrix.template block<Dim,HDim>(0,0) = internal::toRotationMatrix<Scalar,Dim>(rotation)
@@ -905,7 +968,7 @@
   * \sa preshear()
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::shear(const Scalar& sx, const Scalar& sy)
 {
   EIGEN_STATIC_ASSERT(int(Dim)==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
@@ -921,7 +984,7 @@
   * \sa shear()
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::preshear(const Scalar& sx, const Scalar& sy)
 {
   EIGEN_STATIC_ASSERT(int(Dim)==2, YOU_MADE_A_PROGRAMMING_MISTAKE)
@@ -935,7 +998,7 @@
 ******************************************************/
 
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const TranslationType& t)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const TranslationType& t)
 {
   linear().setIdentity();
   translation() = t.vector();
@@ -944,7 +1007,7 @@
 }
 
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const TranslationType& t) const
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const TranslationType& t) const
 {
   Transform res = *this;
   res.translate(t.vector());
@@ -952,7 +1015,7 @@
 }
 
 template<typename Scalar, int Dim, int Mode, int Options>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const UniformScaling<Scalar>& s)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const UniformScaling<Scalar>& s)
 {
   m_matrix.setZero();
   linear().diagonal().fill(s.factor());
@@ -962,7 +1025,7 @@
 
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename Derived>
-inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const RotationBase<Derived,Dim>& r)
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options>& Transform<Scalar,Dim,Mode,Options>::operator=(const RotationBase<Derived,Dim>& r)
 {
   linear() = internal::toRotationMatrix<Scalar,Dim>(r);
   translation().setZero();
@@ -972,7 +1035,7 @@
 
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename Derived>
-inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const RotationBase<Derived,Dim>& r) const
+EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode,Options> Transform<Scalar,Dim,Mode,Options>::operator*(const RotationBase<Derived,Dim>& r) const
 {
   Transform res = *this;
   res.rotate(r.derived());
@@ -991,7 +1054,7 @@
   * \sa computeRotationScaling(), computeScalingRotation(), class SVD
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-const typename Transform<Scalar,Dim,Mode,Options>::LinearMatrixType
+EIGEN_DEVICE_FUNC const typename Transform<Scalar,Dim,Mode,Options>::LinearMatrixType
 Transform<Scalar,Dim,Mode,Options>::rotation() const
 {
   LinearMatrixType result;
@@ -1013,7 +1076,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename RotationMatrixType, typename ScalingMatrixType>
-void Transform<Scalar,Dim,Mode,Options>::computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const
+EIGEN_DEVICE_FUNC void Transform<Scalar,Dim,Mode,Options>::computeRotationScaling(RotationMatrixType *rotation, ScalingMatrixType *scaling) const
 {
   JacobiSVD<LinearMatrixType> svd(linear(), ComputeFullU | ComputeFullV);
 
@@ -1029,7 +1092,7 @@
   }
 }
 
-/** decomposes the linear part of the transformation as a product rotation x scaling, the scaling being
+/** decomposes the linear part of the transformation as a product scaling x rotation, the scaling being
   * not necessarily positive.
   *
   * If either pointer is zero, the corresponding computation is skipped.
@@ -1042,7 +1105,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename ScalingMatrixType, typename RotationMatrixType>
-void Transform<Scalar,Dim,Mode,Options>::computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const
+EIGEN_DEVICE_FUNC void Transform<Scalar,Dim,Mode,Options>::computeScalingRotation(ScalingMatrixType *scaling, RotationMatrixType *rotation) const
 {
   JacobiSVD<LinearMatrixType> svd(linear(), ComputeFullU | ComputeFullV);
 
@@ -1063,7 +1126,7 @@
   */
 template<typename Scalar, int Dim, int Mode, int Options>
 template<typename PositionDerived, typename OrientationType, typename ScaleDerived>
-Transform<Scalar,Dim,Mode,Options>&
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>&
 Transform<Scalar,Dim,Mode,Options>::fromPositionOrientationScale(const MatrixBase<PositionDerived> &position,
   const OrientationType& orientation, const MatrixBase<ScaleDerived> &scale)
 {
@@ -1080,7 +1143,7 @@
 struct transform_make_affine
 {
   template<typename MatrixType>
-  static void run(MatrixType &mat)
+  EIGEN_DEVICE_FUNC static void run(MatrixType &mat)
   {
     static const int Dim = MatrixType::ColsAtCompileTime-1;
     mat.template block<1,Dim>(Dim,0).setZero();
@@ -1091,21 +1154,21 @@
 template<>
 struct transform_make_affine<AffineCompact>
 {
-  template<typename MatrixType> static void run(MatrixType &) { }
+  template<typename MatrixType> EIGEN_DEVICE_FUNC static void run(MatrixType &) { }
 };
     
 // selector needed to avoid taking the inverse of a 3x4 matrix
 template<typename TransformType, int Mode=TransformType::Mode>
 struct projective_transform_inverse
 {
-  static inline void run(const TransformType&, TransformType&)
+  EIGEN_DEVICE_FUNC static inline void run(const TransformType&, TransformType&)
   {}
 };
 
 template<typename TransformType>
 struct projective_transform_inverse<TransformType, Projective>
 {
-  static inline void run(const TransformType& m, TransformType& res)
+  EIGEN_DEVICE_FUNC static inline void run(const TransformType& m, TransformType& res)
   {
     res.matrix() = m.matrix().inverse();
   }
@@ -1135,7 +1198,7 @@
   * \sa MatrixBase::inverse()
   */
 template<typename Scalar, int Dim, int Mode, int Options>
-Transform<Scalar,Dim,Mode,Options>
+EIGEN_DEVICE_FUNC Transform<Scalar,Dim,Mode,Options>
 Transform<Scalar,Dim,Mode,Options>::inverse(TransformTraits hint) const
 {
   Transform res;
@@ -1244,8 +1307,8 @@
   };
 };
 
-template< typename TransformType, typename MatrixType >
-struct transform_right_product_impl< TransformType, MatrixType, 0 >
+template< typename TransformType, typename MatrixType, int RhsCols>
+struct transform_right_product_impl< TransformType, MatrixType, 0, RhsCols>
 {
   typedef typename MatrixType::PlainObject ResultType;
 
@@ -1255,8 +1318,8 @@
   }
 };
 
-template< typename TransformType, typename MatrixType >
-struct transform_right_product_impl< TransformType, MatrixType, 1 >
+template< typename TransformType, typename MatrixType, int RhsCols>
+struct transform_right_product_impl< TransformType, MatrixType, 1, RhsCols>
 {
   enum { 
     Dim = TransformType::Dim, 
@@ -1281,8 +1344,8 @@
   }
 };
 
-template< typename TransformType, typename MatrixType >
-struct transform_right_product_impl< TransformType, MatrixType, 2 >
+template< typename TransformType, typename MatrixType, int RhsCols>
+struct transform_right_product_impl< TransformType, MatrixType, 2, RhsCols>
 {
   enum { 
     Dim = TransformType::Dim, 
@@ -1305,6 +1368,30 @@
   }
 };
 
+template< typename TransformType, typename MatrixType >
+struct transform_right_product_impl< TransformType, MatrixType, 2, 1> // rhs is a vector of size Dim
+{
+  typedef typename TransformType::MatrixType TransformMatrix;
+  enum {
+    Dim = TransformType::Dim,
+    HDim = TransformType::HDim,
+    OtherRows = MatrixType::RowsAtCompileTime,
+    WorkingRows = EIGEN_PLAIN_ENUM_MIN(TransformMatrix::RowsAtCompileTime,HDim)
+  };
+
+  typedef typename MatrixType::PlainObject ResultType;
+
+  static EIGEN_STRONG_INLINE ResultType run(const TransformType& T, const MatrixType& other)
+  {
+    EIGEN_STATIC_ASSERT(OtherRows==Dim, YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES);
+
+    Matrix<typename ResultType::Scalar, Dim+1, 1> rhs;
+    rhs.template head<Dim>() = other; rhs[Dim] = typename ResultType::Scalar(1);
+    Matrix<typename ResultType::Scalar, WorkingRows, 1> res(T.matrix() * rhs);
+    return res.template head<Dim>();
+  }
+};
+
 /**********************************************************
 ***   Specializations of operator* with lhs EigenBase   ***
 **********************************************************/
diff --git a/Eigen/src/Geometry/Translation.h b/Eigen/src/Geometry/Translation.h
index 7fda179..51d9a82 100644
--- a/Eigen/src/Geometry/Translation.h
+++ b/Eigen/src/Geometry/Translation.h
@@ -18,8 +18,8 @@
   *
   * \brief Represents a translation transformation
   *
-  * \param _Scalar the scalar type, i.e., the type of the coefficients.
-  * \param _Dim the  dimension of the space, can be a compile time value or Dynamic
+  * \tparam _Scalar the scalar type, i.e., the type of the coefficients.
+  * \tparam _Dim the  dimension of the space, can be a compile time value or Dynamic
   *
   * \note This class is not aimed to be used to store a translation transformation,
   * but rather to make easier the constructions and updates of Transform objects.
@@ -51,16 +51,16 @@
 public:
 
   /** Default constructor without initialization. */
-  Translation() {}
+  EIGEN_DEVICE_FUNC Translation() {}
   /**  */
-  inline Translation(const Scalar& sx, const Scalar& sy)
+  EIGEN_DEVICE_FUNC inline Translation(const Scalar& sx, const Scalar& sy)
   {
     eigen_assert(Dim==2);
     m_coeffs.x() = sx;
     m_coeffs.y() = sy;
   }
   /**  */
-  inline Translation(const Scalar& sx, const Scalar& sy, const Scalar& sz)
+  EIGEN_DEVICE_FUNC inline Translation(const Scalar& sx, const Scalar& sy, const Scalar& sz)
   {
     eigen_assert(Dim==3);
     m_coeffs.x() = sx;
@@ -68,48 +68,48 @@
     m_coeffs.z() = sz;
   }
   /** Constructs and initialize the translation transformation from a vector of translation coefficients */
-  explicit inline Translation(const VectorType& vector) : m_coeffs(vector) {}
+  EIGEN_DEVICE_FUNC explicit inline Translation(const VectorType& vector) : m_coeffs(vector) {}
 
   /** \brief Retruns the x-translation by value. **/
-  inline Scalar x() const { return m_coeffs.x(); }
+  EIGEN_DEVICE_FUNC inline Scalar x() const { return m_coeffs.x(); }
   /** \brief Retruns the y-translation by value. **/
-  inline Scalar y() const { return m_coeffs.y(); }
+  EIGEN_DEVICE_FUNC inline Scalar y() const { return m_coeffs.y(); }
   /** \brief Retruns the z-translation by value. **/
-  inline Scalar z() const { return m_coeffs.z(); }
+  EIGEN_DEVICE_FUNC inline Scalar z() const { return m_coeffs.z(); }
 
   /** \brief Retruns the x-translation as a reference. **/
-  inline Scalar& x() { return m_coeffs.x(); }
+  EIGEN_DEVICE_FUNC inline Scalar& x() { return m_coeffs.x(); }
   /** \brief Retruns the y-translation as a reference. **/
-  inline Scalar& y() { return m_coeffs.y(); }
+  EIGEN_DEVICE_FUNC inline Scalar& y() { return m_coeffs.y(); }
   /** \brief Retruns the z-translation as a reference. **/
-  inline Scalar& z() { return m_coeffs.z(); }
+  EIGEN_DEVICE_FUNC inline Scalar& z() { return m_coeffs.z(); }
 
-  const VectorType& vector() const { return m_coeffs; }
-  VectorType& vector() { return m_coeffs; }
+  EIGEN_DEVICE_FUNC const VectorType& vector() const { return m_coeffs; }
+  EIGEN_DEVICE_FUNC VectorType& vector() { return m_coeffs; }
 
-  const VectorType& translation() const { return m_coeffs; }
-  VectorType& translation() { return m_coeffs; }
+  EIGEN_DEVICE_FUNC const VectorType& translation() const { return m_coeffs; }
+  EIGEN_DEVICE_FUNC VectorType& translation() { return m_coeffs; }
 
   /** Concatenates two translation */
-  inline Translation operator* (const Translation& other) const
+  EIGEN_DEVICE_FUNC inline Translation operator* (const Translation& other) const
   { return Translation(m_coeffs + other.m_coeffs); }
 
   /** Concatenates a translation and a uniform scaling */
-  inline AffineTransformType operator* (const UniformScaling<Scalar>& other) const;
+  EIGEN_DEVICE_FUNC inline AffineTransformType operator* (const UniformScaling<Scalar>& other) const;
 
   /** Concatenates a translation and a linear transformation */
   template<typename OtherDerived>
-  inline AffineTransformType operator* (const EigenBase<OtherDerived>& linear) const;
+  EIGEN_DEVICE_FUNC inline AffineTransformType operator* (const EigenBase<OtherDerived>& linear) const;
 
   /** Concatenates a translation and a rotation */
   template<typename Derived>
-  inline IsometryTransformType operator*(const RotationBase<Derived,Dim>& r) const
+  EIGEN_DEVICE_FUNC inline IsometryTransformType operator*(const RotationBase<Derived,Dim>& r) const
   { return *this * IsometryTransformType(r); }
 
   /** \returns the concatenation of a linear transformation \a l with the translation \a t */
   // its a nightmare to define a templated friend function outside its declaration
   template<typename OtherDerived> friend
-  inline AffineTransformType operator*(const EigenBase<OtherDerived>& linear, const Translation& t)
+  EIGEN_DEVICE_FUNC inline AffineTransformType operator*(const EigenBase<OtherDerived>& linear, const Translation& t)
   {
     AffineTransformType res;
     res.matrix().setZero();
@@ -122,7 +122,7 @@
 
   /** Concatenates a translation and a transformation */
   template<int Mode, int Options>
-  inline Transform<Scalar,Dim,Mode> operator* (const Transform<Scalar,Dim,Mode,Options>& t) const
+  EIGEN_DEVICE_FUNC inline Transform<Scalar,Dim,Mode> operator* (const Transform<Scalar,Dim,Mode,Options>& t) const
   {
     Transform<Scalar,Dim,Mode> res = t;
     res.pretranslate(m_coeffs);
@@ -130,8 +130,10 @@
   }
 
   /** Applies translation to vector */
-  inline VectorType operator* (const VectorType& other) const
-  { return m_coeffs + other; }
+  template<typename Derived>
+  inline typename internal::enable_if<Derived::IsVectorAtCompileTime,VectorType>::type
+  operator* (const MatrixBase<Derived>& vec) const
+  { return m_coeffs + vec.derived(); }
 
   /** \returns the inverse translation (opposite) */
   Translation inverse() const { return Translation(-m_coeffs); }
@@ -150,19 +152,19 @@
     * then this function smartly returns a const reference to \c *this.
     */
   template<typename NewScalarType>
-  inline typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type cast() const
+  EIGEN_DEVICE_FUNC inline typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type cast() const
   { return typename internal::cast_return_type<Translation,Translation<NewScalarType,Dim> >::type(*this); }
 
   /** Copy constructor with scalar type conversion */
   template<typename OtherScalarType>
-  inline explicit Translation(const Translation<OtherScalarType,Dim>& other)
+  EIGEN_DEVICE_FUNC inline explicit Translation(const Translation<OtherScalarType,Dim>& other)
   { m_coeffs = other.vector().template cast<Scalar>(); }
 
   /** \returns \c true if \c *this is approximately equal to \a other, within the precision
     * determined by \a prec.
     *
     * \sa MatrixBase::isApprox() */
-  bool isApprox(const Translation& other, typename NumTraits<Scalar>::Real prec = NumTraits<Scalar>::dummy_precision()) const
+  EIGEN_DEVICE_FUNC bool isApprox(const Translation& other, const typename NumTraits<Scalar>::Real& prec = NumTraits<Scalar>::dummy_precision()) const
   { return m_coeffs.isApprox(other.m_coeffs, prec); }
 
 };
@@ -176,7 +178,7 @@
 //@}
 
 template<typename Scalar, int Dim>
-inline typename Translation<Scalar,Dim>::AffineTransformType
+EIGEN_DEVICE_FUNC inline typename Translation<Scalar,Dim>::AffineTransformType
 Translation<Scalar,Dim>::operator* (const UniformScaling<Scalar>& other) const
 {
   AffineTransformType res;
@@ -189,7 +191,7 @@
 
 template<typename Scalar, int Dim>
 template<typename OtherDerived>
-inline typename Translation<Scalar,Dim>::AffineTransformType
+EIGEN_DEVICE_FUNC inline typename Translation<Scalar,Dim>::AffineTransformType
 Translation<Scalar,Dim>::operator* (const EigenBase<OtherDerived>& linear) const
 {
   AffineTransformType res;
diff --git a/Eigen/src/Geometry/Umeyama.h b/Eigen/src/Geometry/Umeyama.h
index 5e20662..7e933fc 100644
--- a/Eigen/src/Geometry/Umeyama.h
+++ b/Eigen/src/Geometry/Umeyama.h
@@ -97,7 +97,6 @@
   typedef typename internal::umeyama_transform_matrix_type<Derived, OtherDerived>::type TransformationMatrixType;
   typedef typename internal::traits<TransformationMatrixType>::Scalar Scalar;
   typedef typename NumTraits<Scalar>::Real RealScalar;
-  typedef typename Derived::Index Index;
 
   EIGEN_STATIC_ASSERT(!NumTraits<Scalar>::IsComplex, NUMERIC_TYPE_MUST_BE_REAL)
   EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename internal::traits<OtherDerived>::Scalar>::value),
@@ -136,22 +135,12 @@
 
   // Eq. (39)
   VectorType S = VectorType::Ones(m);
-  if (sigma.determinant()<Scalar(0)) S(m-1) = Scalar(-1);
+
+  if  ( svd.matrixU().determinant() * svd.matrixV().determinant() < 0 )
+    S(m-1) = -1;
 
   // Eq. (40) and (43)
-  const VectorType& d = svd.singularValues();
-  Index rank = 0; for (Index i=0; i<m; ++i) if (!internal::isMuchSmallerThan(d.coeff(i),d.coeff(0))) ++rank;
-  if (rank == m-1) {
-    if ( svd.matrixU().determinant() * svd.matrixV().determinant() > Scalar(0) ) {
-      Rt.block(0,0,m,m).noalias() = svd.matrixU()*svd.matrixV().transpose();
-    } else {
-      const Scalar s = S(m-1); S(m-1) = Scalar(-1);
-      Rt.block(0,0,m,m).noalias() = svd.matrixU() * S.asDiagonal() * svd.matrixV().transpose();
-      S(m-1) = s;
-    }
-  } else {
-    Rt.block(0,0,m,m).noalias() = svd.matrixU() * S.asDiagonal() * svd.matrixV().transpose();
-  }
+  Rt.block(0,0,m,m).noalias() = svd.matrixU() * S.asDiagonal() * svd.matrixV().transpose();
 
   if (with_scaling)
   {
diff --git a/Eigen/src/Geometry/arch/CMakeLists.txt b/Eigen/src/Geometry/arch/CMakeLists.txt
deleted file mode 100644
index 1267a79..0000000
--- a/Eigen/src/Geometry/arch/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Geometry_arch_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Geometry_arch_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Geometry/arch COMPONENT Devel
-  )
diff --git a/Eigen/src/Geometry/arch/Geometry_SSE.h b/Eigen/src/Geometry/arch/Geometry_SSE.h
index 3d8284f..f68cab5 100644
--- a/Eigen/src/Geometry/arch/Geometry_SSE.h
+++ b/Eigen/src/Geometry/arch/Geometry_SSE.h
@@ -16,39 +16,63 @@
 namespace internal {
 
 template<class Derived, class OtherDerived>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, float, Aligned>
+struct quat_product<Architecture::SSE, Derived, OtherDerived, float>
 {
+  enum {
+    AAlignment = traits<Derived>::Alignment,
+    BAlignment = traits<OtherDerived>::Alignment,
+    ResAlignment = traits<Quaternion<float> >::Alignment
+  };
   static inline Quaternion<float> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
   {
-    const __m128 mask = _mm_castsi128_ps(_mm_setr_epi32(0,0,0,0x80000000));
     Quaternion<float> res;
-    __m128 a = _a.coeffs().template packet<Aligned>(0);
-    __m128 b = _b.coeffs().template packet<Aligned>(0);
-    __m128 flip1 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),
-                                         vec4f_swizzle1(b,2,0,1,2)),mask);
-    __m128 flip2 = _mm_xor_ps(_mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),
-                                         vec4f_swizzle1(b,0,1,2,1)),mask);
-    pstore(&res.x(),
+    const __m128 mask = _mm_setr_ps(0.f,0.f,0.f,-0.f);
+    __m128 a = _a.coeffs().template packet<AAlignment>(0);
+    __m128 b = _b.coeffs().template packet<BAlignment>(0);
+    __m128 s1 = _mm_mul_ps(vec4f_swizzle1(a,1,2,0,2),vec4f_swizzle1(b,2,0,1,2));
+    __m128 s2 = _mm_mul_ps(vec4f_swizzle1(a,3,3,3,1),vec4f_swizzle1(b,0,1,2,1));
+    pstoret<float,Packet4f,ResAlignment>(
+              &res.x(),
               _mm_add_ps(_mm_sub_ps(_mm_mul_ps(a,vec4f_swizzle1(b,3,3,3,3)),
                                     _mm_mul_ps(vec4f_swizzle1(a,2,0,1,0),
                                                vec4f_swizzle1(b,1,2,0,0))),
-                         _mm_add_ps(flip1,flip2)));
+                         _mm_xor_ps(mask,_mm_add_ps(s1,s2))));
+    
     return res;
   }
 };
 
+template<class Derived>
+struct quat_conj<Architecture::SSE, Derived, float>
+{
+  enum {
+    ResAlignment = traits<Quaternion<float> >::Alignment
+  };
+  static inline Quaternion<float> run(const QuaternionBase<Derived>& q)
+  {
+    Quaternion<float> res;
+    const __m128 mask = _mm_setr_ps(-0.f,-0.f,-0.f,0.f);
+    pstoret<float,Packet4f,ResAlignment>(&res.x(), _mm_xor_ps(mask, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
+    return res;
+  }
+};
+
+
 template<typename VectorLhs,typename VectorRhs>
 struct cross3_impl<Architecture::SSE,VectorLhs,VectorRhs,float,true>
 {
+  enum {
+    ResAlignment = traits<typename plain_matrix_type<VectorLhs>::type>::Alignment
+  };
   static inline typename plain_matrix_type<VectorLhs>::type
   run(const VectorLhs& lhs, const VectorRhs& rhs)
   {
-    __m128 a = lhs.template packet<VectorLhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
-    __m128 b = rhs.template packet<VectorRhs::Flags&AlignedBit ? Aligned : Unaligned>(0);
+    __m128 a = lhs.template packet<traits<VectorLhs>::Alignment>(0);
+    __m128 b = rhs.template packet<traits<VectorRhs>::Alignment>(0);
     __m128 mul1=_mm_mul_ps(vec4f_swizzle1(a,1,2,0,3),vec4f_swizzle1(b,2,0,1,3));
     __m128 mul2=_mm_mul_ps(vec4f_swizzle1(a,2,0,1,3),vec4f_swizzle1(b,1,2,0,3));
     typename plain_matrix_type<VectorLhs>::type res;
-    pstore(&res.x(),_mm_sub_ps(mul1,mul2));
+    pstoret<float,Packet4f,ResAlignment>(&res.x(),_mm_sub_ps(mul1,mul2));
     return res;
   }
 };
@@ -57,8 +81,13 @@
 
 
 template<class Derived, class OtherDerived>
-struct quat_product<Architecture::SSE, Derived, OtherDerived, double, Aligned>
+struct quat_product<Architecture::SSE, Derived, OtherDerived, double>
 {
+  enum {
+    BAlignment = traits<OtherDerived>::Alignment,
+    ResAlignment = traits<Quaternion<double> >::Alignment
+  };
+
   static inline Quaternion<double> run(const QuaternionBase<Derived>& _a, const QuaternionBase<OtherDerived>& _b)
   {
   const Packet2d mask = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
@@ -66,8 +95,8 @@
   Quaternion<double> res;
 
   const double* a = _a.coeffs().data();
-  Packet2d b_xy = _b.coeffs().template packet<Aligned>(0);
-  Packet2d b_zw = _b.coeffs().template packet<Aligned>(2);
+  Packet2d b_xy = _b.coeffs().template packet<BAlignment>(0);
+  Packet2d b_zw = _b.coeffs().template packet<BAlignment>(2);
   Packet2d a_xx = pset1<Packet2d>(a[0]);
   Packet2d a_yy = pset1<Packet2d>(a[1]);
   Packet2d a_zz = pset1<Packet2d>(a[2]);
@@ -85,9 +114,9 @@
   t2 = psub(pmul(a_zz, b_xy), pmul(a_xx, b_zw));
 #ifdef EIGEN_VECTORIZE_SSE3
   EIGEN_UNUSED_VARIABLE(mask)
-  pstore(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
+  pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_addsub_pd(t1, preverse(t2)));
 #else
-  pstore(&res.x(), padd(t1, pxor(mask,preverse(t2))));
+  pstoret<double,Packet2d,ResAlignment>(&res.x(), padd(t1, pxor(mask,preverse(t2))));
 #endif
   
   /*
@@ -99,15 +128,32 @@
   t2 = padd(pmul(a_zz, b_zw), pmul(a_xx, b_xy));
 #ifdef EIGEN_VECTORIZE_SSE3
   EIGEN_UNUSED_VARIABLE(mask)
-  pstore(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
+  pstoret<double,Packet2d,ResAlignment>(&res.z(), preverse(_mm_addsub_pd(preverse(t1), t2)));
 #else
-  pstore(&res.z(), psub(t1, pxor(mask,preverse(t2))));
+  pstoret<double,Packet2d,ResAlignment>(&res.z(), psub(t1, pxor(mask,preverse(t2))));
 #endif
 
   return res;
 }
 };
 
+template<class Derived>
+struct quat_conj<Architecture::SSE, Derived, double>
+{
+  enum {
+    ResAlignment = traits<Quaternion<double> >::Alignment
+  };
+  static inline Quaternion<double> run(const QuaternionBase<Derived>& q)
+  {
+    Quaternion<double> res;
+    const __m128d mask0 = _mm_setr_pd(-0.,-0.);
+    const __m128d mask2 = _mm_setr_pd(-0.,0.);
+    pstoret<double,Packet2d,ResAlignment>(&res.x(), _mm_xor_pd(mask0, q.coeffs().template packet<traits<Derived>::Alignment>(0)));
+    pstoret<double,Packet2d,ResAlignment>(&res.z(), _mm_xor_pd(mask2, q.coeffs().template packet<traits<Derived>::Alignment>(2)));
+    return res;
+  }
+};
+
 } // end namespace internal
 
 } // end namespace Eigen
diff --git a/Eigen/src/Householder/BlockHouseholder.h b/Eigen/src/Householder/BlockHouseholder.h
index 60dbea5..01a7ed1 100644
--- a/Eigen/src/Householder/BlockHouseholder.h
+++ b/Eigen/src/Householder/BlockHouseholder.h
@@ -16,48 +16,83 @@
 namespace Eigen { 
 
 namespace internal {
+  
+/** \internal */
+// template<typename TriangularFactorType,typename VectorsType,typename CoeffsType>
+// void make_block_householder_triangular_factor(TriangularFactorType& triFactor, const VectorsType& vectors, const CoeffsType& hCoeffs)
+// {
+//   typedef typename VectorsType::Scalar Scalar;
+//   const Index nbVecs = vectors.cols();
+//   eigen_assert(triFactor.rows() == nbVecs && triFactor.cols() == nbVecs && vectors.rows()>=nbVecs);
+// 
+//   for(Index i = 0; i < nbVecs; i++)
+//   {
+//     Index rs = vectors.rows() - i;
+//     // Warning, note that hCoeffs may alias with vectors.
+//     // It is then necessary to copy it before modifying vectors(i,i). 
+//     typename CoeffsType::Scalar h = hCoeffs(i);
+//     // This hack permits to pass trough nested Block<> and Transpose<> expressions.
+//     Scalar *Vii_ptr = const_cast<Scalar*>(vectors.data() + vectors.outerStride()*i + vectors.innerStride()*i);
+//     Scalar Vii = *Vii_ptr;
+//     *Vii_ptr = Scalar(1);
+//     triFactor.col(i).head(i).noalias() = -h * vectors.block(i, 0, rs, i).adjoint()
+//                                        * vectors.col(i).tail(rs);
+//     *Vii_ptr = Vii;
+//     // FIXME add .noalias() once the triangular product can work inplace
+//     triFactor.col(i).head(i) = triFactor.block(0,0,i,i).template triangularView<Upper>()
+//                              * triFactor.col(i).head(i);
+//     triFactor(i,i) = hCoeffs(i);
+//   }
+// }
 
 /** \internal */
+// This variant avoid modifications in vectors
 template<typename TriangularFactorType,typename VectorsType,typename CoeffsType>
 void make_block_householder_triangular_factor(TriangularFactorType& triFactor, const VectorsType& vectors, const CoeffsType& hCoeffs)
 {
-  typedef typename TriangularFactorType::Index Index;
-  typedef typename VectorsType::Scalar Scalar;
   const Index nbVecs = vectors.cols();
   eigen_assert(triFactor.rows() == nbVecs && triFactor.cols() == nbVecs && vectors.rows()>=nbVecs);
 
-  for(Index i = 0; i < nbVecs; i++)
+  for(Index i = nbVecs-1; i >=0 ; --i)
   {
-    Index rs = vectors.rows() - i;
-    Scalar Vii = vectors(i,i);
-    vectors.const_cast_derived().coeffRef(i,i) = Scalar(1);
-    triFactor.col(i).head(i).noalias() = -hCoeffs(i) * vectors.block(i, 0, rs, i).adjoint()
-                                       * vectors.col(i).tail(rs);
-    vectors.const_cast_derived().coeffRef(i, i) = Vii;
-    // FIXME add .noalias() once the triangular product can work inplace
-    triFactor.col(i).head(i) = triFactor.block(0,0,i,i).template triangularView<Upper>()
-                             * triFactor.col(i).head(i);
+    Index rs = vectors.rows() - i - 1;
+    Index rt = nbVecs-i-1;
+
+    if(rt>0)
+    {
+      triFactor.row(i).tail(rt).noalias() = -hCoeffs(i) * vectors.col(i).tail(rs).adjoint()
+                                                        * vectors.bottomRightCorner(rs, rt).template triangularView<UnitLower>();
+            
+      // FIXME add .noalias() once the triangular product can work inplace
+      triFactor.row(i).tail(rt) = triFactor.row(i).tail(rt) * triFactor.bottomRightCorner(rt,rt).template triangularView<Upper>();
+      
+    }
     triFactor(i,i) = hCoeffs(i);
   }
 }
 
-/** \internal */
+/** \internal
+  * if forward then perform   mat = H0 * H1 * H2 * mat
+  * otherwise perform         mat = H2 * H1 * H0 * mat
+  */
 template<typename MatrixType,typename VectorsType,typename CoeffsType>
-void apply_block_householder_on_the_left(MatrixType& mat, const VectorsType& vectors, const CoeffsType& hCoeffs)
+void apply_block_householder_on_the_left(MatrixType& mat, const VectorsType& vectors, const CoeffsType& hCoeffs, bool forward)
 {
-  typedef typename MatrixType::Index Index;
   enum { TFactorSize = MatrixType::ColsAtCompileTime };
   Index nbVecs = vectors.cols();
-  Matrix<typename MatrixType::Scalar, TFactorSize, TFactorSize, ColMajor> T(nbVecs,nbVecs);
-  make_block_householder_triangular_factor(T, vectors, hCoeffs);
-
-  const TriangularView<const VectorsType, UnitLower>& V(vectors);
+  Matrix<typename MatrixType::Scalar, TFactorSize, TFactorSize, RowMajor> T(nbVecs,nbVecs);
+  
+  if(forward) make_block_householder_triangular_factor(T, vectors, hCoeffs);
+  else        make_block_householder_triangular_factor(T, vectors, hCoeffs.conjugate());  
+  const TriangularView<const VectorsType, UnitLower> V(vectors);
 
   // A -= V T V^* A
-  Matrix<typename MatrixType::Scalar,VectorsType::ColsAtCompileTime,MatrixType::ColsAtCompileTime,0,
+  Matrix<typename MatrixType::Scalar,VectorsType::ColsAtCompileTime,MatrixType::ColsAtCompileTime,
+         (VectorsType::MaxColsAtCompileTime==1 && MatrixType::MaxColsAtCompileTime!=1)?RowMajor:ColMajor,
          VectorsType::MaxColsAtCompileTime,MatrixType::MaxColsAtCompileTime> tmp = V.adjoint() * mat;
   // FIXME add .noalias() once the triangular product can work inplace
-  tmp = T.template triangularView<Upper>().adjoint() * tmp;
+  if(forward) tmp = T.template triangularView<Upper>()           * tmp;
+  else        tmp = T.template triangularView<Upper>().adjoint() * tmp;
   mat.noalias() -= V * tmp;
 }
 
diff --git a/Eigen/src/Householder/CMakeLists.txt b/Eigen/src/Householder/CMakeLists.txt
deleted file mode 100644
index ce4937d..0000000
--- a/Eigen/src/Householder/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Householder_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Householder_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Householder COMPONENT Devel
-  )
diff --git a/Eigen/src/Householder/Householder.h b/Eigen/src/Householder/Householder.h
index 32112af..80de2c3 100644
--- a/Eigen/src/Householder/Householder.h
+++ b/Eigen/src/Householder/Householder.h
@@ -75,8 +75,9 @@
   
   RealScalar tailSqNorm = size()==1 ? RealScalar(0) : tail.squaredNorm();
   Scalar c0 = coeff(0);
+  const RealScalar tol = (std::numeric_limits<RealScalar>::min)();
 
-  if(tailSqNorm == RealScalar(0) && numext::imag(c0)==RealScalar(0))
+  if(tailSqNorm <= tol && numext::abs2(numext::imag(c0))<=tol)
   {
     tau = RealScalar(0);
     beta = numext::real(c0);
@@ -118,7 +119,7 @@
   {
     *this *= Scalar(1)-tau;
   }
-  else
+  else if(tau!=Scalar(0))
   {
     Map<typename internal::plain_row_type<PlainObject>::type> tmp(workspace,cols());
     Block<Derived, EssentialPart::SizeAtCompileTime, Derived::ColsAtCompileTime> bottom(derived(), 1, 0, rows()-1, cols());
@@ -155,7 +156,7 @@
   {
     *this *= Scalar(1)-tau;
   }
-  else
+  else if(tau!=Scalar(0))
   {
     Map<typename internal::plain_col_type<PlainObject>::type> tmp(workspace,rows());
     Block<Derived, Derived::RowsAtCompileTime, EssentialPart::SizeAtCompileTime> right(derived(), 0, 1, rows(), cols()-1);
diff --git a/Eigen/src/Householder/HouseholderSequence.h b/Eigen/src/Householder/HouseholderSequence.h
index d800ca1..3ce0a69 100644
--- a/Eigen/src/Householder/HouseholderSequence.h
+++ b/Eigen/src/Householder/HouseholderSequence.h
@@ -60,7 +60,7 @@
 struct traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
 {
   typedef typename VectorsType::Scalar Scalar;
-  typedef typename VectorsType::Index Index;
+  typedef typename VectorsType::StorageIndex StorageIndex;
   typedef typename VectorsType::StorageKind StorageKind;
   enum {
     RowsAtCompileTime = Side==OnTheLeft ? traits<VectorsType>::RowsAtCompileTime
@@ -73,12 +73,20 @@
   };
 };
 
+struct HouseholderSequenceShape {};
+
+template<typename VectorsType, typename CoeffsType, int Side>
+struct evaluator_traits<HouseholderSequence<VectorsType,CoeffsType,Side> >
+  : public evaluator_traits_base<HouseholderSequence<VectorsType,CoeffsType,Side> >
+{
+  typedef HouseholderSequenceShape Shape;
+};
+
 template<typename VectorsType, typename CoeffsType, int Side>
 struct hseq_side_dependent_impl
 {
   typedef Block<const VectorsType, Dynamic, 1> EssentialVectorType;
   typedef HouseholderSequence<VectorsType, CoeffsType, OnTheLeft> HouseholderSequenceType;
-  typedef typename VectorsType::Index Index;
   static inline const EssentialVectorType essentialVector(const HouseholderSequenceType& h, Index k)
   {
     Index start = k+1+h.m_shift;
@@ -91,7 +99,6 @@
 {
   typedef Transpose<Block<const VectorsType, 1, Dynamic> > EssentialVectorType;
   typedef HouseholderSequence<VectorsType, CoeffsType, OnTheRight> HouseholderSequenceType;
-  typedef typename VectorsType::Index Index;
   static inline const EssentialVectorType essentialVector(const HouseholderSequenceType& h, Index k)
   {
     Index start = k+1+h.m_shift;
@@ -101,7 +108,7 @@
 
 template<typename OtherScalarType, typename MatrixType> struct matrix_type_times_scalar_type
 {
-  typedef typename scalar_product_traits<OtherScalarType, typename MatrixType::Scalar>::ReturnType
+  typedef typename ScalarBinaryOpTraits<OtherScalarType, typename MatrixType::Scalar>::ReturnType
     ResultScalar;
   typedef Matrix<ResultScalar, MatrixType::RowsAtCompileTime, MatrixType::ColsAtCompileTime,
                  0, MatrixType::MaxRowsAtCompileTime, MatrixType::MaxColsAtCompileTime> Type;
@@ -122,7 +129,6 @@
       MaxColsAtCompileTime = internal::traits<HouseholderSequence>::MaxColsAtCompileTime
     };
     typedef typename internal::traits<HouseholderSequence>::Scalar Scalar;
-    typedef typename VectorsType::Index Index;
 
     typedef HouseholderSequence<
       typename internal::conditional<NumTraits<Scalar>::IsComplex,
@@ -237,8 +243,7 @@
     {
       workspace.resize(rows());
       Index vecs = m_length;
-      if(    internal::is_same<typename internal::remove_all<VectorsType>::type,Dest>::value
-          && internal::extract_data(dst) == internal::extract_data(m_vectors))
+      if(internal::is_same_dense(dst,m_vectors))
       {
         // in-place
         dst.diagonal().setOnes();
@@ -299,7 +304,7 @@
     /** \internal */
     template<typename Dest> inline void applyThisOnTheLeft(Dest& dst) const
     {
-      Matrix<Scalar,1,Dest::ColsAtCompileTime,RowMajor,1,Dest::MaxColsAtCompileTime> workspace(dst.cols());
+      Matrix<Scalar,1,Dest::ColsAtCompileTime,RowMajor,1,Dest::MaxColsAtCompileTime> workspace;
       applyThisOnTheLeft(dst, workspace);
     }
 
@@ -307,12 +312,36 @@
     template<typename Dest, typename Workspace>
     inline void applyThisOnTheLeft(Dest& dst, Workspace& workspace) const
     {
-      workspace.resize(dst.cols());
-      for(Index k = 0; k < m_length; ++k)
+      const Index BlockSize = 48;
+      // if the entries are large enough, then apply the reflectors by block
+      if(m_length>=BlockSize && dst.cols()>1)
       {
-        Index actual_k = m_trans ? k : m_length-k-1;
-        dst.bottomRows(rows()-m_shift-actual_k)
-           .applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
+        for(Index i = 0; i < m_length; i+=BlockSize)
+        {
+          Index end = m_trans ? (std::min)(m_length,i+BlockSize) : m_length-i;
+          Index k = m_trans ? i : (std::max)(Index(0),end-BlockSize);
+          Index bs = end-k;
+          Index start = k + m_shift;
+          
+          typedef Block<typename internal::remove_all<VectorsType>::type,Dynamic,Dynamic> SubVectorsType;
+          SubVectorsType sub_vecs1(m_vectors.const_cast_derived(), Side==OnTheRight ? k : start,
+                                                                   Side==OnTheRight ? start : k,
+                                                                   Side==OnTheRight ? bs : m_vectors.rows()-start,
+                                                                   Side==OnTheRight ? m_vectors.cols()-start : bs);
+          typename internal::conditional<Side==OnTheRight, Transpose<SubVectorsType>, SubVectorsType&>::type sub_vecs(sub_vecs1);
+          Block<Dest,Dynamic,Dynamic> sub_dst(dst,dst.rows()-rows()+m_shift+k,0, rows()-m_shift-k,dst.cols());
+          apply_block_householder_on_the_left(sub_dst, sub_vecs, m_coeffs.segment(k, bs), !m_trans);
+        }
+      }
+      else
+      {
+        workspace.resize(dst.cols());
+        for(Index k = 0; k < m_length; ++k)
+        {
+          Index actual_k = m_trans ? k : m_length-k-1;
+          dst.bottomRows(rows()-m_shift-actual_k)
+            .applyHouseholderOnTheLeft(essentialVector(actual_k), m_coeffs.coeff(actual_k), workspace.data());
+        }
       }
     }
 
diff --git a/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h b/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h
index 73ca9bf..f66c846 100644
--- a/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h
+++ b/Eigen/src/IterativeLinearSolvers/BasicPreconditioners.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -17,33 +17,37 @@
   *
   * This class allows to approximately solve for A.x = b problems assuming A is a diagonal matrix.
   * In other words, this preconditioner neglects all off diagonal entries and, in Eigen's language, solves for:
-  * \code
-  * A.diagonal().asDiagonal() . x = b
-  * \endcode
+    \code
+    A.diagonal().asDiagonal() . x = b
+    \endcode
   *
   * \tparam _Scalar the type of the scalar.
   *
+  * \implsparsesolverconcept
+  *
   * This preconditioner is suitable for both selfadjoint and general problems.
   * The diagonal entries are pre-inverted and stored into a dense vector.
   *
   * \note A variant that has yet to be implemented would attempt to preserve the norm of each column.
   *
+  * \sa class LeastSquareDiagonalPreconditioner, class ConjugateGradient
   */
 template <typename _Scalar>
 class DiagonalPreconditioner
 {
     typedef _Scalar Scalar;
     typedef Matrix<Scalar,Dynamic,1> Vector;
-    typedef typename Vector::Index Index;
-
   public:
-    // this typedef is only to export the scalar type and compile-time dimensions to solve_retval
-    typedef Matrix<Scalar,Dynamic,Dynamic> MatrixType;
+    typedef typename Vector::StorageIndex StorageIndex;
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
 
     DiagonalPreconditioner() : m_isInitialized(false) {}
 
     template<typename MatType>
-    DiagonalPreconditioner(const MatType& mat) : m_invdiag(mat.cols())
+    explicit DiagonalPreconditioner(const MatType& mat) : m_invdiag(mat.cols())
     {
       compute(mat);
     }
@@ -65,10 +69,10 @@
       {
         typename MatType::InnerIterator it(mat,j);
         while(it && it.index()!=j) ++it;
-        if(it && it.index()==j)
+        if(it && it.index()==j && it.value()!=Scalar(0))
           m_invdiag(j) = Scalar(1)/it.value();
         else
-          m_invdiag(j) = 0;
+          m_invdiag(j) = Scalar(1);
       }
       m_isInitialized = true;
       return *this;
@@ -80,46 +84,117 @@
       return factorize(mat);
     }
 
+    /** \internal */
     template<typename Rhs, typename Dest>
-    void _solve(const Rhs& b, Dest& x) const
+    void _solve_impl(const Rhs& b, Dest& x) const
     {
       x = m_invdiag.array() * b.array() ;
     }
 
-    template<typename Rhs> inline const internal::solve_retval<DiagonalPreconditioner, Rhs>
+    template<typename Rhs> inline const Solve<DiagonalPreconditioner, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "DiagonalPreconditioner is not initialized.");
       eigen_assert(m_invdiag.size()==b.rows()
                 && "DiagonalPreconditioner::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<DiagonalPreconditioner, Rhs>(*this, b.derived());
+      return Solve<DiagonalPreconditioner, Rhs>(*this, b.derived());
     }
+    
+    ComputationInfo info() { return Success; }
 
   protected:
     Vector m_invdiag;
     bool m_isInitialized;
 };
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<DiagonalPreconditioner<_MatrixType>, Rhs>
-  : solve_retval_base<DiagonalPreconditioner<_MatrixType>, Rhs>
+/** \ingroup IterativeLinearSolvers_Module
+  * \brief Jacobi preconditioner for LeastSquaresConjugateGradient
+  *
+  * This class allows to approximately solve for A' A x  = A' b problems assuming A' A is a diagonal matrix.
+  * In other words, this preconditioner neglects all off diagonal entries and, in Eigen's language, solves for:
+    \code
+    (A.adjoint() * A).diagonal().asDiagonal() * x = b
+    \endcode
+  *
+  * \tparam _Scalar the type of the scalar.
+  *
+  * \implsparsesolverconcept
+  *
+  * The diagonal entries are pre-inverted and stored into a dense vector.
+  * 
+  * \sa class LeastSquaresConjugateGradient, class DiagonalPreconditioner
+  */
+template <typename _Scalar>
+class LeastSquareDiagonalPreconditioner : public DiagonalPreconditioner<_Scalar>
 {
-  typedef DiagonalPreconditioner<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
+    typedef _Scalar Scalar;
+    typedef typename NumTraits<Scalar>::Real RealScalar;
+    typedef DiagonalPreconditioner<_Scalar> Base;
+    using Base::m_invdiag;
+  public:
 
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
+    LeastSquareDiagonalPreconditioner() : Base() {}
+
+    template<typename MatType>
+    explicit LeastSquareDiagonalPreconditioner(const MatType& mat) : Base()
+    {
+      compute(mat);
+    }
+
+    template<typename MatType>
+    LeastSquareDiagonalPreconditioner& analyzePattern(const MatType& )
+    {
+      return *this;
+    }
+    
+    template<typename MatType>
+    LeastSquareDiagonalPreconditioner& factorize(const MatType& mat)
+    {
+      // Compute the inverse squared-norm of each column of mat
+      m_invdiag.resize(mat.cols());
+      if(MatType::IsRowMajor)
+      {
+        m_invdiag.setZero();
+        for(Index j=0; j<mat.outerSize(); ++j)
+        {
+          for(typename MatType::InnerIterator it(mat,j); it; ++it)
+            m_invdiag(it.index()) += numext::abs2(it.value());
+        }
+        for(Index j=0; j<mat.cols(); ++j)
+          if(numext::real(m_invdiag(j))>RealScalar(0))
+            m_invdiag(j) = RealScalar(1)/numext::real(m_invdiag(j));
+      }
+      else
+      {
+        for(Index j=0; j<mat.outerSize(); ++j)
+        {
+          RealScalar sum = mat.col(j).squaredNorm();
+          if(sum>RealScalar(0))
+            m_invdiag(j) = RealScalar(1)/sum;
+          else
+            m_invdiag(j) = RealScalar(1);
+        }
+      }
+      Base::m_isInitialized = true;
+      return *this;
+    }
+    
+    template<typename MatType>
+    LeastSquareDiagonalPreconditioner& compute(const MatType& mat)
+    {
+      return factorize(mat);
+    }
+    
+    ComputationInfo info() { return Success; }
+
+  protected:
 };
 
-}
-
 /** \ingroup IterativeLinearSolvers_Module
   * \brief A naive preconditioner which approximates any matrix as the identity matrix
   *
+  * \implsparsesolverconcept
+  *
   * \sa class DiagonalPreconditioner
   */
 class IdentityPreconditioner
@@ -129,7 +204,7 @@
     IdentityPreconditioner() {}
 
     template<typename MatrixType>
-    IdentityPreconditioner(const MatrixType& ) {}
+    explicit IdentityPreconditioner(const MatrixType& ) {}
     
     template<typename MatrixType>
     IdentityPreconditioner& analyzePattern(const MatrixType& ) { return *this; }
@@ -142,6 +217,8 @@
     
     template<typename Rhs>
     inline const Rhs& solve(const Rhs& b) const { return b; }
+    
+    ComputationInfo info() { return Success; }
 };
 
 } // end namespace Eigen
diff --git a/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h b/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h
index 2625c4d..454f468 100644
--- a/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h
+++ b/Eigen/src/IterativeLinearSolvers/BiCGSTAB.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -27,7 +27,7 @@
   */
 template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
 bool bicgstab(const MatrixType& mat, const Rhs& rhs, Dest& x,
-              const Preconditioner& precond, int& iters,
+              const Preconditioner& precond, Index& iters,
               typename Dest::RealScalar& tol_error)
 {
   using std::sqrt;
@@ -36,9 +36,9 @@
   typedef typename Dest::Scalar Scalar;
   typedef Matrix<Scalar,Dynamic,1> VectorType;
   RealScalar tol = tol_error;
-  int maxIters = iters;
+  Index maxIters = iters;
 
-  int n = mat.cols();
+  Index n = mat.cols();
   VectorType r  = rhs - mat * x;
   VectorType r0 = r;
   
@@ -59,20 +59,21 @@
 
   VectorType s(n), t(n);
 
-  RealScalar tol2 = tol*tol;
+  RealScalar tol2 = tol*tol*rhs_sqnorm;
   RealScalar eps2 = NumTraits<Scalar>::epsilon()*NumTraits<Scalar>::epsilon();
-  int i = 0;
-  int restarts = 0;
+  Index i = 0;
+  Index restarts = 0;
 
-  while ( r.squaredNorm()/rhs_sqnorm > tol2 && i<maxIters )
+  while ( r.squaredNorm() > tol2 && i<maxIters )
   {
     Scalar rho_old = rho;
 
     rho = r0.dot(r);
     if (abs(rho) < eps2*r0_sqnorm)
     {
-      // The new residual vector became too orthogonal to the arbitrarily choosen direction r0
+      // The new residual vector became too orthogonal to the arbitrarily chosen direction r0
       // Let's restart with a new r0:
+      r  = rhs - mat * x;
       r0 = r;
       rho = r0_sqnorm = r.squaredNorm();
       if(restarts++ == 0)
@@ -131,35 +132,33 @@
   * \tparam _MatrixType the type of the sparse matrix A, can be a dense or a sparse matrix.
   * \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
   *
+  * \implsparsesolverconcept
+  *
   * The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
   * and setTolerance() methods. The defaults are the size of the problem for the maximal number of iterations
   * and NumTraits<Scalar>::epsilon() for the tolerance.
   * 
+  * The tolerance corresponds to the relative residual error: |Ax-b|/|b|
+  * 
+  * \b Performance: when using sparse matrices, best performance is achied for a row-major sparse matrix format.
+  * Moreover, in this case multi-threading can be exploited if the user code is compiled with OpenMP enabled.
+  * See \ref TopicMultiThreading for details.
+  * 
   * This class can be used as the direct solver classes. Here is a typical usage example:
-  * \code
-  * int n = 10000;
-  * VectorXd x(n), b(n);
-  * SparseMatrix<double> A(n,n);
-  * // fill A and b
-  * BiCGSTAB<SparseMatrix<double> > solver;
-  * solver.compute(A);
-  * x = solver.solve(b);
-  * std::cout << "#iterations:     " << solver.iterations() << std::endl;
-  * std::cout << "estimated error: " << solver.error()      << std::endl;
-  * // update b, and solve again
-  * x = solver.solve(b);
-  * \endcode
+  * \include BiCGSTAB_simple.cpp
   * 
   * By default the iterations start with x=0 as an initial guess of the solution.
   * One can control the start using the solveWithGuess() method.
   * 
+  * BiCGSTAB can also be used in a matrix-free context, see the following \link MatrixfreeSolverExample example \endlink.
+  *
   * \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
   */
 template< typename _MatrixType, typename _Preconditioner>
 class BiCGSTAB : public IterativeSolverBase<BiCGSTAB<_MatrixType,_Preconditioner> >
 {
   typedef IterativeSolverBase<BiCGSTAB> Base;
-  using Base::mp_matrix;
+  using Base::matrix;
   using Base::m_error;
   using Base::m_iterations;
   using Base::m_info;
@@ -167,7 +166,6 @@
 public:
   typedef _MatrixType MatrixType;
   typedef typename MatrixType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::RealScalar RealScalar;
   typedef _Preconditioner Preconditioner;
 
@@ -186,38 +184,23 @@
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  BiCGSTAB(const MatrixType& A) : Base(A) {}
+  template<typename MatrixDerived>
+  explicit BiCGSTAB(const EigenBase<MatrixDerived>& A) : Base(A.derived()) {}
 
   ~BiCGSTAB() {}
-  
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
-    * \a x0 as an initial solution.
-    *
-    * \sa compute()
-    */
-  template<typename Rhs,typename Guess>
-  inline const internal::solve_retval_with_guess<BiCGSTAB, Rhs, Guess>
-  solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
-  {
-    eigen_assert(m_isInitialized && "BiCGSTAB is not initialized.");
-    eigen_assert(Base::rows()==b.rows()
-              && "BiCGSTAB::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::solve_retval_with_guess
-            <BiCGSTAB, Rhs, Guess>(*this, b.derived(), x0);
-  }
-  
+
   /** \internal */
   template<typename Rhs,typename Dest>
-  void _solveWithGuess(const Rhs& b, Dest& x) const
+  void _solve_with_guess_impl(const Rhs& b, Dest& x) const
   {    
     bool failed = false;
-    for(int j=0; j<b.cols(); ++j)
+    for(Index j=0; j<b.cols(); ++j)
     {
       m_iterations = Base::maxIterations();
       m_error = Base::m_tolerance;
       
       typename Dest::ColXpr xj(x,j);
-      if(!internal::bicgstab(*mp_matrix, b.col(j), xj, Base::m_preconditioner, m_iterations, m_error))
+      if(!internal::bicgstab(matrix(), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error))
         failed = true;
     }
     m_info = failed ? NumericalIssue
@@ -227,36 +210,19 @@
   }
 
   /** \internal */
+  using Base::_solve_impl;
   template<typename Rhs,typename Dest>
-  void _solve(const Rhs& b, Dest& x) const
+  void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
   {
-//     x.setZero();
-  x = b;
-    _solveWithGuess(b,x);
+    x.resize(this->rows(),b.cols());
+    x.setZero();
+    _solve_with_guess_impl(b,x);
   }
 
 protected:
 
 };
 
-
-namespace internal {
-
-  template<typename _MatrixType, typename _Preconditioner, typename Rhs>
-struct solve_retval<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
-  : solve_retval_base<BiCGSTAB<_MatrixType, _Preconditioner>, Rhs>
-{
-  typedef BiCGSTAB<_MatrixType, _Preconditioner> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_BICGSTAB_H
diff --git a/Eigen/src/IterativeLinearSolvers/CMakeLists.txt b/Eigen/src/IterativeLinearSolvers/CMakeLists.txt
deleted file mode 100644
index 59ccc00..0000000
--- a/Eigen/src/IterativeLinearSolvers/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_IterativeLinearSolvers_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_IterativeLinearSolvers_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/IterativeLinearSolvers COMPONENT Devel
-  )
diff --git a/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h b/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h
index 8ba4a8d..f7ce471 100644
--- a/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h
+++ b/Eigen/src/IterativeLinearSolvers/ConjugateGradient.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -26,7 +26,7 @@
 template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
 EIGEN_DONT_INLINE
 void conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
-                        const Preconditioner& precond, int& iters,
+                        const Preconditioner& precond, Index& iters,
                         typename Dest::RealScalar& tol_error)
 {
   using std::sqrt;
@@ -36,9 +36,9 @@
   typedef Matrix<Scalar,Dynamic,1> VectorType;
   
   RealScalar tol = tol_error;
-  int maxIters = iters;
+  Index maxIters = iters;
   
-  int n = mat.cols();
+  Index n = mat.cols();
 
   VectorType residual = rhs - mat * x; //initial residual
 
@@ -50,7 +50,8 @@
     tol_error = 0;
     return;
   }
-  RealScalar threshold = tol*tol*rhsNorm2;
+  const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
+  RealScalar threshold = numext::maxi(tol*tol*rhsNorm2,considerAsZero);
   RealScalar residualNorm2 = residual.squaredNorm();
   if (residualNorm2 < threshold)
   {
@@ -58,31 +59,31 @@
     tol_error = sqrt(residualNorm2 / rhsNorm2);
     return;
   }
-  
+
   VectorType p(n);
-  p = precond.solve(residual);      //initial search direction
+  p = precond.solve(residual);      // initial search direction
 
   VectorType z(n), tmp(n);
   RealScalar absNew = numext::real(residual.dot(p));  // the square of the absolute value of r scaled by invM
-  int i = 0;
+  Index i = 0;
   while(i < maxIters)
   {
-    tmp.noalias() = mat * p;              // the bottleneck of the algorithm
+    tmp.noalias() = mat * p;                    // the bottleneck of the algorithm
 
-    Scalar alpha = absNew / p.dot(tmp);   // the amount we travel on dir
-    x += alpha * p;                       // update solution
-    residual -= alpha * tmp;              // update residue
+    Scalar alpha = absNew / p.dot(tmp);         // the amount we travel on dir
+    x += alpha * p;                             // update solution
+    residual -= alpha * tmp;                    // update residual
     
     residualNorm2 = residual.squaredNorm();
     if(residualNorm2 < threshold)
       break;
     
-    z = precond.solve(residual);          // approximately solve for "A z = residual"
+    z = precond.solve(residual);                // approximately solve for "A z = residual"
 
     RealScalar absOld = absNew;
     absNew = numext::real(residual.dot(z));     // update the absolute value of r
-    RealScalar beta = absNew / absOld;            // calculate the Gram-Schmidt value used to create the new search direction
-    p = z + beta * p;                             // update search direction
+    RealScalar beta = absNew / absOld;          // calculate the Gram-Schmidt value used to create the new search direction
+    p = z + beta * p;                           // update search direction
     i++;
   }
   tol_error = sqrt(residualNorm2 / rhsNorm2);
@@ -107,45 +108,57 @@
 }
 
 /** \ingroup IterativeLinearSolvers_Module
-  * \brief A conjugate gradient solver for sparse self-adjoint problems
+  * \brief A conjugate gradient solver for sparse (or dense) self-adjoint problems
   *
-  * This class allows to solve for A.x = b sparse linear problems using a conjugate gradient algorithm.
-  * The sparse matrix A must be selfadjoint. The vectors x and b can be either dense or sparse.
+  * This class allows to solve for A.x = b linear problems using an iterative conjugate gradient algorithm.
+  * The matrix A must be selfadjoint. The matrix A and the vectors x and b can be either dense or sparse.
   *
   * \tparam _MatrixType the type of the matrix A, can be a dense or a sparse matrix.
   * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower,
-  *               Upper, or Lower|Upper in which the full matrix entries will be considered. Default is Lower.
+  *               \c Upper, or \c Lower|Upper in which the full matrix entries will be considered.
+  *               Default is \c Lower, best performance is \c Lower|Upper.
   * \tparam _Preconditioner the type of the preconditioner. Default is DiagonalPreconditioner
   *
+  * \implsparsesolverconcept
+  *
   * The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
   * and setTolerance() methods. The defaults are the size of the problem for the maximal number of iterations
   * and NumTraits<Scalar>::epsilon() for the tolerance.
   * 
+  * The tolerance corresponds to the relative residual error: |Ax-b|/|b|
+  * 
+  * \b Performance: Even though the default value of \c _UpLo is \c Lower, significantly higher performance is
+  * achieved when using a complete matrix and \b Lower|Upper as the \a _UpLo template parameter. Moreover, in this
+  * case multi-threading can be exploited if the user code is compiled with OpenMP enabled.
+  * See \ref TopicMultiThreading for details.
+  * 
   * This class can be used as the direct solver classes. Here is a typical usage example:
-  * \code
-  * int n = 10000;
-  * VectorXd x(n), b(n);
-  * SparseMatrix<double> A(n,n);
-  * // fill A and b
-  * ConjugateGradient<SparseMatrix<double> > cg;
-  * cg.compute(A);
-  * x = cg.solve(b);
-  * std::cout << "#iterations:     " << cg.iterations() << std::endl;
-  * std::cout << "estimated error: " << cg.error()      << std::endl;
-  * // update b, and solve again
-  * x = cg.solve(b);
-  * \endcode
+    \code
+    int n = 10000;
+    VectorXd x(n), b(n);
+    SparseMatrix<double> A(n,n);
+    // fill A and b
+    ConjugateGradient<SparseMatrix<double>, Lower|Upper> cg;
+    cg.compute(A);
+    x = cg.solve(b);
+    std::cout << "#iterations:     " << cg.iterations() << std::endl;
+    std::cout << "estimated error: " << cg.error()      << std::endl;
+    // update b, and solve again
+    x = cg.solve(b);
+    \endcode
   * 
   * By default the iterations start with x=0 as an initial guess of the solution.
   * One can control the start using the solveWithGuess() method.
   * 
-  * \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
+  * ConjugateGradient can also be used in a matrix-free context, see the following \link MatrixfreeSolverExample example \endlink.
+  *
+  * \sa class LeastSquaresConjugateGradient, class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
   */
 template< typename _MatrixType, int _UpLo, typename _Preconditioner>
 class ConjugateGradient : public IterativeSolverBase<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> >
 {
   typedef IterativeSolverBase<ConjugateGradient> Base;
-  using Base::mp_matrix;
+  using Base::matrix;
   using Base::m_error;
   using Base::m_iterations;
   using Base::m_info;
@@ -153,7 +166,6 @@
 public:
   typedef _MatrixType MatrixType;
   typedef typename MatrixType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::RealScalar RealScalar;
   typedef _Preconditioner Preconditioner;
 
@@ -176,44 +188,40 @@
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  ConjugateGradient(const MatrixType& A) : Base(A) {}
+  template<typename MatrixDerived>
+  explicit ConjugateGradient(const EigenBase<MatrixDerived>& A) : Base(A.derived()) {}
 
   ~ConjugateGradient() {}
-  
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
-    * \a x0 as an initial solution.
-    *
-    * \sa compute()
-    */
-  template<typename Rhs,typename Guess>
-  inline const internal::solve_retval_with_guess<ConjugateGradient, Rhs, Guess>
-  solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
-  {
-    eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
-    eigen_assert(Base::rows()==b.rows()
-              && "ConjugateGradient::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::solve_retval_with_guess
-            <ConjugateGradient, Rhs, Guess>(*this, b.derived(), x0);
-  }
 
   /** \internal */
   template<typename Rhs,typename Dest>
-  void _solveWithGuess(const Rhs& b, Dest& x) const
+  void _solve_with_guess_impl(const Rhs& b, Dest& x) const
   {
+    typedef typename Base::MatrixWrapper MatrixWrapper;
+    typedef typename Base::ActualMatrixType ActualMatrixType;
+    enum {
+      TransposeInput  =   (!MatrixWrapper::MatrixFree)
+                      &&  (UpLo==(Lower|Upper))
+                      &&  (!MatrixType::IsRowMajor)
+                      &&  (!NumTraits<Scalar>::IsComplex)
+    };
+    typedef typename internal::conditional<TransposeInput,Transpose<const ActualMatrixType>, ActualMatrixType const&>::type RowMajorWrapper;
+    EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(MatrixWrapper::MatrixFree,UpLo==(Lower|Upper)),MATRIX_FREE_CONJUGATE_GRADIENT_IS_COMPATIBLE_WITH_UPPER_UNION_LOWER_MODE_ONLY);
     typedef typename internal::conditional<UpLo==(Lower|Upper),
-                                           const MatrixType&,
-                                           SparseSelfAdjointView<const MatrixType, UpLo>
-                                          >::type MatrixWrapperType;
+                                           RowMajorWrapper,
+                                           typename MatrixWrapper::template ConstSelfAdjointViewReturnType<UpLo>::Type
+                                          >::type SelfAdjointWrapper;
     m_iterations = Base::maxIterations();
     m_error = Base::m_tolerance;
 
-    for(int j=0; j<b.cols(); ++j)
+    for(Index j=0; j<b.cols(); ++j)
     {
       m_iterations = Base::maxIterations();
       m_error = Base::m_tolerance;
 
       typename Dest::ColXpr xj(x,j);
-      internal::conjugate_gradient(MatrixWrapperType(*mp_matrix), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
+      RowMajorWrapper row_mat(matrix());
+      internal::conjugate_gradient(SelfAdjointWrapper(row_mat), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
     }
 
     m_isInitialized = true;
@@ -221,35 +229,18 @@
   }
   
   /** \internal */
+  using Base::_solve_impl;
   template<typename Rhs,typename Dest>
-  void _solve(const Rhs& b, Dest& x) const
+  void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
   {
     x.setZero();
-    _solveWithGuess(b,x);
+    _solve_with_guess_impl(b.derived(),x);
   }
 
 protected:
 
 };
 
-
-namespace internal {
-
-template<typename _MatrixType, int _UpLo, typename _Preconditioner, typename Rhs>
-struct solve_retval<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
-  : solve_retval_base<ConjugateGradient<_MatrixType,_UpLo,_Preconditioner>, Rhs>
-{
-  typedef ConjugateGradient<_MatrixType,_UpLo,_Preconditioner> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_CONJUGATE_GRADIENT_H
diff --git a/Eigen/src/IterativeLinearSolvers/IncompleteCholesky.h b/Eigen/src/IterativeLinearSolvers/IncompleteCholesky.h
new file mode 100644
index 0000000..e45c272
--- /dev/null
+++ b/Eigen/src/IterativeLinearSolvers/IncompleteCholesky.h
@@ -0,0 +1,400 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_INCOMPLETE_CHOlESKY_H
+#define EIGEN_INCOMPLETE_CHOlESKY_H
+
+#include <vector>
+#include <list>
+
+namespace Eigen {  
+/** 
+  * \brief Modified Incomplete Cholesky with dual threshold
+  *
+  * References : C-J. Lin and J. J. Moré, Incomplete Cholesky Factorizations with
+  *              Limited memory, SIAM J. Sci. Comput.  21(1), pp. 24-45, 1999
+  *
+  * \tparam Scalar the scalar type of the input matrices
+  * \tparam _UpLo The triangular part that will be used for the computations. It can be Lower
+    *               or Upper. Default is Lower.
+  * \tparam _OrderingType The ordering method to use, either AMDOrdering<> or NaturalOrdering<>. Default is AMDOrdering<int>,
+  *                       unless EIGEN_MPL2_ONLY is defined, in which case the default is NaturalOrdering<int>.
+  *
+  * \implsparsesolverconcept
+  *
+  * It performs the following incomplete factorization: \f$ S P A P' S \approx L L' \f$
+  * where L is a lower triangular factor, S is a diagonal scaling matrix, and P is a
+  * fill-in reducing permutation as computed by the ordering method.
+  *
+  * \b Shifting \b strategy: Let \f$ B = S P A P' S \f$  be the scaled matrix on which the factorization is carried out,
+  * and \f$ \beta \f$ be the minimum value of the diagonal. If \f$ \beta > 0 \f$ then, the factorization is directly performed
+  * on the matrix B. Otherwise, the factorization is performed on the shifted matrix \f$ B + (\sigma+|\beta| I \f$ where
+  * \f$ \sigma \f$ is the initial shift value as returned and set by setInitialShift() method. The default value is \f$ \sigma = 10^{-3} \f$.
+  * If the factorization fails, then the shift in doubled until it succeed or a maximum of ten attempts. If it still fails, as returned by
+  * the info() method, then you can either increase the initial shift, or better use another preconditioning technique.
+  *
+  */
+template <typename Scalar, int _UpLo = Lower, typename _OrderingType =
+#ifndef EIGEN_MPL2_ONLY
+AMDOrdering<int>
+#else
+NaturalOrdering<int>
+#endif
+>
+class IncompleteCholesky : public SparseSolverBase<IncompleteCholesky<Scalar,_UpLo,_OrderingType> >
+{
+  protected:
+    typedef SparseSolverBase<IncompleteCholesky<Scalar,_UpLo,_OrderingType> > Base;
+    using Base::m_isInitialized;
+  public:
+    typedef typename NumTraits<Scalar>::Real RealScalar; 
+    typedef _OrderingType OrderingType;
+    typedef typename OrderingType::PermutationType PermutationType;
+    typedef typename PermutationType::StorageIndex StorageIndex; 
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> FactorType;
+    typedef Matrix<Scalar,Dynamic,1> VectorSx;
+    typedef Matrix<RealScalar,Dynamic,1> VectorRx;
+    typedef Matrix<StorageIndex,Dynamic, 1> VectorIx;
+    typedef std::vector<std::list<StorageIndex> > VectorList; 
+    enum { UpLo = _UpLo };
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
+  public:
+
+    /** Default constructor leaving the object in a partly non-initialized stage.
+      *
+      * You must call compute() or the pair analyzePattern()/factorize() to make it valid.
+      *
+      * \sa IncompleteCholesky(const MatrixType&)
+      */
+    IncompleteCholesky() : m_initialShift(1e-3),m_factorizationIsOk(false) {}
+    
+    /** Constructor computing the incomplete factorization for the given matrix \a matrix.
+      */
+    template<typename MatrixType>
+    IncompleteCholesky(const MatrixType& matrix) : m_initialShift(1e-3),m_factorizationIsOk(false)
+    {
+      compute(matrix);
+    }
+    
+    /** \returns number of rows of the factored matrix */
+    Index rows() const { return m_L.rows(); }
+    
+    /** \returns number of columns of the factored matrix */
+    Index cols() const { return m_L.cols(); }
+    
+
+    /** \brief Reports whether previous computation was successful.
+      *
+      * It triggers an assertion if \c *this has not been initialized through the respective constructor,
+      * or a call to compute() or analyzePattern().
+      *
+      * \returns \c Success if computation was successful,
+      *          \c NumericalIssue if the matrix appears to be negative.
+      */
+    ComputationInfo info() const
+    {
+      eigen_assert(m_isInitialized && "IncompleteCholesky is not initialized.");
+      return m_info;
+    }
+    
+    /** \brief Set the initial shift parameter \f$ \sigma \f$.
+      */
+    void setInitialShift(RealScalar shift) { m_initialShift = shift; }
+    
+    /** \brief Computes the fill reducing permutation vector using the sparsity pattern of \a mat
+      */
+    template<typename MatrixType>
+    void analyzePattern(const MatrixType& mat)
+    {
+      OrderingType ord; 
+      PermutationType pinv;
+      ord(mat.template selfadjointView<UpLo>(), pinv); 
+      if(pinv.size()>0) m_perm = pinv.inverse();
+      else              m_perm.resize(0);
+      m_L.resize(mat.rows(), mat.cols());
+      m_analysisIsOk = true;
+      m_isInitialized = true;
+      m_info = Success;
+    }
+    
+    /** \brief Performs the numerical factorization of the input matrix \a mat
+      *
+      * The method analyzePattern() or compute() must have been called beforehand
+      * with a matrix having the same pattern.
+      *
+      * \sa compute(), analyzePattern()
+      */
+    template<typename MatrixType>
+    void factorize(const MatrixType& mat);
+    
+    /** Computes or re-computes the incomplete Cholesky factorization of the input matrix \a mat
+      *
+      * It is a shortcut for a sequential call to the analyzePattern() and factorize() methods.
+      *
+      * \sa analyzePattern(), factorize()
+      */
+    template<typename MatrixType>
+    void compute(const MatrixType& mat)
+    {
+      analyzePattern(mat);
+      factorize(mat);
+    }
+    
+    // internal
+    template<typename Rhs, typename Dest>
+    void _solve_impl(const Rhs& b, Dest& x) const
+    {
+      eigen_assert(m_factorizationIsOk && "factorize() should be called first");
+      if (m_perm.rows() == b.rows())  x = m_perm * b;
+      else                            x = b;
+      x = m_scale.asDiagonal() * x;
+      x = m_L.template triangularView<Lower>().solve(x);
+      x = m_L.adjoint().template triangularView<Upper>().solve(x);
+      x = m_scale.asDiagonal() * x;
+      if (m_perm.rows() == b.rows())
+        x = m_perm.inverse() * x;
+    }
+
+    /** \returns the sparse lower triangular factor L */
+    const FactorType& matrixL() const { eigen_assert("m_factorizationIsOk"); return m_L; }
+
+    /** \returns a vector representing the scaling factor S */
+    const VectorRx& scalingS() const { eigen_assert("m_factorizationIsOk"); return m_scale; }
+
+    /** \returns the fill-in reducing permutation P (can be empty for a natural ordering) */
+    const PermutationType& permutationP() const { eigen_assert("m_analysisIsOk"); return m_perm; }
+
+  protected:
+    FactorType m_L;              // The lower part stored in CSC
+    VectorRx m_scale;            // The vector for scaling the matrix 
+    RealScalar m_initialShift;   // The initial shift parameter
+    bool m_analysisIsOk; 
+    bool m_factorizationIsOk; 
+    ComputationInfo m_info;
+    PermutationType m_perm; 
+
+  private:
+    inline void updateList(Ref<const VectorIx> colPtr, Ref<VectorIx> rowIdx, Ref<VectorSx> vals, const Index& col, const Index& jk, VectorIx& firstElt, VectorList& listCol); 
+}; 
+
+// Based on the following paper:
+//   C-J. Lin and J. J. Moré, Incomplete Cholesky Factorizations with
+//   Limited memory, SIAM J. Sci. Comput.  21(1), pp. 24-45, 1999
+//   http://ftp.mcs.anl.gov/pub/tech_reports/reports/P682.pdf
+template<typename Scalar, int _UpLo, typename OrderingType>
+template<typename _MatrixType>
+void IncompleteCholesky<Scalar,_UpLo, OrderingType>::factorize(const _MatrixType& mat)
+{
+  using std::sqrt;
+  eigen_assert(m_analysisIsOk && "analyzePattern() should be called first"); 
+    
+  // Dropping strategy : Keep only the p largest elements per column, where p is the number of elements in the column of the original matrix. Other strategies will be added
+  
+  // Apply the fill-reducing permutation computed in analyzePattern()
+  if (m_perm.rows() == mat.rows() ) // To detect the null permutation
+  {
+    // The temporary is needed to make sure that the diagonal entry is properly sorted
+    FactorType tmp(mat.rows(), mat.cols());
+    tmp = mat.template selfadjointView<_UpLo>().twistedBy(m_perm);
+    m_L.template selfadjointView<Lower>() = tmp.template selfadjointView<Lower>();
+  }
+  else
+  {
+    m_L.template selfadjointView<Lower>() = mat.template selfadjointView<_UpLo>();
+  }
+  
+  Index n = m_L.cols(); 
+  Index nnz = m_L.nonZeros();
+  Map<VectorSx> vals(m_L.valuePtr(), nnz);         //values
+  Map<VectorIx> rowIdx(m_L.innerIndexPtr(), nnz);  //Row indices
+  Map<VectorIx> colPtr( m_L.outerIndexPtr(), n+1); // Pointer to the beginning of each row
+  VectorIx firstElt(n-1); // for each j, points to the next entry in vals that will be used in the factorization
+  VectorList listCol(n);  // listCol(j) is a linked list of columns to update column j
+  VectorSx col_vals(n);   // Store a  nonzero values in each column
+  VectorIx col_irow(n);   // Row indices of nonzero elements in each column
+  VectorIx col_pattern(n);
+  col_pattern.fill(-1);
+  StorageIndex col_nnz;
+  
+  
+  // Computes the scaling factors 
+  m_scale.resize(n);
+  m_scale.setZero();
+  for (Index j = 0; j < n; j++)
+    for (Index k = colPtr[j]; k < colPtr[j+1]; k++)
+    {
+      m_scale(j) += numext::abs2(vals(k));
+      if(rowIdx[k]!=j)
+        m_scale(rowIdx[k]) += numext::abs2(vals(k));
+    }
+  
+  m_scale = m_scale.cwiseSqrt().cwiseSqrt();
+
+  for (Index j = 0; j < n; ++j)
+    if(m_scale(j)>(std::numeric_limits<RealScalar>::min)())
+      m_scale(j) = RealScalar(1)/m_scale(j);
+    else
+      m_scale(j) = 1;
+
+  // TODO disable scaling if not needed, i.e., if it is roughly uniform? (this will make solve() faster)
+  
+  // Scale and compute the shift for the matrix 
+  RealScalar mindiag = NumTraits<RealScalar>::highest();
+  for (Index j = 0; j < n; j++)
+  {
+    for (Index k = colPtr[j]; k < colPtr[j+1]; k++)
+      vals[k] *= (m_scale(j)*m_scale(rowIdx[k]));
+    eigen_internal_assert(rowIdx[colPtr[j]]==j && "IncompleteCholesky: only the lower triangular part must be stored");
+    mindiag = numext::mini(numext::real(vals[colPtr[j]]), mindiag);
+  }
+
+  FactorType L_save = m_L;
+  
+  RealScalar shift = 0;
+  if(mindiag <= RealScalar(0.))
+    shift = m_initialShift - mindiag;
+
+  m_info = NumericalIssue;
+
+  // Try to perform the incomplete factorization using the current shift
+  int iter = 0;
+  do
+  {
+    // Apply the shift to the diagonal elements of the matrix
+    for (Index j = 0; j < n; j++)
+      vals[colPtr[j]] += shift;
+
+    // jki version of the Cholesky factorization
+    Index j=0;
+    for (; j < n; ++j)
+    {
+      // Left-looking factorization of the j-th column
+      // First, load the j-th column into col_vals
+      Scalar diag = vals[colPtr[j]];  // It is assumed that only the lower part is stored
+      col_nnz = 0;
+      for (Index i = colPtr[j] + 1; i < colPtr[j+1]; i++)
+      {
+        StorageIndex l = rowIdx[i];
+        col_vals(col_nnz) = vals[i];
+        col_irow(col_nnz) = l;
+        col_pattern(l) = col_nnz;
+        col_nnz++;
+      }
+      {
+        typename std::list<StorageIndex>::iterator k;
+        // Browse all previous columns that will update column j
+        for(k = listCol[j].begin(); k != listCol[j].end(); k++)
+        {
+          Index jk = firstElt(*k); // First element to use in the column
+          eigen_internal_assert(rowIdx[jk]==j);
+          Scalar v_j_jk = numext::conj(vals[jk]);
+
+          jk += 1;
+          for (Index i = jk; i < colPtr[*k+1]; i++)
+          {
+            StorageIndex l = rowIdx[i];
+            if(col_pattern[l]<0)
+            {
+              col_vals(col_nnz) = vals[i] * v_j_jk;
+              col_irow[col_nnz] = l;
+              col_pattern(l) = col_nnz;
+              col_nnz++;
+            }
+            else
+              col_vals(col_pattern[l]) -= vals[i] * v_j_jk;
+          }
+          updateList(colPtr,rowIdx,vals, *k, jk, firstElt, listCol);
+        }
+      }
+
+      // Scale the current column
+      if(numext::real(diag) <= 0)
+      {
+        if(++iter>=10)
+          return;
+
+        // increase shift
+        shift = numext::maxi(m_initialShift,RealScalar(2)*shift);
+        // restore m_L, col_pattern, and listCol
+        vals = Map<const VectorSx>(L_save.valuePtr(), nnz);
+        rowIdx = Map<const VectorIx>(L_save.innerIndexPtr(), nnz);
+        colPtr = Map<const VectorIx>(L_save.outerIndexPtr(), n+1);
+        col_pattern.fill(-1);
+        for(Index i=0; i<n; ++i)
+          listCol[i].clear();
+
+        break;
+      }
+
+      RealScalar rdiag = sqrt(numext::real(diag));
+      vals[colPtr[j]] = rdiag;
+      for (Index k = 0; k<col_nnz; ++k)
+      {
+        Index i = col_irow[k];
+        //Scale
+        col_vals(k) /= rdiag;
+        //Update the remaining diagonals with col_vals
+        vals[colPtr[i]] -= numext::abs2(col_vals(k));
+      }
+      // Select the largest p elements
+      // p is the original number of elements in the column (without the diagonal)
+      Index p = colPtr[j+1] - colPtr[j] - 1 ;
+      Ref<VectorSx> cvals = col_vals.head(col_nnz);
+      Ref<VectorIx> cirow = col_irow.head(col_nnz);
+      internal::QuickSplit(cvals,cirow, p);
+      // Insert the largest p elements in the matrix
+      Index cpt = 0;
+      for (Index i = colPtr[j]+1; i < colPtr[j+1]; i++)
+      {
+        vals[i] = col_vals(cpt);
+        rowIdx[i] = col_irow(cpt);
+        // restore col_pattern:
+        col_pattern(col_irow(cpt)) = -1;
+        cpt++;
+      }
+      // Get the first smallest row index and put it after the diagonal element
+      Index jk = colPtr(j)+1;
+      updateList(colPtr,rowIdx,vals,j,jk,firstElt,listCol);
+    }
+
+    if(j==n)
+    {
+      m_factorizationIsOk = true;
+      m_info = Success;
+    }
+  } while(m_info!=Success);
+}
+
+template<typename Scalar, int _UpLo, typename OrderingType>
+inline void IncompleteCholesky<Scalar,_UpLo, OrderingType>::updateList(Ref<const VectorIx> colPtr, Ref<VectorIx> rowIdx, Ref<VectorSx> vals, const Index& col, const Index& jk, VectorIx& firstElt, VectorList& listCol)
+{
+  if (jk < colPtr(col+1) )
+  {
+    Index p = colPtr(col+1) - jk;
+    Index minpos; 
+    rowIdx.segment(jk,p).minCoeff(&minpos);
+    minpos += jk;
+    if (rowIdx(minpos) != rowIdx(jk))
+    {
+      //Swap
+      std::swap(rowIdx(jk),rowIdx(minpos));
+      std::swap(vals(jk),vals(minpos));
+    }
+    firstElt(col) = internal::convert_index<StorageIndex,Index>(jk);
+    listCol[rowIdx(jk)].push_back(internal::convert_index<StorageIndex,Index>(col));
+  }
+}
+
+} // end namespace Eigen 
+
+#endif
diff --git a/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h b/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h
index 4c169aa..338e6f1 100644
--- a/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h
+++ b/Eigen/src/IterativeLinearSolvers/IncompleteLUT.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -24,7 +25,7 @@
   * \param ind The array of index for the elements in @p row
   * \param ncut  The number of largest elements to keep
   **/ 
-template <typename VectorV, typename VectorI, typename Index>
+template <typename VectorV, typename VectorI>
 Index QuickSplit(VectorV &row, VectorI &ind, Index ncut)
 {
   typedef typename VectorV::RealScalar RealScalar;
@@ -66,6 +67,8 @@
   * \class IncompleteLUT
   * \brief Incomplete LU factorization with dual-threshold strategy
   *
+  * \implsparsesolverconcept
+  *
   * During the numerical factorization, two dropping rules are used :
   *  1) any element whose magnitude is less than some tolerance is dropped.
   *    This tolerance is obtained by multiplying the input tolerance @p droptol 
@@ -92,28 +95,36 @@
   * alternatively, on GMANE:
   *   http://comments.gmane.org/gmane.comp.lib.eigen/3302
   */
-template <typename _Scalar>
-class IncompleteLUT : internal::noncopyable
+template <typename _Scalar, typename _StorageIndex = int>
+class IncompleteLUT : public SparseSolverBase<IncompleteLUT<_Scalar, _StorageIndex> >
 {
+  protected:
+    typedef SparseSolverBase<IncompleteLUT> Base;
+    using Base::m_isInitialized;
+  public:
     typedef _Scalar Scalar;
+    typedef _StorageIndex StorageIndex;
     typedef typename NumTraits<Scalar>::Real RealScalar;
     typedef Matrix<Scalar,Dynamic,1> Vector;
-    typedef SparseMatrix<Scalar,RowMajor> FactorType;
-    typedef SparseMatrix<Scalar,ColMajor> PermutType;
-    typedef typename FactorType::Index Index;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+    typedef SparseMatrix<Scalar,RowMajor,StorageIndex> FactorType;
+
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
 
   public:
-    typedef Matrix<Scalar,Dynamic,Dynamic> MatrixType;
     
     IncompleteLUT()
       : m_droptol(NumTraits<Scalar>::dummy_precision()), m_fillfactor(10),
-        m_analysisIsOk(false), m_factorizationIsOk(false), m_isInitialized(false)
+        m_analysisIsOk(false), m_factorizationIsOk(false)
     {}
     
     template<typename MatrixType>
-    IncompleteLUT(const MatrixType& mat, const RealScalar& droptol=NumTraits<Scalar>::dummy_precision(), int fillfactor = 10)
+    explicit IncompleteLUT(const MatrixType& mat, const RealScalar& droptol=NumTraits<Scalar>::dummy_precision(), int fillfactor = 10)
       : m_droptol(droptol),m_fillfactor(fillfactor),
-        m_analysisIsOk(false),m_factorizationIsOk(false),m_isInitialized(false)
+        m_analysisIsOk(false),m_factorizationIsOk(false)
     {
       eigen_assert(fillfactor != 0);
       compute(mat); 
@@ -146,7 +157,7 @@
       * 
       **/
     template<typename MatrixType>
-    IncompleteLUT<Scalar>& compute(const MatrixType& amat)
+    IncompleteLUT& compute(const MatrixType& amat)
     {
       analyzePattern(amat); 
       factorize(amat);
@@ -157,23 +168,14 @@
     void setFillfactor(int fillfactor); 
     
     template<typename Rhs, typename Dest>
-    void _solve(const Rhs& b, Dest& x) const
+    void _solve_impl(const Rhs& b, Dest& x) const
     {
-      x = m_Pinv * b;  
+      x = m_Pinv * b;
       x = m_lu.template triangularView<UnitLower>().solve(x);
       x = m_lu.template triangularView<Upper>().solve(x);
       x = m_P * x; 
     }
 
-    template<typename Rhs> inline const internal::solve_retval<IncompleteLUT, Rhs>
-     solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "IncompleteLUT is not initialized.");
-      eigen_assert(cols()==b.rows()
-                && "IncompleteLUT::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<IncompleteLUT, Rhs>(*this, b.derived());
-    }
-
 protected:
 
     /** keeps off-diagonal entries; drops diagonal entries */
@@ -191,18 +193,17 @@
     int m_fillfactor;
     bool m_analysisIsOk;
     bool m_factorizationIsOk;
-    bool m_isInitialized;
     ComputationInfo m_info;
-    PermutationMatrix<Dynamic,Dynamic,Index> m_P;     // Fill-reducing permutation
-    PermutationMatrix<Dynamic,Dynamic,Index> m_Pinv;  // Inverse permutation
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_P;     // Fill-reducing permutation
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_Pinv;  // Inverse permutation
 };
 
 /**
  * Set control parameter droptol
  *  \param droptol   Drop any element whose magnitude is less than this tolerance 
  **/ 
-template<typename Scalar>
-void IncompleteLUT<Scalar>::setDroptol(const RealScalar& droptol)
+template<typename Scalar, typename StorageIndex>
+void IncompleteLUT<Scalar,StorageIndex>::setDroptol(const RealScalar& droptol)
 {
   this->m_droptol = droptol;   
 }
@@ -211,52 +212,62 @@
  * Set control parameter fillfactor
  * \param fillfactor  This is used to compute the  number @p fill_in of largest elements to keep on each row. 
  **/ 
-template<typename Scalar>
-void IncompleteLUT<Scalar>::setFillfactor(int fillfactor)
+template<typename Scalar, typename StorageIndex>
+void IncompleteLUT<Scalar,StorageIndex>::setFillfactor(int fillfactor)
 {
   this->m_fillfactor = fillfactor;   
 }
 
-template <typename Scalar>
+template <typename Scalar, typename StorageIndex>
 template<typename _MatrixType>
-void IncompleteLUT<Scalar>::analyzePattern(const _MatrixType& amat)
+void IncompleteLUT<Scalar,StorageIndex>::analyzePattern(const _MatrixType& amat)
 {
   // Compute the Fill-reducing permutation
-  SparseMatrix<Scalar,ColMajor, Index> mat1 = amat;
-  SparseMatrix<Scalar,ColMajor, Index> mat2 = amat.transpose();
-  // Symmetrize the pattern
+  // Since ILUT does not perform any numerical pivoting,
+  // it is highly preferable to keep the diagonal through symmetric permutations.
+#ifndef EIGEN_MPL2_ONLY
+  // To this end, let's symmetrize the pattern and perform AMD on it.
+  SparseMatrix<Scalar,ColMajor, StorageIndex> mat1 = amat;
+  SparseMatrix<Scalar,ColMajor, StorageIndex> mat2 = amat.transpose();
   // FIXME for a matrix with nearly symmetric pattern, mat2+mat1 is the appropriate choice.
   //       on the other hand for a really non-symmetric pattern, mat2*mat1 should be prefered...
-  SparseMatrix<Scalar,ColMajor, Index> AtA = mat2 + mat1;
-  AtA.prune(keep_diag());
-  internal::minimum_degree_ordering<Scalar, Index>(AtA, m_P);  // Then compute the AMD ordering...
-
-  m_Pinv  = m_P.inverse(); // ... and the inverse permutation
+  SparseMatrix<Scalar,ColMajor, StorageIndex> AtA = mat2 + mat1;
+  AMDOrdering<StorageIndex> ordering;
+  ordering(AtA,m_P);
+  m_Pinv  = m_P.inverse(); // cache the inverse permutation
+#else
+  // If AMD is not available, (MPL2-only), then let's use the slower COLAMD routine.
+  SparseMatrix<Scalar,ColMajor, StorageIndex> mat1 = amat;
+  COLAMDOrdering<StorageIndex> ordering;
+  ordering(mat1,m_Pinv);
+  m_P = m_Pinv.inverse();
+#endif
 
   m_analysisIsOk = true;
   m_factorizationIsOk = false;
-  m_isInitialized = false;
+  m_isInitialized = true;
 }
 
-template <typename Scalar>
+template <typename Scalar, typename StorageIndex>
 template<typename _MatrixType>
-void IncompleteLUT<Scalar>::factorize(const _MatrixType& amat)
+void IncompleteLUT<Scalar,StorageIndex>::factorize(const _MatrixType& amat)
 {
   using std::sqrt;
   using std::swap;
   using std::abs;
+  using internal::convert_index;
 
   eigen_assert((amat.rows() == amat.cols()) && "The factorization should be done on a square matrix");
   Index n = amat.cols();  // Size of the matrix
   m_lu.resize(n,n);
   // Declare Working vectors and variables
   Vector u(n) ;     // real values of the row -- maximum size is n --
-  VectorXi ju(n);   // column position of the values in u -- maximum size  is n
-  VectorXi jr(n);   // Indicate the position of the nonzero elements in the vector u -- A zero location is indicated by -1
+  VectorI ju(n);   // column position of the values in u -- maximum size  is n
+  VectorI jr(n);   // Indicate the position of the nonzero elements in the vector u -- A zero location is indicated by -1
 
   // Apply the fill-reducing permutation
   eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
-  SparseMatrix<Scalar,RowMajor, Index> mat;
+  SparseMatrix<Scalar,RowMajor, StorageIndex> mat;
   mat = amat.twistedBy(m_Pinv);
 
   // Initialization
@@ -265,7 +276,7 @@
   u.fill(0);
 
   // number of largest elements to keep in each row:
-  Index fill_in =   static_cast<Index> (amat.nonZeros()*m_fillfactor)/n+1;
+  Index fill_in = (amat.nonZeros()*m_fillfactor)/n + 1;
   if (fill_in > n) fill_in = n;
 
   // number of largest nonzero elements to keep in the L and the U part of the current row:
@@ -280,9 +291,9 @@
 
     Index sizeu = 1; // number of nonzero elements in the upper part of the current row
     Index sizel = 0; // number of nonzero elements in the lower part of the current row
-    ju(ii)    = ii;
+    ju(ii)    = convert_index<StorageIndex>(ii);
     u(ii)     = 0;
-    jr(ii)    = ii;
+    jr(ii)    = convert_index<StorageIndex>(ii);
     RealScalar rownorm = 0;
 
     typename FactorType::InnerIterator j_it(mat, ii); // Iterate through the current row ii
@@ -292,9 +303,9 @@
       if (k < ii)
       {
         // copy the lower part
-        ju(sizel) = k;
+        ju(sizel) = convert_index<StorageIndex>(k);
         u(sizel) = j_it.value();
-        jr(k) = sizel;
+        jr(k) = convert_index<StorageIndex>(sizel);
         ++sizel;
       }
       else if (k == ii)
@@ -305,9 +316,9 @@
       {
         // copy the upper part
         Index jpos = ii + sizeu;
-        ju(jpos) = k;
+        ju(jpos) = convert_index<StorageIndex>(k);
         u(jpos) = j_it.value();
-        jr(k) = jpos;
+        jr(k) = convert_index<StorageIndex>(jpos);
         ++sizeu;
       }
       rownorm += numext::abs2(j_it.value());
@@ -337,7 +348,8 @@
         // swap the two locations
         Index j = ju(jj);
         swap(ju(jj), ju(k));
-        jr(minrow) = jj;   jr(j) = k;
+        jr(minrow) = convert_index<StorageIndex>(jj);
+        jr(j) = convert_index<StorageIndex>(k);
         swap(u(jj), u(k));
       }
       // Reset this location
@@ -361,8 +373,8 @@
       for (; ki_it; ++ki_it)
       {
         Scalar prod = fact * ki_it.value();
-        Index j       = ki_it.index();
-        Index jpos    = jr(j);
+        Index j     = ki_it.index();
+        Index jpos  = jr(j);
         if (jpos == -1) // fill-in element
         {
           Index newpos;
@@ -378,16 +390,16 @@
             sizel++;
             eigen_internal_assert(sizel<=ii);
           }
-          ju(newpos) = j;
+          ju(newpos) = convert_index<StorageIndex>(j);
           u(newpos) = -prod;
-          jr(j) = newpos;
+          jr(j) = convert_index<StorageIndex>(newpos);
         }
         else
           u(jpos) -= prod;
       }
       // store the pivot element
-      u(len) = fact;
-      ju(len) = minrow;
+      u(len)  = fact;
+      ju(len) = convert_index<StorageIndex>(minrow);
       ++len;
 
       jj++;
@@ -402,7 +414,7 @@
     sizel = len;
     len = (std::min)(sizel, nnzL);
     typename Vector::SegmentReturnType ul(u.segment(0, sizel));
-    typename VectorXi::SegmentReturnType jul(ju.segment(0, sizel));
+    typename VectorI::SegmentReturnType jul(ju.segment(0, sizel));
     internal::QuickSplit(ul, jul, len);
 
     // store the largest m_fill elements of the L part
@@ -431,39 +443,20 @@
     sizeu = len + 1; // +1 to take into account the diagonal element
     len = (std::min)(sizeu, nnzU);
     typename Vector::SegmentReturnType uu(u.segment(ii+1, sizeu-1));
-    typename VectorXi::SegmentReturnType juu(ju.segment(ii+1, sizeu-1));
+    typename VectorI::SegmentReturnType juu(ju.segment(ii+1, sizeu-1));
     internal::QuickSplit(uu, juu, len);
 
     // store the largest elements of the U part
     for(Index k = ii + 1; k < ii + len; k++)
       m_lu.insertBackByOuterInnerUnordered(ii,ju(k)) = u(k);
   }
-
   m_lu.finalize();
   m_lu.makeCompressed();
 
   m_factorizationIsOk = true;
-  m_isInitialized = m_factorizationIsOk;
   m_info = Success;
 }
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<IncompleteLUT<_MatrixType>, Rhs>
-  : solve_retval_base<IncompleteLUT<_MatrixType>, Rhs>
-{
-  typedef IncompleteLUT<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_INCOMPLETE_LUT_H
diff --git a/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h b/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h
index 2036922..7c2326e 100644
--- a/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h
+++ b/Eigen/src/IterativeLinearSolvers/IterativeSolverBase.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,29 +12,158 @@
 
 namespace Eigen { 
 
+namespace internal {
+
+template<typename MatrixType>
+struct is_ref_compatible_impl
+{
+private:
+  template <typename T0>
+  struct any_conversion
+  {
+    template <typename T> any_conversion(const volatile T&);
+    template <typename T> any_conversion(T&);
+  };
+  struct yes {int a[1];};
+  struct no  {int a[2];};
+
+  template<typename T>
+  static yes test(const Ref<const T>&, int);
+  template<typename T>
+  static no  test(any_conversion<T>, ...);
+
+public:
+  static MatrixType ms_from;
+  enum { value = sizeof(test<MatrixType>(ms_from, 0))==sizeof(yes) };
+};
+
+template<typename MatrixType>
+struct is_ref_compatible
+{
+  enum { value = is_ref_compatible_impl<typename remove_all<MatrixType>::type>::value };
+};
+
+template<typename MatrixType, bool MatrixFree = !internal::is_ref_compatible<MatrixType>::value>
+class generic_matrix_wrapper;
+
+// We have an explicit matrix at hand, compatible with Ref<>
+template<typename MatrixType>
+class generic_matrix_wrapper<MatrixType,false>
+{
+public:
+  typedef Ref<const MatrixType> ActualMatrixType;
+  template<int UpLo> struct ConstSelfAdjointViewReturnType {
+    typedef typename ActualMatrixType::template ConstSelfAdjointViewReturnType<UpLo>::Type Type;
+  };
+
+  enum {
+    MatrixFree = false
+  };
+
+  generic_matrix_wrapper()
+    : m_dummy(0,0), m_matrix(m_dummy)
+  {}
+
+  template<typename InputType>
+  generic_matrix_wrapper(const InputType &mat)
+    : m_matrix(mat)
+  {}
+
+  const ActualMatrixType& matrix() const
+  {
+    return m_matrix;
+  }
+
+  template<typename MatrixDerived>
+  void grab(const EigenBase<MatrixDerived> &mat)
+  {
+    m_matrix.~Ref<const MatrixType>();
+    ::new (&m_matrix) Ref<const MatrixType>(mat.derived());
+  }
+
+  void grab(const Ref<const MatrixType> &mat)
+  {
+    if(&(mat.derived()) != &m_matrix)
+    {
+      m_matrix.~Ref<const MatrixType>();
+      ::new (&m_matrix) Ref<const MatrixType>(mat);
+    }
+  }
+
+protected:
+  MatrixType m_dummy; // used to default initialize the Ref<> object
+  ActualMatrixType m_matrix;
+};
+
+// MatrixType is not compatible with Ref<> -> matrix-free wrapper
+template<typename MatrixType>
+class generic_matrix_wrapper<MatrixType,true>
+{
+public:
+  typedef MatrixType ActualMatrixType;
+  template<int UpLo> struct ConstSelfAdjointViewReturnType
+  {
+    typedef ActualMatrixType Type;
+  };
+
+  enum {
+    MatrixFree = true
+  };
+
+  generic_matrix_wrapper()
+    : mp_matrix(0)
+  {}
+
+  generic_matrix_wrapper(const MatrixType &mat)
+    : mp_matrix(&mat)
+  {}
+
+  const ActualMatrixType& matrix() const
+  {
+    return *mp_matrix;
+  }
+
+  void grab(const MatrixType &mat)
+  {
+    mp_matrix = &mat;
+  }
+
+protected:
+  const ActualMatrixType *mp_matrix;
+};
+
+}
+
 /** \ingroup IterativeLinearSolvers_Module
   * \brief Base class for linear iterative solvers
   *
   * \sa class SimplicialCholesky, DiagonalPreconditioner, IdentityPreconditioner
   */
 template< typename Derived>
-class IterativeSolverBase : internal::noncopyable
+class IterativeSolverBase : public SparseSolverBase<Derived>
 {
+protected:
+  typedef SparseSolverBase<Derived> Base;
+  using Base::m_isInitialized;
+  
 public:
   typedef typename internal::traits<Derived>::MatrixType MatrixType;
   typedef typename internal::traits<Derived>::Preconditioner Preconditioner;
   typedef typename MatrixType::Scalar Scalar;
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef typename MatrixType::RealScalar RealScalar;
 
+  enum {
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+  };
+
 public:
 
-  Derived& derived() { return *static_cast<Derived*>(this); }
-  const Derived& derived() const { return *static_cast<const Derived*>(this); }
+  using Base::derived;
 
   /** Default constructor. */
   IterativeSolverBase()
-    : mp_matrix(0)
   {
     init();
   }
@@ -49,77 +178,90 @@
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  IterativeSolverBase(const MatrixType& A)
+  template<typename MatrixDerived>
+  explicit IterativeSolverBase(const EigenBase<MatrixDerived>& A)
+    : m_matrixWrapper(A.derived())
   {
     init();
-    compute(A);
+    compute(matrix());
   }
 
   ~IterativeSolverBase() {}
   
-  /** Initializes the iterative solver for the sparcity pattern of the matrix \a A for further solving \c Ax=b problems.
+  /** Initializes the iterative solver for the sparsity pattern of the matrix \a A for further solving \c Ax=b problems.
     *
-    * Currently, this function mostly call analyzePattern on the preconditioner. In the future
-    * we might, for instance, implement column reodering for faster matrix vector products.
+    * Currently, this function mostly calls analyzePattern on the preconditioner. In the future
+    * we might, for instance, implement column reordering for faster matrix vector products.
     */
-  Derived& analyzePattern(const MatrixType& A)
+  template<typename MatrixDerived>
+  Derived& analyzePattern(const EigenBase<MatrixDerived>& A)
   {
-    m_preconditioner.analyzePattern(A);
+    grab(A.derived());
+    m_preconditioner.analyzePattern(matrix());
     m_isInitialized = true;
     m_analysisIsOk = true;
-    m_info = Success;
+    m_info = m_preconditioner.info();
     return derived();
   }
   
   /** Initializes the iterative solver with the numerical values of the matrix \a A for further solving \c Ax=b problems.
     *
-    * Currently, this function mostly call factorize on the preconditioner.
+    * Currently, this function mostly calls factorize on the preconditioner.
     *
     * \warning this class stores a reference to the matrix A as well as some
     * precomputed values that depend on it. Therefore, if \a A is changed
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  Derived& factorize(const MatrixType& A)
+  template<typename MatrixDerived>
+  Derived& factorize(const EigenBase<MatrixDerived>& A)
   {
     eigen_assert(m_analysisIsOk && "You must first call analyzePattern()"); 
-    mp_matrix = &A;
-    m_preconditioner.factorize(A);
+    grab(A.derived());
+    m_preconditioner.factorize(matrix());
     m_factorizationIsOk = true;
-    m_info = Success;
+    m_info = m_preconditioner.info();
     return derived();
   }
 
   /** Initializes the iterative solver with the matrix \a A for further solving \c Ax=b problems.
     *
-    * Currently, this function mostly initialized/compute the preconditioner. In the future
-    * we might, for instance, implement column reodering for faster matrix vector products.
+    * Currently, this function mostly initializes/computes the preconditioner. In the future
+    * we might, for instance, implement column reordering for faster matrix vector products.
     *
     * \warning this class stores a reference to the matrix A as well as some
     * precomputed values that depend on it. Therefore, if \a A is changed
     * this class becomes invalid. Call compute() to update it with the new
     * matrix A, or modify a copy of A.
     */
-  Derived& compute(const MatrixType& A)
+  template<typename MatrixDerived>
+  Derived& compute(const EigenBase<MatrixDerived>& A)
   {
-    mp_matrix = &A;
-    m_preconditioner.compute(A);
+    grab(A.derived());
+    m_preconditioner.compute(matrix());
     m_isInitialized = true;
     m_analysisIsOk = true;
     m_factorizationIsOk = true;
-    m_info = Success;
+    m_info = m_preconditioner.info();
     return derived();
   }
 
   /** \internal */
-  Index rows() const { return mp_matrix ? mp_matrix->rows() : 0; }
-  /** \internal */
-  Index cols() const { return mp_matrix ? mp_matrix->cols() : 0; }
+  Index rows() const { return matrix().rows(); }
 
-  /** \returns the tolerance threshold used by the stopping criteria */
+  /** \internal */
+  Index cols() const { return matrix().cols(); }
+
+  /** \returns the tolerance threshold used by the stopping criteria.
+    * \sa setTolerance()
+    */
   RealScalar tolerance() const { return m_tolerance; }
   
-  /** Sets the tolerance threshold used by the stopping criteria */
+  /** Sets the tolerance threshold used by the stopping criteria.
+    *
+    * This value is used as an upper bound to the relative residual error: |Ax-b|/|b|.
+    * The default value is the machine precision given by NumTraits<Scalar>::epsilon()
+    */
   Derived& setTolerance(const RealScalar& tolerance)
   {
     m_tolerance = tolerance;
@@ -132,58 +274,52 @@
   /** \returns a read-only reference to the preconditioner. */
   const Preconditioner& preconditioner() const { return m_preconditioner; }
 
-  /** \returns the max number of iterations */
-  int maxIterations() const
+  /** \returns the max number of iterations.
+    * It is either the value setted by setMaxIterations or, by default,
+    * twice the number of columns of the matrix.
+    */
+  Index maxIterations() const
   {
-    return (mp_matrix && m_maxIterations<0) ? mp_matrix->cols() : m_maxIterations;
+    return (m_maxIterations<0) ? 2*matrix().cols() : m_maxIterations;
   }
   
-  /** Sets the max number of iterations */
-  Derived& setMaxIterations(int maxIters)
+  /** Sets the max number of iterations.
+    * Default is twice the number of columns of the matrix.
+    */
+  Derived& setMaxIterations(Index maxIters)
   {
     m_maxIterations = maxIters;
     return derived();
   }
 
   /** \returns the number of iterations performed during the last solve */
-  int iterations() const
+  Index iterations() const
   {
     eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
     return m_iterations;
   }
 
-  /** \returns the tolerance error reached during the last solve */
+  /** \returns the tolerance error reached during the last solve.
+    * It is a close approximation of the true relative residual error |Ax-b|/|b|.
+    */
   RealScalar error() const
   {
     eigen_assert(m_isInitialized && "ConjugateGradient is not initialized.");
     return m_error;
   }
 
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
+  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A
+    * and \a x0 as an initial solution.
     *
-    * \sa compute()
+    * \sa solve(), compute()
     */
-  template<typename Rhs> inline const internal::solve_retval<Derived, Rhs>
-  solve(const MatrixBase<Rhs>& b) const
+  template<typename Rhs,typename Guess>
+  inline const SolveWithGuess<Derived, Rhs, Guess>
+  solveWithGuess(const MatrixBase<Rhs>& b, const Guess& x0) const
   {
-    eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
-    eigen_assert(rows()==b.rows()
-              && "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::solve_retval<Derived, Rhs>(derived(), b.derived());
-  }
-  
-  /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-    *
-    * \sa compute()
-    */
-  template<typename Rhs>
-  inline const internal::sparse_solve_retval<IterativeSolverBase, Rhs>
-  solve(const SparseMatrixBase<Rhs>& b) const
-  {
-    eigen_assert(m_isInitialized && "IterativeSolverBase is not initialized.");
-    eigen_assert(rows()==b.rows()
-              && "IterativeSolverBase::solve(): invalid number of rows of the right hand side matrix b");
-    return internal::sparse_solve_retval<IterativeSolverBase, Rhs>(*this, b.derived());
+    eigen_assert(m_isInitialized && "Solver is not initialized.");
+    eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+    return SolveWithGuess<Derived, Rhs, Guess>(derived(), b.derived(), x0);
   }
 
   /** \returns Success if the iterations converged, and NoConvergence otherwise. */
@@ -194,21 +330,27 @@
   }
   
   /** \internal */
-  template<typename Rhs, typename DestScalar, int DestOptions, typename DestIndex>
-  void _solve_sparse(const Rhs& b, SparseMatrix<DestScalar,DestOptions,DestIndex> &dest) const
+  template<typename Rhs, typename DestDerived>
+  void _solve_impl(const Rhs& b, SparseMatrixBase<DestDerived> &aDest) const
   {
     eigen_assert(rows()==b.rows());
     
-    int rhsCols = b.cols();
-    int size = b.rows();
+    Index rhsCols = b.cols();
+    Index size = b.rows();
+    DestDerived& dest(aDest.derived());
+    typedef typename DestDerived::Scalar DestScalar;
     Eigen::Matrix<DestScalar,Dynamic,1> tb(size);
-    Eigen::Matrix<DestScalar,Dynamic,1> tx(size);
-    for(int k=0; k<rhsCols; ++k)
+    Eigen::Matrix<DestScalar,Dynamic,1> tx(cols());
+    // We do not directly fill dest because sparse expressions have to be free of aliasing issue.
+    // For non square least-square problems, b and dest might not have the same size whereas they might alias each-other.
+    typename DestDerived::PlainObject tmp(cols(),rhsCols);
+    for(Index k=0; k<rhsCols; ++k)
     {
       tb = b.col(k);
       tx = derived().solve(tb);
-      dest.col(k) = tx.sparseView(0);
+      tmp.col(k) = tx.sparseView(0);
     }
+    dest.swap(tmp);
   }
 
 protected:
@@ -220,35 +362,33 @@
     m_maxIterations = -1;
     m_tolerance = NumTraits<Scalar>::epsilon();
   }
-  const MatrixType* mp_matrix;
+
+  typedef internal::generic_matrix_wrapper<MatrixType> MatrixWrapper;
+  typedef typename MatrixWrapper::ActualMatrixType ActualMatrixType;
+
+  const ActualMatrixType& matrix() const
+  {
+    return m_matrixWrapper.matrix();
+  }
+  
+  template<typename InputType>
+  void grab(const InputType &A)
+  {
+    m_matrixWrapper.grab(A);
+  }
+  
+  MatrixWrapper m_matrixWrapper;
   Preconditioner m_preconditioner;
 
-  int m_maxIterations;
+  Index m_maxIterations;
   RealScalar m_tolerance;
   
   mutable RealScalar m_error;
-  mutable int m_iterations;
+  mutable Index m_iterations;
   mutable ComputationInfo m_info;
-  mutable bool m_isInitialized, m_analysisIsOk, m_factorizationIsOk;
+  mutable bool m_analysisIsOk, m_factorizationIsOk;
 };
 
-namespace internal {
- 
-template<typename Derived, typename Rhs>
-struct sparse_solve_retval<IterativeSolverBase<Derived>, Rhs>
-  : sparse_solve_retval_base<IterativeSolverBase<Derived>, Rhs>
-{
-  typedef IterativeSolverBase<Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec().derived()._solve_sparse(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_ITERATIVE_SOLVER_BASE_H
diff --git a/Eigen/src/IterativeLinearSolvers/LeastSquareConjugateGradient.h b/Eigen/src/IterativeLinearSolvers/LeastSquareConjugateGradient.h
new file mode 100644
index 0000000..0aea0e0
--- /dev/null
+++ b/Eigen/src/IterativeLinearSolvers/LeastSquareConjugateGradient.h
@@ -0,0 +1,216 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
+#define EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
+
+namespace Eigen { 
+
+namespace internal {
+
+/** \internal Low-level conjugate gradient algorithm for least-square problems
+  * \param mat The matrix A
+  * \param rhs The right hand side vector b
+  * \param x On input and initial solution, on output the computed solution.
+  * \param precond A preconditioner being able to efficiently solve for an
+  *                approximation of A'Ax=b (regardless of b)
+  * \param iters On input the max number of iteration, on output the number of performed iterations.
+  * \param tol_error On input the tolerance error, on output an estimation of the relative error.
+  */
+template<typename MatrixType, typename Rhs, typename Dest, typename Preconditioner>
+EIGEN_DONT_INLINE
+void least_square_conjugate_gradient(const MatrixType& mat, const Rhs& rhs, Dest& x,
+                                     const Preconditioner& precond, Index& iters,
+                                     typename Dest::RealScalar& tol_error)
+{
+  using std::sqrt;
+  using std::abs;
+  typedef typename Dest::RealScalar RealScalar;
+  typedef typename Dest::Scalar Scalar;
+  typedef Matrix<Scalar,Dynamic,1> VectorType;
+  
+  RealScalar tol = tol_error;
+  Index maxIters = iters;
+  
+  Index m = mat.rows(), n = mat.cols();
+
+  VectorType residual        = rhs - mat * x;
+  VectorType normal_residual = mat.adjoint() * residual;
+
+  RealScalar rhsNorm2 = (mat.adjoint()*rhs).squaredNorm();
+  if(rhsNorm2 == 0) 
+  {
+    x.setZero();
+    iters = 0;
+    tol_error = 0;
+    return;
+  }
+  RealScalar threshold = tol*tol*rhsNorm2;
+  RealScalar residualNorm2 = normal_residual.squaredNorm();
+  if (residualNorm2 < threshold)
+  {
+    iters = 0;
+    tol_error = sqrt(residualNorm2 / rhsNorm2);
+    return;
+  }
+  
+  VectorType p(n);
+  p = precond.solve(normal_residual);                         // initial search direction
+
+  VectorType z(n), tmp(m);
+  RealScalar absNew = numext::real(normal_residual.dot(p));  // the square of the absolute value of r scaled by invM
+  Index i = 0;
+  while(i < maxIters)
+  {
+    tmp.noalias() = mat * p;
+
+    Scalar alpha = absNew / tmp.squaredNorm();      // the amount we travel on dir
+    x += alpha * p;                                 // update solution
+    residual -= alpha * tmp;                        // update residual
+    normal_residual = mat.adjoint() * residual;     // update residual of the normal equation
+    
+    residualNorm2 = normal_residual.squaredNorm();
+    if(residualNorm2 < threshold)
+      break;
+    
+    z = precond.solve(normal_residual);             // approximately solve for "A'A z = normal_residual"
+
+    RealScalar absOld = absNew;
+    absNew = numext::real(normal_residual.dot(z));  // update the absolute value of r
+    RealScalar beta = absNew / absOld;              // calculate the Gram-Schmidt value used to create the new search direction
+    p = z + beta * p;                               // update search direction
+    i++;
+  }
+  tol_error = sqrt(residualNorm2 / rhsNorm2);
+  iters = i;
+}
+
+}
+
+template< typename _MatrixType,
+          typename _Preconditioner = LeastSquareDiagonalPreconditioner<typename _MatrixType::Scalar> >
+class LeastSquaresConjugateGradient;
+
+namespace internal {
+
+template< typename _MatrixType, typename _Preconditioner>
+struct traits<LeastSquaresConjugateGradient<_MatrixType,_Preconditioner> >
+{
+  typedef _MatrixType MatrixType;
+  typedef _Preconditioner Preconditioner;
+};
+
+}
+
+/** \ingroup IterativeLinearSolvers_Module
+  * \brief A conjugate gradient solver for sparse (or dense) least-square problems
+  *
+  * This class allows to solve for A x = b linear problems using an iterative conjugate gradient algorithm.
+  * The matrix A can be non symmetric and rectangular, but the matrix A' A should be positive-definite to guaranty stability.
+  * Otherwise, the SparseLU or SparseQR classes might be preferable.
+  * The matrix A and the vectors x and b can be either dense or sparse.
+  *
+  * \tparam _MatrixType the type of the matrix A, can be a dense or a sparse matrix.
+  * \tparam _Preconditioner the type of the preconditioner. Default is LeastSquareDiagonalPreconditioner
+  *
+  * \implsparsesolverconcept
+  * 
+  * The maximal number of iterations and tolerance value can be controlled via the setMaxIterations()
+  * and setTolerance() methods. The defaults are the size of the problem for the maximal number of iterations
+  * and NumTraits<Scalar>::epsilon() for the tolerance.
+  * 
+  * This class can be used as the direct solver classes. Here is a typical usage example:
+    \code
+    int m=1000000, n = 10000;
+    VectorXd x(n), b(m);
+    SparseMatrix<double> A(m,n);
+    // fill A and b
+    LeastSquaresConjugateGradient<SparseMatrix<double> > lscg;
+    lscg.compute(A);
+    x = lscg.solve(b);
+    std::cout << "#iterations:     " << lscg.iterations() << std::endl;
+    std::cout << "estimated error: " << lscg.error()      << std::endl;
+    // update b, and solve again
+    x = lscg.solve(b);
+    \endcode
+  * 
+  * By default the iterations start with x=0 as an initial guess of the solution.
+  * One can control the start using the solveWithGuess() method.
+  * 
+  * \sa class ConjugateGradient, SparseLU, SparseQR
+  */
+template< typename _MatrixType, typename _Preconditioner>
+class LeastSquaresConjugateGradient : public IterativeSolverBase<LeastSquaresConjugateGradient<_MatrixType,_Preconditioner> >
+{
+  typedef IterativeSolverBase<LeastSquaresConjugateGradient> Base;
+  using Base::matrix;
+  using Base::m_error;
+  using Base::m_iterations;
+  using Base::m_info;
+  using Base::m_isInitialized;
+public:
+  typedef _MatrixType MatrixType;
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef _Preconditioner Preconditioner;
+
+public:
+
+  /** Default constructor. */
+  LeastSquaresConjugateGradient() : Base() {}
+
+  /** Initialize the solver with matrix \a A for further \c Ax=b solving.
+    * 
+    * This constructor is a shortcut for the default constructor followed
+    * by a call to compute().
+    * 
+    * \warning this class stores a reference to the matrix A as well as some
+    * precomputed values that depend on it. Therefore, if \a A is changed
+    * this class becomes invalid. Call compute() to update it with the new
+    * matrix A, or modify a copy of A.
+    */
+  template<typename MatrixDerived>
+  explicit LeastSquaresConjugateGradient(const EigenBase<MatrixDerived>& A) : Base(A.derived()) {}
+
+  ~LeastSquaresConjugateGradient() {}
+
+  /** \internal */
+  template<typename Rhs,typename Dest>
+  void _solve_with_guess_impl(const Rhs& b, Dest& x) const
+  {
+    m_iterations = Base::maxIterations();
+    m_error = Base::m_tolerance;
+
+    for(Index j=0; j<b.cols(); ++j)
+    {
+      m_iterations = Base::maxIterations();
+      m_error = Base::m_tolerance;
+
+      typename Dest::ColXpr xj(x,j);
+      internal::least_square_conjugate_gradient(matrix(), b.col(j), xj, Base::m_preconditioner, m_iterations, m_error);
+    }
+
+    m_isInitialized = true;
+    m_info = m_error <= Base::m_tolerance ? Success : NoConvergence;
+  }
+  
+  /** \internal */
+  using Base::_solve_impl;
+  template<typename Rhs,typename Dest>
+  void _solve_impl(const MatrixBase<Rhs>& b, Dest& x) const
+  {
+    x.setZero();
+    _solve_with_guess_impl(b.derived(),x);
+  }
+
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_LEAST_SQUARE_CONJUGATE_GRADIENT_H
diff --git a/Eigen/src/IterativeLinearSolvers/SolveWithGuess.h b/Eigen/src/IterativeLinearSolvers/SolveWithGuess.h
new file mode 100644
index 0000000..0ace451
--- /dev/null
+++ b/Eigen/src/IterativeLinearSolvers/SolveWithGuess.h
@@ -0,0 +1,115 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SOLVEWITHGUESS_H
+#define EIGEN_SOLVEWITHGUESS_H
+
+namespace Eigen {
+
+template<typename Decomposition, typename RhsType, typename GuessType> class SolveWithGuess;
+  
+/** \class SolveWithGuess
+  * \ingroup IterativeLinearSolvers_Module
+  *
+  * \brief Pseudo expression representing a solving operation
+  *
+  * \tparam Decomposition the type of the matrix or decomposion object
+  * \tparam Rhstype the type of the right-hand side
+  *
+  * This class represents an expression of A.solve(B)
+  * and most of the time this is the only way it is used.
+  *
+  */
+namespace internal {
+
+
+template<typename Decomposition, typename RhsType, typename GuessType>
+struct traits<SolveWithGuess<Decomposition, RhsType, GuessType> >
+  : traits<Solve<Decomposition,RhsType> >
+{};
+
+}
+
+
+template<typename Decomposition, typename RhsType, typename GuessType>
+class SolveWithGuess : public internal::generic_xpr_base<SolveWithGuess<Decomposition,RhsType,GuessType>, MatrixXpr, typename internal::traits<RhsType>::StorageKind>::type
+{
+public:
+  typedef typename internal::traits<SolveWithGuess>::Scalar Scalar;
+  typedef typename internal::traits<SolveWithGuess>::PlainObject PlainObject;
+  typedef typename internal::generic_xpr_base<SolveWithGuess<Decomposition,RhsType,GuessType>, MatrixXpr, typename internal::traits<RhsType>::StorageKind>::type Base;
+  typedef typename internal::ref_selector<SolveWithGuess>::type Nested;
+  
+  SolveWithGuess(const Decomposition &dec, const RhsType &rhs, const GuessType &guess)
+    : m_dec(dec), m_rhs(rhs), m_guess(guess)
+  {}
+  
+  EIGEN_DEVICE_FUNC Index rows() const { return m_dec.cols(); }
+  EIGEN_DEVICE_FUNC Index cols() const { return m_rhs.cols(); }
+
+  EIGEN_DEVICE_FUNC const Decomposition& dec()   const { return m_dec; }
+  EIGEN_DEVICE_FUNC const RhsType&       rhs()   const { return m_rhs; }
+  EIGEN_DEVICE_FUNC const GuessType&     guess() const { return m_guess; }
+
+protected:
+  const Decomposition &m_dec;
+  const RhsType       &m_rhs;
+  const GuessType     &m_guess;
+  
+private:
+  Scalar coeff(Index row, Index col) const;
+  Scalar coeff(Index i) const;
+};
+
+namespace internal {
+
+// Evaluator of SolveWithGuess -> eval into a temporary
+template<typename Decomposition, typename RhsType, typename GuessType>
+struct evaluator<SolveWithGuess<Decomposition,RhsType, GuessType> >
+  : public evaluator<typename SolveWithGuess<Decomposition,RhsType,GuessType>::PlainObject>
+{
+  typedef SolveWithGuess<Decomposition,RhsType,GuessType> SolveType;
+  typedef typename SolveType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  evaluator(const SolveType& solve)
+    : m_result(solve.rows(), solve.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    m_result = solve.guess();
+    solve.dec()._solve_with_guess_impl(solve.rhs(), m_result);
+  }
+  
+protected:  
+  PlainObject m_result;
+};
+
+// Specialization for "dst = dec.solveWithGuess(rhs)"
+// NOTE we need to specialize it for Dense2Dense to avoid ambiguous specialization error and a Sparse2Sparse specialization must exist somewhere
+template<typename DstXprType, typename DecType, typename RhsType, typename GuessType, typename Scalar>
+struct Assignment<DstXprType, SolveWithGuess<DecType,RhsType,GuessType>, internal::assign_op<Scalar,Scalar>, Dense2Dense>
+{
+  typedef SolveWithGuess<DecType,RhsType,GuessType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    dst = src.guess();
+    src.dec()._solve_with_guess_impl(src.rhs(), dst/*, src.guess()*/);
+  }
+};
+
+} // end namepsace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SOLVEWITHGUESS_H
diff --git a/Eigen/src/Jacobi/CMakeLists.txt b/Eigen/src/Jacobi/CMakeLists.txt
deleted file mode 100644
index 490dac6..0000000
--- a/Eigen/src/Jacobi/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_Jacobi_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_Jacobi_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Jacobi COMPONENT Devel
-  )
diff --git a/Eigen/src/Jacobi/Jacobi.h b/Eigen/src/Jacobi/Jacobi.h
index 956f72d..1998c63 100644
--- a/Eigen/src/Jacobi/Jacobi.h
+++ b/Eigen/src/Jacobi/Jacobi.h
@@ -62,14 +62,14 @@
     JacobiRotation adjoint() const { using numext::conj; return JacobiRotation(conj(m_c), -m_s); }
 
     template<typename Derived>
-    bool makeJacobi(const MatrixBase<Derived>&, typename Derived::Index p, typename Derived::Index q);
+    bool makeJacobi(const MatrixBase<Derived>&, Index p, Index q);
     bool makeJacobi(const RealScalar& x, const Scalar& y, const RealScalar& z);
 
-    void makeGivens(const Scalar& p, const Scalar& q, Scalar* z=0);
+    void makeGivens(const Scalar& p, const Scalar& q, Scalar* r=0);
 
   protected:
-    void makeGivens(const Scalar& p, const Scalar& q, Scalar* z, internal::true_type);
-    void makeGivens(const Scalar& p, const Scalar& q, Scalar* z, internal::false_type);
+    void makeGivens(const Scalar& p, const Scalar& q, Scalar* r, internal::true_type);
+    void makeGivens(const Scalar& p, const Scalar& q, Scalar* r, internal::false_type);
 
     Scalar m_c, m_s;
 };
@@ -84,8 +84,8 @@
 {
   using std::sqrt;
   using std::abs;
-  typedef typename NumTraits<Scalar>::Real RealScalar;
-  if(y == Scalar(0))
+  RealScalar deno = RealScalar(2)*abs(y);
+  if(deno < (std::numeric_limits<RealScalar>::min)())
   {
     m_c = Scalar(1);
     m_s = Scalar(0);
@@ -93,7 +93,7 @@
   }
   else
   {
-    RealScalar tau = (x-z)/(RealScalar(2)*abs(y));
+    RealScalar tau = (x-z)/deno;
     RealScalar w = sqrt(numext::abs2(tau) + RealScalar(1));
     RealScalar t;
     if(tau>RealScalar(0))
@@ -123,7 +123,7 @@
   */
 template<typename Scalar>
 template<typename Derived>
-inline bool JacobiRotation<Scalar>::makeJacobi(const MatrixBase<Derived>& m, typename Derived::Index p, typename Derived::Index q)
+inline bool JacobiRotation<Scalar>::makeJacobi(const MatrixBase<Derived>& m, Index p, Index q)
 {
   return makeJacobi(numext::real(m.coeff(p,p)), m.coeff(p,q), numext::real(m.coeff(q,q)));
 }
@@ -132,7 +132,7 @@
   * \f$ V = \left ( \begin{array}{c} p \\ q \end{array} \right )\f$ yields:
   * \f$ G^* V = \left ( \begin{array}{c} r \\ 0 \end{array} \right )\f$.
   *
-  * The value of \a z is returned if \a z is not null (the default is null).
+  * The value of \a r is returned if \a r is not null (the default is null).
   * Also note that G is built such that the cosine is always real.
   *
   * Example: \include Jacobi_makeGivens.cpp
@@ -145,9 +145,9 @@
   * \sa MatrixBase::applyOnTheLeft(), MatrixBase::applyOnTheRight()
   */
 template<typename Scalar>
-void JacobiRotation<Scalar>::makeGivens(const Scalar& p, const Scalar& q, Scalar* z)
+void JacobiRotation<Scalar>::makeGivens(const Scalar& p, const Scalar& q, Scalar* r)
 {
-  makeGivens(p, q, z, typename internal::conditional<NumTraits<Scalar>::IsComplex, internal::true_type, internal::false_type>::type());
+  makeGivens(p, q, r, typename internal::conditional<NumTraits<Scalar>::IsComplex, internal::true_type, internal::false_type>::type());
 }
 
 
@@ -255,15 +255,15 @@
 *   Implementation of MatrixBase methods
 ****************************************************************************************/
 
+namespace internal {
 /** \jacobi_module
   * Applies the clock wise 2D rotation \a j to the set of 2D vectors of cordinates \a x and \a y:
   * \f$ \left ( \begin{array}{cc} x \\ y \end{array} \right )  =  J \left ( \begin{array}{cc} x \\ y \end{array} \right ) \f$
   *
   * \sa MatrixBase::applyOnTheLeft(), MatrixBase::applyOnTheRight()
   */
-namespace internal {
 template<typename VectorX, typename VectorY, typename OtherScalar>
-void apply_rotation_in_the_plane(VectorX& _x, VectorY& _y, const JacobiRotation<OtherScalar>& j);
+void apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x, DenseBase<VectorY>& xpr_y, const JacobiRotation<OtherScalar>& j);
 }
 
 /** \jacobi_module
@@ -297,122 +297,12 @@
 }
 
 namespace internal {
-template<typename VectorX, typename VectorY, typename OtherScalar>
-void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(VectorX& _x, VectorY& _y, const JacobiRotation<OtherScalar>& j)
+
+template<typename Scalar, typename OtherScalar,
+         int SizeAtCompileTime, int MinAlignment, bool Vectorizable>
+struct apply_rotation_in_the_plane_selector
 {
-  typedef typename VectorX::Index Index;
-  typedef typename VectorX::Scalar Scalar;
-  enum { PacketSize = packet_traits<Scalar>::size };
-  typedef typename packet_traits<Scalar>::type Packet;
-  eigen_assert(_x.size() == _y.size());
-  Index size = _x.size();
-  Index incrx = _x.innerStride();
-  Index incry = _y.innerStride();
-
-  Scalar* EIGEN_RESTRICT x = &_x.coeffRef(0);
-  Scalar* EIGEN_RESTRICT y = &_y.coeffRef(0);
-  
-  OtherScalar c = j.c();
-  OtherScalar s = j.s();
-  if (c==OtherScalar(1) && s==OtherScalar(0))
-    return;
-
-  /*** dynamic-size vectorized paths ***/
-
-  if(VectorX::SizeAtCompileTime == Dynamic &&
-    (VectorX::Flags & VectorY::Flags & PacketAccessBit) &&
-    ((incrx==1 && incry==1) || PacketSize == 1))
-  {
-    // both vectors are sequentially stored in memory => vectorization
-    enum { Peeling = 2 };
-
-    Index alignedStart = internal::first_aligned(y, size);
-    Index alignedEnd = alignedStart + ((size-alignedStart)/PacketSize)*PacketSize;
-
-    const Packet pc = pset1<Packet>(c);
-    const Packet ps = pset1<Packet>(s);
-    conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex,false> pcj;
-
-    for(Index i=0; i<alignedStart; ++i)
-    {
-      Scalar xi = x[i];
-      Scalar yi = y[i];
-      x[i] =  c * xi + numext::conj(s) * yi;
-      y[i] = -s * xi + numext::conj(c) * yi;
-    }
-
-    Scalar* EIGEN_RESTRICT px = x + alignedStart;
-    Scalar* EIGEN_RESTRICT py = y + alignedStart;
-
-    if(internal::first_aligned(x, size)==alignedStart)
-    {
-      for(Index i=alignedStart; i<alignedEnd; i+=PacketSize)
-      {
-        Packet xi = pload<Packet>(px);
-        Packet yi = pload<Packet>(py);
-        pstore(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstore(py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-        px += PacketSize;
-        py += PacketSize;
-      }
-    }
-    else
-    {
-      Index peelingEnd = alignedStart + ((size-alignedStart)/(Peeling*PacketSize))*(Peeling*PacketSize);
-      for(Index i=alignedStart; i<peelingEnd; i+=Peeling*PacketSize)
-      {
-        Packet xi   = ploadu<Packet>(px);
-        Packet xi1  = ploadu<Packet>(px+PacketSize);
-        Packet yi   = pload <Packet>(py);
-        Packet yi1  = pload <Packet>(py+PacketSize);
-        pstoreu(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstoreu(px+PacketSize, padd(pmul(pc,xi1),pcj.pmul(ps,yi1)));
-        pstore (py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-        pstore (py+PacketSize, psub(pcj.pmul(pc,yi1),pmul(ps,xi1)));
-        px += Peeling*PacketSize;
-        py += Peeling*PacketSize;
-      }
-      if(alignedEnd!=peelingEnd)
-      {
-        Packet xi = ploadu<Packet>(x+peelingEnd);
-        Packet yi = pload <Packet>(y+peelingEnd);
-        pstoreu(x+peelingEnd, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-        pstore (y+peelingEnd, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-      }
-    }
-
-    for(Index i=alignedEnd; i<size; ++i)
-    {
-      Scalar xi = x[i];
-      Scalar yi = y[i];
-      x[i] =  c * xi + numext::conj(s) * yi;
-      y[i] = -s * xi + numext::conj(c) * yi;
-    }
-  }
-
-  /*** fixed-size vectorized path ***/
-  else if(VectorX::SizeAtCompileTime != Dynamic &&
-          (VectorX::Flags & VectorY::Flags & PacketAccessBit) &&
-          (VectorX::Flags & VectorY::Flags & AlignedBit))
-  {
-    const Packet pc = pset1<Packet>(c);
-    const Packet ps = pset1<Packet>(s);
-    conj_helper<Packet,Packet,NumTraits<Scalar>::IsComplex,false> pcj;
-    Scalar* EIGEN_RESTRICT px = x;
-    Scalar* EIGEN_RESTRICT py = y;
-    for(Index i=0; i<size; i+=PacketSize)
-    {
-      Packet xi = pload<Packet>(px);
-      Packet yi = pload<Packet>(py);
-      pstore(px, padd(pmul(pc,xi),pcj.pmul(ps,yi)));
-      pstore(py, psub(pcj.pmul(pc,yi),pmul(ps,xi)));
-      px += PacketSize;
-      py += PacketSize;
-    }
-  }
-
-  /*** non-vectorized path ***/
-  else
+  static inline void run(Scalar *x, Index incrx, Scalar *y, Index incry, Index size, OtherScalar c, OtherScalar s)
   {
     for(Index i=0; i<size; ++i)
     {
@@ -424,6 +314,145 @@
       y += incry;
     }
   }
+};
+
+template<typename Scalar, typename OtherScalar,
+         int SizeAtCompileTime, int MinAlignment>
+struct apply_rotation_in_the_plane_selector<Scalar,OtherScalar,SizeAtCompileTime,MinAlignment,true /* vectorizable */>
+{
+  static inline void run(Scalar *x, Index incrx, Scalar *y, Index incry, Index size, OtherScalar c, OtherScalar s)
+  {
+    enum {
+      PacketSize = packet_traits<Scalar>::size,
+      OtherPacketSize = packet_traits<OtherScalar>::size
+    };
+    typedef typename packet_traits<Scalar>::type Packet;
+    typedef typename packet_traits<OtherScalar>::type OtherPacket;
+
+    /*** dynamic-size vectorized paths ***/
+    if(SizeAtCompileTime == Dynamic && ((incrx==1 && incry==1) || PacketSize == 1))
+    {
+      // both vectors are sequentially stored in memory => vectorization
+      enum { Peeling = 2 };
+
+      Index alignedStart = internal::first_default_aligned(y, size);
+      Index alignedEnd = alignedStart + ((size-alignedStart)/PacketSize)*PacketSize;
+
+      const OtherPacket pc = pset1<OtherPacket>(c);
+      const OtherPacket ps = pset1<OtherPacket>(s);
+      conj_helper<OtherPacket,Packet,NumTraits<OtherScalar>::IsComplex,false> pcj;
+      conj_helper<OtherPacket,Packet,false,false> pm;
+
+      for(Index i=0; i<alignedStart; ++i)
+      {
+        Scalar xi = x[i];
+        Scalar yi = y[i];
+        x[i] =  c * xi + numext::conj(s) * yi;
+        y[i] = -s * xi + numext::conj(c) * yi;
+      }
+
+      Scalar* EIGEN_RESTRICT px = x + alignedStart;
+      Scalar* EIGEN_RESTRICT py = y + alignedStart;
+
+      if(internal::first_default_aligned(x, size)==alignedStart)
+      {
+        for(Index i=alignedStart; i<alignedEnd; i+=PacketSize)
+        {
+          Packet xi = pload<Packet>(px);
+          Packet yi = pload<Packet>(py);
+          pstore(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+          pstore(py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+          px += PacketSize;
+          py += PacketSize;
+        }
+      }
+      else
+      {
+        Index peelingEnd = alignedStart + ((size-alignedStart)/(Peeling*PacketSize))*(Peeling*PacketSize);
+        for(Index i=alignedStart; i<peelingEnd; i+=Peeling*PacketSize)
+        {
+          Packet xi   = ploadu<Packet>(px);
+          Packet xi1  = ploadu<Packet>(px+PacketSize);
+          Packet yi   = pload <Packet>(py);
+          Packet yi1  = pload <Packet>(py+PacketSize);
+          pstoreu(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+          pstoreu(px+PacketSize, padd(pm.pmul(pc,xi1),pcj.pmul(ps,yi1)));
+          pstore (py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+          pstore (py+PacketSize, psub(pcj.pmul(pc,yi1),pm.pmul(ps,xi1)));
+          px += Peeling*PacketSize;
+          py += Peeling*PacketSize;
+        }
+        if(alignedEnd!=peelingEnd)
+        {
+          Packet xi = ploadu<Packet>(x+peelingEnd);
+          Packet yi = pload <Packet>(y+peelingEnd);
+          pstoreu(x+peelingEnd, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+          pstore (y+peelingEnd, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+        }
+      }
+
+      for(Index i=alignedEnd; i<size; ++i)
+      {
+        Scalar xi = x[i];
+        Scalar yi = y[i];
+        x[i] =  c * xi + numext::conj(s) * yi;
+        y[i] = -s * xi + numext::conj(c) * yi;
+      }
+    }
+
+    /*** fixed-size vectorized path ***/
+    else if(SizeAtCompileTime != Dynamic && MinAlignment>0) // FIXME should be compared to the required alignment
+    {
+      const OtherPacket pc = pset1<OtherPacket>(c);
+      const OtherPacket ps = pset1<OtherPacket>(s);
+      conj_helper<OtherPacket,Packet,NumTraits<OtherPacket>::IsComplex,false> pcj;
+      conj_helper<OtherPacket,Packet,false,false> pm;
+      Scalar* EIGEN_RESTRICT px = x;
+      Scalar* EIGEN_RESTRICT py = y;
+      for(Index i=0; i<size; i+=PacketSize)
+      {
+        Packet xi = pload<Packet>(px);
+        Packet yi = pload<Packet>(py);
+        pstore(px, padd(pm.pmul(pc,xi),pcj.pmul(ps,yi)));
+        pstore(py, psub(pcj.pmul(pc,yi),pm.pmul(ps,xi)));
+        px += PacketSize;
+        py += PacketSize;
+      }
+    }
+
+    /*** non-vectorized path ***/
+    else
+    {
+      apply_rotation_in_the_plane_selector<Scalar,OtherScalar,SizeAtCompileTime,MinAlignment,false>::run(x,incrx,y,incry,size,c,s);
+    }
+  }
+};
+
+template<typename VectorX, typename VectorY, typename OtherScalar>
+void /*EIGEN_DONT_INLINE*/ apply_rotation_in_the_plane(DenseBase<VectorX>& xpr_x, DenseBase<VectorY>& xpr_y, const JacobiRotation<OtherScalar>& j)
+{
+  typedef typename VectorX::Scalar Scalar;
+  const bool Vectorizable =    (VectorX::Flags & VectorY::Flags & PacketAccessBit)
+                            && (int(packet_traits<Scalar>::size) == int(packet_traits<OtherScalar>::size));
+
+  eigen_assert(xpr_x.size() == xpr_y.size());
+  Index size = xpr_x.size();
+  Index incrx = xpr_x.derived().innerStride();
+  Index incry = xpr_y.derived().innerStride();
+
+  Scalar* EIGEN_RESTRICT x = &xpr_x.derived().coeffRef(0);
+  Scalar* EIGEN_RESTRICT y = &xpr_y.derived().coeffRef(0);
+  
+  OtherScalar c = j.c();
+  OtherScalar s = j.s();
+  if (c==OtherScalar(1) && s==OtherScalar(0))
+    return;
+
+  apply_rotation_in_the_plane_selector<
+    Scalar,OtherScalar,
+    VectorX::SizeAtCompileTime,
+    EIGEN_PLAIN_ENUM_MIN(evaluator<VectorX>::Alignment, evaluator<VectorY>::Alignment),
+    Vectorizable>::run(x,incrx,y,incry,size,c,s);
 }
 
 } // end namespace internal
diff --git a/Eigen/src/LU/CMakeLists.txt b/Eigen/src/LU/CMakeLists.txt
deleted file mode 100644
index e0d8d78..0000000
--- a/Eigen/src/LU/CMakeLists.txt
+++ /dev/null
@@ -1,8 +0,0 @@
-FILE(GLOB Eigen_LU_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_LU_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/LU COMPONENT Devel
-  )
-
-ADD_SUBDIRECTORY(arch)
diff --git a/Eigen/src/LU/Determinant.h b/Eigen/src/LU/Determinant.h
index bb8e78a..d6a3c1e 100644
--- a/Eigen/src/LU/Determinant.h
+++ b/Eigen/src/LU/Determinant.h
@@ -92,7 +92,7 @@
 inline typename internal::traits<Derived>::Scalar MatrixBase<Derived>::determinant() const
 {
   eigen_assert(rows() == cols());
-  typedef typename internal::nested<Derived,Base::RowsAtCompileTime>::type Nested;
+  typedef typename internal::nested_eval<Derived,Base::RowsAtCompileTime>::type Nested;
   return internal::determinant_impl<typename internal::remove_all<Nested>::type>::run(derived());
 }
 
diff --git a/Eigen/src/LU/FullPivLU.h b/Eigen/src/LU/FullPivLU.h
index 26bc714..03b6af7 100644
--- a/Eigen/src/LU/FullPivLU.h
+++ b/Eigen/src/LU/FullPivLU.h
@@ -10,7 +10,18 @@
 #ifndef EIGEN_LU_H
 #define EIGEN_LU_H
 
-namespace Eigen { 
+namespace Eigen {
+
+namespace internal {
+template<typename _MatrixType> struct traits<FullPivLU<_MatrixType> >
+ : traits<_MatrixType>
+{
+  typedef MatrixXpr XprKind;
+  typedef SolverStorage StorageKind;
+  enum { Flags = 0 };
+};
+
+} // end namespace internal
 
 /** \ingroup LU_Module
   *
@@ -18,7 +29,7 @@
   *
   * \brief LU decomposition of a matrix with complete pivoting, and related features
   *
-  * \param MatrixType the type of the matrix of which we are computing the LU decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the LU decomposition
   *
   * This class represents a LU decomposition of any matrix, with complete pivoting: the matrix A is
   * decomposed as \f$ A = P^{-1} L U Q^{-1} \f$ where L is unit-lower-triangular, U is
@@ -41,27 +52,28 @@
   * \include class_FullPivLU.cpp
   * Output: \verbinclude class_FullPivLU.out
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::fullPivLu(), MatrixBase::determinant(), MatrixBase::inverse()
   */
 template<typename _MatrixType> class FullPivLU
+  : public SolverBase<FullPivLU<_MatrixType> >
 {
   public:
     typedef _MatrixType MatrixType;
+    typedef SolverBase<FullPivLU> Base;
+
+    EIGEN_GENERIC_PUBLIC_INTERFACE(FullPivLU)
+    // FIXME StorageIndex defined in EIGEN_GENERIC_PUBLIC_INTERFACE should be int
     enum {
-      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
-      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
-    typedef typename MatrixType::Scalar Scalar;
-    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename internal::traits<MatrixType>::StorageKind StorageKind;
-    typedef typename MatrixType::Index Index;
-    typedef typename internal::plain_row_type<MatrixType, Index>::type IntRowVectorType;
-    typedef typename internal::plain_col_type<MatrixType, Index>::type IntColVectorType;
+    typedef typename internal::plain_row_type<MatrixType, StorageIndex>::type IntRowVectorType;
+    typedef typename internal::plain_col_type<MatrixType, StorageIndex>::type IntColVectorType;
     typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationQType;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationPType;
+    typedef typename MatrixType::PlainObject PlainObject;
 
     /**
       * \brief Default Constructor.
@@ -84,7 +96,17 @@
       * \param matrix the matrix of which to compute the LU decomposition.
       *               It is required to be nonzero.
       */
-    FullPivLU(const MatrixType& matrix);
+    template<typename InputType>
+    explicit FullPivLU(const EigenBase<InputType>& matrix);
+
+    /** \brief Constructs a LU factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa FullPivLU(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit FullPivLU(EigenBase<InputType>& matrix);
 
     /** Computes the LU decomposition of the given matrix.
       *
@@ -93,7 +115,12 @@
       *
       * \returns a reference to *this
       */
-    FullPivLU& compute(const MatrixType& matrix);
+    template<typename InputType>
+    FullPivLU& compute(const EigenBase<InputType>& matrix) {
+      m_lu = matrix.derived();
+      computeInPlace();
+      return *this;
+    }
 
     /** \returns the LU decomposition matrix: the upper-triangular part is U, the
       * unit-lower-triangular part is L (at least for square matrices; in the non-square
@@ -129,7 +156,7 @@
       *
       * \sa permutationQ()
       */
-    inline const PermutationPType& permutationP() const
+    EIGEN_DEVICE_FUNC inline const PermutationPType& permutationP() const
     {
       eigen_assert(m_isInitialized && "LU is not initialized.");
       return m_p;
@@ -166,7 +193,7 @@
     }
 
     /** \returns the image of the matrix, also called its column-space. The columns of the returned matrix
-      * will form a basis of the kernel.
+      * will form a basis of the image (column-space).
       *
       * \param originalMatrix the original matrix, of which *this is the LU decomposition.
       *                       The reason why it is needed to pass it here, is that this allows
@@ -210,12 +237,22 @@
       *
       * \sa TriangularView::solve(), kernel(), inverse()
       */
+    // FIXME this is a copy-paste of the base-class member to add the isInitialized assertion.
     template<typename Rhs>
-    inline const internal::solve_retval<FullPivLU, Rhs>
+    inline const Solve<FullPivLU, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "LU is not initialized.");
-      return internal::solve_retval<FullPivLU, Rhs>(*this, b.derived());
+      return Solve<FullPivLU, Rhs>(*this, b.derived());
+    }
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of which \c *this is
+        the LU decomposition.
+      */
+    inline RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
     }
 
     /** \returns the determinant of the matrix of which
@@ -360,33 +397,46 @@
       *
       * \sa MatrixBase::inverse()
       */
-    inline const internal::solve_retval<FullPivLU,typename MatrixType::IdentityReturnType> inverse() const
+    inline const Inverse<FullPivLU> inverse() const
     {
       eigen_assert(m_isInitialized && "LU is not initialized.");
       eigen_assert(m_lu.rows() == m_lu.cols() && "You can't take the inverse of a non-square matrix!");
-      return internal::solve_retval<FullPivLU,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_lu.rows(), m_lu.cols()));
+      return Inverse<FullPivLU>(*this);
     }
 
     MatrixType reconstructedMatrix() const;
 
-    inline Index rows() const { return m_lu.rows(); }
-    inline Index cols() const { return m_lu.cols(); }
+    EIGEN_DEVICE_FUNC inline Index rows() const { return m_lu.rows(); }
+    EIGEN_DEVICE_FUNC inline Index cols() const { return m_lu.cols(); }
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+
+    template<bool Conjugate, typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
+    void computeInPlace();
+
     MatrixType m_lu;
     PermutationPType m_p;
     PermutationQType m_q;
     IntColVectorType m_rowsTranspositions;
     IntRowVectorType m_colsTranspositions;
-    Index m_det_pq, m_nonzero_pivots;
+    Index m_nonzero_pivots;
+    RealScalar m_l1_norm;
     RealScalar m_maxpivot, m_prescribedThreshold;
+    signed char m_det_pq;
     bool m_isInitialized, m_usePrescribedThreshold;
 };
 
@@ -409,7 +459,8 @@
 }
 
 template<typename MatrixType>
-FullPivLU<MatrixType>::FullPivLU(const MatrixType& matrix)
+template<typename InputType>
+FullPivLU<MatrixType>::FullPivLU(const EigenBase<InputType>& matrix)
   : m_lu(matrix.rows(), matrix.cols()),
     m_p(matrix.rows()),
     m_q(matrix.cols()),
@@ -418,28 +469,41 @@
     m_isInitialized(false),
     m_usePrescribedThreshold(false)
 {
-  compute(matrix);
+  compute(matrix.derived());
 }
 
 template<typename MatrixType>
-FullPivLU<MatrixType>& FullPivLU<MatrixType>::compute(const MatrixType& matrix)
+template<typename InputType>
+FullPivLU<MatrixType>::FullPivLU(EigenBase<InputType>& matrix)
+  : m_lu(matrix.derived()),
+    m_p(matrix.rows()),
+    m_q(matrix.cols()),
+    m_rowsTranspositions(matrix.rows()),
+    m_colsTranspositions(matrix.cols()),
+    m_isInitialized(false),
+    m_usePrescribedThreshold(false)
+{
+  computeInPlace();
+}
+
+template<typename MatrixType>
+void FullPivLU<MatrixType>::computeInPlace()
 {
   check_template_parameters();
-  
-  // the permutations are stored as int indices, so just to be sure:
-  eigen_assert(matrix.rows()<=NumTraits<int>::highest() && matrix.cols()<=NumTraits<int>::highest());
-  
-  m_isInitialized = true;
-  m_lu = matrix;
 
-  const Index size = matrix.diagonalSize();
-  const Index rows = matrix.rows();
-  const Index cols = matrix.cols();
+  // the permutations are stored as int indices, so just to be sure:
+  eigen_assert(m_lu.rows()<=NumTraits<int>::highest() && m_lu.cols()<=NumTraits<int>::highest());
+
+  m_l1_norm = m_lu.cwiseAbs().colwise().sum().maxCoeff();
+
+  const Index size = m_lu.diagonalSize();
+  const Index rows = m_lu.rows();
+  const Index cols = m_lu.cols();
 
   // will store the transpositions, before we accumulate them at the end.
   // can't accumulate on-the-fly because that will be done in reverse order for the rows.
-  m_rowsTranspositions.resize(matrix.rows());
-  m_colsTranspositions.resize(matrix.cols());
+  m_rowsTranspositions.resize(m_lu.rows());
+  m_colsTranspositions.resize(m_lu.cols());
   Index number_of_transpositions = 0; // number of NONTRIVIAL transpositions, i.e. m_rowsTranspositions[i]!=i
 
   m_nonzero_pivots = size; // the generic case is that in which all pivots are nonzero (invertible case)
@@ -451,14 +515,16 @@
 
     // biggest coefficient in the remaining bottom-right corner (starting at row k, col k)
     Index row_of_biggest_in_corner, col_of_biggest_in_corner;
-    RealScalar biggest_in_corner;
+    typedef internal::scalar_score_coeff_op<Scalar> Scoring;
+    typedef typename Scoring::result_type Score;
+    Score biggest_in_corner;
     biggest_in_corner = m_lu.bottomRightCorner(rows-k, cols-k)
-                        .cwiseAbs()
+                        .unaryExpr(Scoring())
                         .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
     row_of_biggest_in_corner += k; // correct the values! since they were computed in the corner,
     col_of_biggest_in_corner += k; // need to add k to them.
 
-    if(biggest_in_corner==RealScalar(0))
+    if(biggest_in_corner==Score(0))
     {
       // before exiting, make sure to initialize the still uninitialized transpositions
       // in a sane state without destroying what we already have.
@@ -471,7 +537,8 @@
       break;
     }
 
-    if(biggest_in_corner > m_maxpivot) m_maxpivot = biggest_in_corner;
+    RealScalar abs_pivot = internal::abs_knowing_score<Scalar>()(m_lu(row_of_biggest_in_corner, col_of_biggest_in_corner), biggest_in_corner);
+    if(abs_pivot > m_maxpivot) m_maxpivot = abs_pivot;
 
     // Now that we've found the pivot, we need to apply the row/col swaps to
     // bring it to the location (k,k).
@@ -508,7 +575,8 @@
     m_q.applyTranspositionOnTheRight(k, m_colsTranspositions.coeff(k));
 
   m_det_pq = (number_of_transpositions%2) ? -1 : 1;
-  return *this;
+
+  m_isInitialized = true;
 }
 
 template<typename MatrixType>
@@ -671,64 +739,136 @@
 
 /***** Implementation of solve() *****************************************************/
 
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<FullPivLU<_MatrixType>, Rhs>
-  : solve_retval_base<FullPivLU<_MatrixType>, Rhs>
+} // end namespace internal
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void FullPivLU<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  EIGEN_MAKE_SOLVE_HELPERS(FullPivLU<_MatrixType>,Rhs)
+  /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1}.
+  * So we proceed as follows:
+  * Step 1: compute c = P * rhs.
+  * Step 2: replace c by the solution x to Lx = c. Exists because L is invertible.
+  * Step 3: replace c by the solution x to Ux = c. May or may not exist.
+  * Step 4: result = Q * c;
+  */
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  const Index rows = this->rows(),
+              cols = this->cols(),
+              nonzero_pivots = this->rank();
+  eigen_assert(rhs.rows() == rows);
+  const Index smalldim = (std::min)(rows, cols);
+
+  if(nonzero_pivots == 0)
   {
-    /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1}.
-     * So we proceed as follows:
-     * Step 1: compute c = P * rhs.
-     * Step 2: replace c by the solution x to Lx = c. Exists because L is invertible.
-     * Step 3: replace c by the solution x to Ux = c. May or may not exist.
-     * Step 4: result = Q * c;
-     */
+    dst.setZero();
+    return;
+  }
 
-    const Index rows = dec().rows(), cols = dec().cols(),
-              nonzero_pivots = dec().nonzeroPivots();
-    eigen_assert(rhs().rows() == rows);
-    const Index smalldim = (std::min)(rows, cols);
+  typename RhsType::PlainObject c(rhs.rows(), rhs.cols());
 
-    if(nonzero_pivots == 0)
-    {
-      dst.setZero();
-      return;
-    }
+  // Step 1
+  c = permutationP() * rhs;
 
-    typename Rhs::PlainObject c(rhs().rows(), rhs().cols());
+  // Step 2
+  m_lu.topLeftCorner(smalldim,smalldim)
+      .template triangularView<UnitLower>()
+      .solveInPlace(c.topRows(smalldim));
+  if(rows>cols)
+    c.bottomRows(rows-cols) -= m_lu.bottomRows(rows-cols) * c.topRows(cols);
 
-    // Step 1
-    c = dec().permutationP() * rhs();
+  // Step 3
+  m_lu.topLeftCorner(nonzero_pivots, nonzero_pivots)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(nonzero_pivots));
 
+  // Step 4
+  for(Index i = 0; i < nonzero_pivots; ++i)
+    dst.row(permutationQ().indices().coeff(i)) = c.row(i);
+  for(Index i = nonzero_pivots; i < m_lu.cols(); ++i)
+    dst.row(permutationQ().indices().coeff(i)).setZero();
+}
+
+template<typename _MatrixType>
+template<bool Conjugate, typename RhsType, typename DstType>
+void FullPivLU<_MatrixType>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const
+{
+  /* The decomposition PAQ = LU can be rewritten as A = P^{-1} L U Q^{-1},
+   * and since permutations are real and unitary, we can write this
+   * as   A^T = Q U^T L^T P,
+   * So we proceed as follows:
+   * Step 1: compute c = Q^T rhs.
+   * Step 2: replace c by the solution x to U^T x = c. May or may not exist.
+   * Step 3: replace c by the solution x to L^T x = c.
+   * Step 4: result = P^T c.
+   * If Conjugate is true, replace "^T" by "^*" above.
+   */
+
+  const Index rows = this->rows(), cols = this->cols(),
+    nonzero_pivots = this->rank();
+   eigen_assert(rhs.rows() == cols);
+  const Index smalldim = (std::min)(rows, cols);
+
+  if(nonzero_pivots == 0)
+  {
+    dst.setZero();
+    return;
+  }
+
+  typename RhsType::PlainObject c(rhs.rows(), rhs.cols());
+
+  // Step 1
+  c = permutationQ().inverse() * rhs;
+
+  if (Conjugate) {
     // Step 2
-    dec().matrixLU()
-        .topLeftCorner(smalldim,smalldim)
-        .template triangularView<UnitLower>()
-        .solveInPlace(c.topRows(smalldim));
-    if(rows>cols)
-    {
-      c.bottomRows(rows-cols)
-        -= dec().matrixLU().bottomRows(rows-cols)
-         * c.topRows(cols);
-    }
-
-    // Step 3
-    dec().matrixLU()
-        .topLeftCorner(nonzero_pivots, nonzero_pivots)
+    m_lu.topLeftCorner(nonzero_pivots, nonzero_pivots)
         .template triangularView<Upper>()
+        .adjoint()
         .solveInPlace(c.topRows(nonzero_pivots));
+    // Step 3
+    m_lu.topLeftCorner(smalldim, smalldim)
+        .template triangularView<UnitLower>()
+        .adjoint()
+        .solveInPlace(c.topRows(smalldim));
+  } else {
+    // Step 2
+    m_lu.topLeftCorner(nonzero_pivots, nonzero_pivots)
+        .template triangularView<Upper>()
+        .transpose()
+        .solveInPlace(c.topRows(nonzero_pivots));
+    // Step 3
+    m_lu.topLeftCorner(smalldim, smalldim)
+        .template triangularView<UnitLower>()
+        .transpose()
+        .solveInPlace(c.topRows(smalldim));
+  }
 
-    // Step 4
-    for(Index i = 0; i < nonzero_pivots; ++i)
-      dst.row(dec().permutationQ().indices().coeff(i)) = c.row(i);
-    for(Index i = nonzero_pivots; i < dec().matrixLU().cols(); ++i)
-      dst.row(dec().permutationQ().indices().coeff(i)).setZero();
+  // Step 4
+  PermutationPType invp = permutationP().inverse().eval();
+  for(Index i = 0; i < smalldim; ++i)
+    dst.row(invp.indices().coeff(i)) = c.row(i);
+  for(Index i = smalldim; i < rows; ++i)
+    dst.row(invp.indices().coeff(i)).setZero();
+}
+
+#endif
+
+namespace internal {
+
+
+/***** Implementation of inverse() *****************************************************/
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<FullPivLU<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename FullPivLU<MatrixType>::Scalar>, Dense2Dense>
+{
+  typedef FullPivLU<MatrixType> LuType;
+  typedef Inverse<LuType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename MatrixType::Scalar> &)
+  {
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
-
 } // end namespace internal
 
 /******* MatrixBase methods *****************************************************************/
diff --git a/Eigen/src/LU/Inverse.h b/Eigen/src/LU/InverseImpl.h
similarity index 86%
rename from Eigen/src/LU/Inverse.h
rename to Eigen/src/LU/InverseImpl.h
index 3cf8871..f49f233 100644
--- a/Eigen/src/LU/Inverse.h
+++ b/Eigen/src/LU/InverseImpl.h
@@ -2,13 +2,14 @@
 // for linear algebra.
 //
 // Copyright (C) 2008-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-#ifndef EIGEN_INVERSE_H
-#define EIGEN_INVERSE_H
+#ifndef EIGEN_INVERSE_IMPL_H
+#define EIGEN_INVERSE_IMPL_H
 
 namespace Eigen { 
 
@@ -21,6 +22,7 @@
 template<typename MatrixType, typename ResultType, int Size = MatrixType::RowsAtCompileTime>
 struct compute_inverse
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(const MatrixType& matrix, ResultType& result)
   {
     result = matrix.partialPivLu().inverse();
@@ -37,16 +39,19 @@
 template<typename MatrixType, typename ResultType>
 struct compute_inverse<MatrixType, ResultType, 1>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(const MatrixType& matrix, ResultType& result)
   {
     typedef typename MatrixType::Scalar Scalar;
-    result.coeffRef(0,0) = Scalar(1) / matrix.coeff(0,0);
+    internal::evaluator<MatrixType> matrixEval(matrix);
+    result.coeffRef(0,0) = Scalar(1) / matrixEval.coeff(0,0);
   }
 };
 
 template<typename MatrixType, typename ResultType>
 struct compute_inverse_and_det_with_check<MatrixType, ResultType, 1>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(
     const MatrixType& matrix,
     const typename MatrixType::RealScalar& absDeterminantThreshold,
@@ -67,19 +72,21 @@
 ****************************/
 
 template<typename MatrixType, typename ResultType>
+EIGEN_DEVICE_FUNC 
 inline void compute_inverse_size2_helper(
     const MatrixType& matrix, const typename ResultType::Scalar& invdet,
     ResultType& result)
 {
-  result.coeffRef(0,0) = matrix.coeff(1,1) * invdet;
+  result.coeffRef(0,0) =  matrix.coeff(1,1) * invdet;
   result.coeffRef(1,0) = -matrix.coeff(1,0) * invdet;
   result.coeffRef(0,1) = -matrix.coeff(0,1) * invdet;
-  result.coeffRef(1,1) = matrix.coeff(0,0) * invdet;
+  result.coeffRef(1,1) =  matrix.coeff(0,0) * invdet;
 }
 
 template<typename MatrixType, typename ResultType>
 struct compute_inverse<MatrixType, ResultType, 2>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(const MatrixType& matrix, ResultType& result)
   {
     typedef typename ResultType::Scalar Scalar;
@@ -91,6 +98,7 @@
 template<typename MatrixType, typename ResultType>
 struct compute_inverse_and_det_with_check<MatrixType, ResultType, 2>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(
     const MatrixType& matrix,
     const typename MatrixType::RealScalar& absDeterminantThreshold,
@@ -114,6 +122,7 @@
 ****************************/
 
 template<typename MatrixType, int i, int j>
+EIGEN_DEVICE_FUNC 
 inline typename MatrixType::Scalar cofactor_3x3(const MatrixType& m)
 {
   enum {
@@ -127,6 +136,7 @@
 }
 
 template<typename MatrixType, typename ResultType>
+EIGEN_DEVICE_FUNC
 inline void compute_inverse_size3_helper(
     const MatrixType& matrix,
     const typename ResultType::Scalar& invdet,
@@ -145,6 +155,7 @@
 template<typename MatrixType, typename ResultType>
 struct compute_inverse<MatrixType, ResultType, 3>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(const MatrixType& matrix, ResultType& result)
   {
     typedef typename ResultType::Scalar Scalar;
@@ -161,6 +172,7 @@
 template<typename MatrixType, typename ResultType>
 struct compute_inverse_and_det_with_check<MatrixType, ResultType, 3>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(
     const MatrixType& matrix,
     const typename MatrixType::RealScalar& absDeterminantThreshold,
@@ -188,6 +200,7 @@
 ****************************/
 
 template<typename Derived>
+EIGEN_DEVICE_FUNC 
 inline const typename Derived::Scalar general_det3_helper
 (const MatrixBase<Derived>& matrix, int i1, int i2, int i3, int j1, int j2, int j3)
 {
@@ -196,6 +209,7 @@
 }
 
 template<typename MatrixType, int i, int j>
+EIGEN_DEVICE_FUNC 
 inline typename MatrixType::Scalar cofactor_4x4(const MatrixType& matrix)
 {
   enum {
@@ -214,6 +228,7 @@
 template<int Arch, typename Scalar, typename MatrixType, typename ResultType>
 struct compute_inverse_size4
 {
+  EIGEN_DEVICE_FUNC
   static void run(const MatrixType& matrix, ResultType& result)
   {
     result.coeffRef(0,0) =  cofactor_4x4<MatrixType,0,0>(matrix);
@@ -246,6 +261,7 @@
 template<typename MatrixType, typename ResultType>
 struct compute_inverse_and_det_with_check<MatrixType, ResultType, 4>
 {
+  EIGEN_DEVICE_FUNC
   static inline void run(
     const MatrixType& matrix,
     const typename MatrixType::RealScalar& absDeterminantThreshold,
@@ -265,38 +281,37 @@
 *** MatrixBase methods ***
 *************************/
 
-template<typename MatrixType>
-struct traits<inverse_impl<MatrixType> >
+} // end namespace internal
+
+namespace internal {
+
+// Specialization for "dense = dense_xpr.inverse()"
+template<typename DstXprType, typename XprType>
+struct Assignment<DstXprType, Inverse<XprType>, internal::assign_op<typename DstXprType::Scalar,typename XprType::Scalar>, Dense2Dense>
 {
-  typedef typename MatrixType::PlainObject ReturnType;
-};
-
-template<typename MatrixType>
-struct inverse_impl : public ReturnByValue<inverse_impl<MatrixType> >
-{
-  typedef typename MatrixType::Index Index;
-  typedef typename internal::eval<MatrixType>::type MatrixTypeNested;
-  typedef typename remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
-  MatrixTypeNested m_matrix;
-
-  inverse_impl(const MatrixType& matrix)
-    : m_matrix(matrix)
-  {}
-
-  inline Index rows() const { return m_matrix.rows(); }
-  inline Index cols() const { return m_matrix.cols(); }
-
-  template<typename Dest> inline void evalTo(Dest& dst) const
+  typedef Inverse<XprType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename XprType::Scalar> &)
   {
-    const int Size = EIGEN_PLAIN_ENUM_MIN(MatrixType::ColsAtCompileTime,Dest::ColsAtCompileTime);
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    const int Size = EIGEN_PLAIN_ENUM_MIN(XprType::ColsAtCompileTime,DstXprType::ColsAtCompileTime);
     EIGEN_ONLY_USED_FOR_DEBUG(Size);
-    eigen_assert(( (Size<=1) || (Size>4) || (extract_data(m_matrix)!=extract_data(dst)))
+    eigen_assert(( (Size<=1) || (Size>4) || (extract_data(src.nestedExpression())!=extract_data(dst)))
               && "Aliasing problem detected in inverse(), you need to do inverse().eval() here.");
 
-    compute_inverse<MatrixTypeNestedCleaned, Dest>::run(m_matrix, dst);
+    typedef typename internal::nested_eval<XprType,XprType::ColsAtCompileTime>::type  ActualXprType;
+    typedef typename internal::remove_all<ActualXprType>::type                        ActualXprTypeCleanded;
+    
+    ActualXprType actual_xpr(src.nestedExpression());
+    
+    compute_inverse<ActualXprTypeCleanded, DstXprType>::run(actual_xpr, dst);
   }
 };
 
+  
 } // end namespace internal
 
 /** \lu_module
@@ -317,11 +332,11 @@
   * \sa computeInverseAndDetWithCheck()
   */
 template<typename Derived>
-inline const internal::inverse_impl<Derived> MatrixBase<Derived>::inverse() const
+inline const Inverse<Derived> MatrixBase<Derived>::inverse() const
 {
   EIGEN_STATIC_ASSERT(!NumTraits<Scalar>::IsInteger,THIS_FUNCTION_IS_NOT_FOR_INTEGER_NUMERIC_TYPES)
   eigen_assert(rows() == cols());
-  return internal::inverse_impl<Derived>(derived());
+  return Inverse<Derived>(derived());
 }
 
 /** \lu_module
@@ -357,7 +372,7 @@
   // for larger sizes, evaluating has negligible cost and limits code size.
   typedef typename internal::conditional<
     RowsAtCompileTime == 2,
-    typename internal::remove_all<typename internal::nested<Derived, 2>::type>::type,
+    typename internal::remove_all<typename internal::nested_eval<Derived, 2>::type>::type,
     PlainObject
   >::type MatrixType;
   internal::compute_inverse_and_det_with_check<MatrixType, ResultType>::run
@@ -389,7 +404,7 @@
     const RealScalar& absDeterminantThreshold
   ) const
 {
-  RealScalar determinant;
+  Scalar determinant;
   // i'd love to put some static assertions there, but SFINAE means that they have no effect...
   eigen_assert(rows() == cols());
   computeInverseAndDetWithCheck(inverse,determinant,invertible,absDeterminantThreshold);
@@ -397,4 +412,4 @@
 
 } // end namespace Eigen
 
-#endif // EIGEN_INVERSE_H
+#endif // EIGEN_INVERSE_IMPL_H
diff --git a/Eigen/src/LU/PartialPivLU.h b/Eigen/src/LU/PartialPivLU.h
index 7d1db94..d439618 100644
--- a/Eigen/src/LU/PartialPivLU.h
+++ b/Eigen/src/LU/PartialPivLU.h
@@ -11,7 +11,33 @@
 #ifndef EIGEN_PARTIALLU_H
 #define EIGEN_PARTIALLU_H
 
-namespace Eigen { 
+namespace Eigen {
+
+namespace internal {
+template<typename _MatrixType> struct traits<PartialPivLU<_MatrixType> >
+ : traits<_MatrixType>
+{
+  typedef MatrixXpr XprKind;
+  typedef SolverStorage StorageKind;
+  typedef traits<_MatrixType> BaseTraits;
+  enum {
+    Flags = BaseTraits::Flags & RowMajorBit,
+    CoeffReadCost = Dynamic
+  };
+};
+
+template<typename T,typename Derived>
+struct enable_if_ref;
+// {
+//   typedef Derived type;
+// };
+
+template<typename T,typename Derived>
+struct enable_if_ref<Ref<T>,Derived> {
+  typedef Derived type;
+};
+
+} // end namespace internal
 
 /** \ingroup LU_Module
   *
@@ -19,7 +45,7 @@
   *
   * \brief LU decomposition of a matrix with partial pivoting, and related features
   *
-  * \param MatrixType the type of the matrix of which we are computing the LU decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the LU decomposition
   *
   * This class represents a LU decomposition of a \b square \b invertible matrix, with partial pivoting: the matrix A
   * is decomposed as A = PLU where L is unit-lower-triangular, U is upper-triangular, and P
@@ -42,34 +68,33 @@
   *
   * The data of the LU decomposition can be directly accessed through the methods matrixLU(), permutationP().
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::partialPivLu(), MatrixBase::determinant(), MatrixBase::inverse(), MatrixBase::computeInverse(), class FullPivLU
   */
 template<typename _MatrixType> class PartialPivLU
+  : public SolverBase<PartialPivLU<_MatrixType> >
 {
   public:
 
     typedef _MatrixType MatrixType;
+    typedef SolverBase<PartialPivLU> Base;
+    EIGEN_GENERIC_PUBLIC_INTERFACE(PartialPivLU)
+    // FIXME StorageIndex defined in EIGEN_GENERIC_PUBLIC_INTERFACE should be int
     enum {
-      RowsAtCompileTime = MatrixType::RowsAtCompileTime,
-      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
-    typedef typename MatrixType::Scalar Scalar;
-    typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename internal::traits<MatrixType>::StorageKind StorageKind;
-    typedef typename MatrixType::Index Index;
     typedef PermutationMatrix<RowsAtCompileTime, MaxRowsAtCompileTime> PermutationType;
     typedef Transpositions<RowsAtCompileTime, MaxRowsAtCompileTime> TranspositionType;
-
+    typedef typename MatrixType::PlainObject PlainObject;
 
     /**
-    * \brief Default Constructor.
-    *
-    * The default constructor is useful in cases in which the user intends to
-    * perform decompositions via PartialPivLU::compute(const MatrixType&).
-    */
+      * \brief Default Constructor.
+      *
+      * The default constructor is useful in cases in which the user intends to
+      * perform decompositions via PartialPivLU::compute(const MatrixType&).
+      */
     PartialPivLU();
 
     /** \brief Default Constructor with memory preallocation
@@ -78,7 +103,7 @@
       * according to the specified problem \a size.
       * \sa PartialPivLU()
       */
-    PartialPivLU(Index size);
+    explicit PartialPivLU(Index size);
 
     /** Constructor.
       *
@@ -87,9 +112,25 @@
       * \warning The matrix should have full rank (e.g. if it's square, it should be invertible).
       * If you need to deal with non-full rank, use class FullPivLU instead.
       */
-    PartialPivLU(const MatrixType& matrix);
+    template<typename InputType>
+    explicit PartialPivLU(const EigenBase<InputType>& matrix);
 
-    PartialPivLU& compute(const MatrixType& matrix);
+    /** Constructor for \link InplaceDecomposition inplace decomposition \endlink
+      *
+      * \param matrix the matrix of which to compute the LU decomposition.
+      *
+      * \warning The matrix should have full rank (e.g. if it's square, it should be invertible).
+      * If you need to deal with non-full rank, use class FullPivLU instead.
+      */
+    template<typename InputType>
+    explicit PartialPivLU(EigenBase<InputType>& matrix);
+
+    template<typename InputType>
+    PartialPivLU& compute(const EigenBase<InputType>& matrix) {
+      m_lu = matrix.derived();
+      compute();
+      return *this;
+    }
 
     /** \returns the LU decomposition matrix: the upper-triangular part is U, the
       * unit-lower-triangular part is L (at least for square matrices; in the non-square
@@ -128,12 +169,22 @@
       *
       * \sa TriangularView::solve(), inverse(), computeInverse()
       */
+    // FIXME this is a copy-paste of the base-class member to add the isInitialized assertion.
     template<typename Rhs>
-    inline const internal::solve_retval<PartialPivLU, Rhs>
+    inline const Solve<PartialPivLU, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
-      return internal::solve_retval<PartialPivLU, Rhs>(*this, b.derived());
+      return Solve<PartialPivLU, Rhs>(*this, b.derived());
+    }
+
+    /** \returns an estimate of the reciprocal condition number of the matrix of which \c *this is
+        the LU decomposition.
+      */
+    inline RealScalar rcond() const
+    {
+      eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
+      return internal::rcond_estimate_helper(m_l1_norm, *this);
     }
 
     /** \returns the inverse of the matrix of which *this is the LU decomposition.
@@ -143,11 +194,10 @@
       *
       * \sa MatrixBase::inverse(), LU::inverse()
       */
-    inline const internal::solve_retval<PartialPivLU,typename MatrixType::IdentityReturnType> inverse() const
+    inline const Inverse<PartialPivLU> inverse() const
     {
       eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
-      return internal::solve_retval<PartialPivLU,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_lu.rows(), m_lu.cols()));
+      return Inverse<PartialPivLU>(*this);
     }
 
     /** \returns the determinant of the matrix of which
@@ -163,24 +213,78 @@
       *
       * \sa MatrixBase::determinant()
       */
-    typename internal::traits<MatrixType>::Scalar determinant() const;
+    Scalar determinant() const;
 
     MatrixType reconstructedMatrix() const;
 
     inline Index rows() const { return m_lu.rows(); }
     inline Index cols() const { return m_lu.cols(); }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const {
+     /* The decomposition PA = LU can be rewritten as A = P^{-1} L U.
+      * So we proceed as follows:
+      * Step 1: compute c = Pb.
+      * Step 2: replace c by the solution x to Lx = c.
+      * Step 3: replace c by the solution x to Ux = c.
+      */
+
+      eigen_assert(rhs.rows() == m_lu.rows());
+
+      // Step 1
+      dst = permutationP() * rhs;
+
+      // Step 2
+      m_lu.template triangularView<UnitLower>().solveInPlace(dst);
+
+      // Step 3
+      m_lu.template triangularView<Upper>().solveInPlace(dst);
+    }
+
+    template<bool Conjugate, typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const {
+     /* The decomposition PA = LU can be rewritten as A = P^{-1} L U.
+      * So we proceed as follows:
+      * Step 1: compute c = Pb.
+      * Step 2: replace c by the solution x to Lx = c.
+      * Step 3: replace c by the solution x to Ux = c.
+      */
+
+      eigen_assert(rhs.rows() == m_lu.cols());
+
+      if (Conjugate) {
+        // Step 1
+        dst = m_lu.template triangularView<Upper>().adjoint().solve(rhs);
+        // Step 2
+        m_lu.template triangularView<UnitLower>().adjoint().solveInPlace(dst);
+      } else {
+        // Step 1
+        dst = m_lu.template triangularView<Upper>().transpose().solve(rhs);
+        // Step 2
+        m_lu.template triangularView<UnitLower>().transpose().solveInPlace(dst);
+      }
+      // Step 3
+      dst = permutationP().transpose() * dst;
+    }
+    #endif
+
   protected:
-    
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
+    void compute();
+
     MatrixType m_lu;
     PermutationType m_p;
     TranspositionType m_rowsTranspositions;
-    Index m_det_p;
+    RealScalar m_l1_norm;
+    signed char m_det_p;
     bool m_isInitialized;
 };
 
@@ -189,6 +293,7 @@
   : m_lu(),
     m_p(),
     m_rowsTranspositions(),
+    m_l1_norm(0),
     m_det_p(0),
     m_isInitialized(false)
 {
@@ -199,20 +304,36 @@
   : m_lu(size, size),
     m_p(size),
     m_rowsTranspositions(size),
+    m_l1_norm(0),
     m_det_p(0),
     m_isInitialized(false)
 {
 }
 
 template<typename MatrixType>
-PartialPivLU<MatrixType>::PartialPivLU(const MatrixType& matrix)
-  : m_lu(matrix.rows(), matrix.rows()),
+template<typename InputType>
+PartialPivLU<MatrixType>::PartialPivLU(const EigenBase<InputType>& matrix)
+  : m_lu(matrix.rows(),matrix.cols()),
     m_p(matrix.rows()),
     m_rowsTranspositions(matrix.rows()),
+    m_l1_norm(0),
     m_det_p(0),
     m_isInitialized(false)
 {
-  compute(matrix);
+  compute(matrix.derived());
+}
+
+template<typename MatrixType>
+template<typename InputType>
+PartialPivLU<MatrixType>::PartialPivLU(EigenBase<InputType>& matrix)
+  : m_lu(matrix.derived()),
+    m_p(matrix.rows()),
+    m_rowsTranspositions(matrix.rows()),
+    m_l1_norm(0),
+    m_det_p(0),
+    m_isInitialized(false)
+{
+  compute();
 }
 
 namespace internal {
@@ -230,7 +351,6 @@
   typedef Block<MapLU, Dynamic, Dynamic> MatrixType;
   typedef Block<MatrixType,Dynamic,Dynamic> BlockType;
   typedef typename MatrixType::RealScalar RealScalar;
-  typedef typename MatrixType::Index Index;
 
   /** \internal performs the LU decomposition in-place of the matrix \a lu
     * using an unblocked algorithm.
@@ -244,6 +364,8 @@
     */
   static Index unblocked_lu(MatrixType& lu, PivIndex* row_transpositions, PivIndex& nb_transpositions)
   {
+    typedef scalar_score_coeff_op<Scalar> Scoring;
+    typedef typename Scoring::result_type Score;
     const Index rows = lu.rows();
     const Index cols = lu.cols();
     const Index size = (std::min)(rows,cols);
@@ -253,15 +375,15 @@
     {
       Index rrows = rows-k-1;
       Index rcols = cols-k-1;
-        
+
       Index row_of_biggest_in_col;
-      RealScalar biggest_in_corner
-        = lu.col(k).tail(rows-k).cwiseAbs().maxCoeff(&row_of_biggest_in_col);
+      Score biggest_in_corner
+        = lu.col(k).tail(rows-k).unaryExpr(Scoring()).maxCoeff(&row_of_biggest_in_col);
       row_of_biggest_in_col += k;
 
       row_transpositions[k] = PivIndex(row_of_biggest_in_col);
 
-      if(biggest_in_corner != RealScalar(0))
+      if(biggest_in_corner != Score(0))
       {
         if(k != row_of_biggest_in_col)
         {
@@ -354,7 +476,7 @@
       // update permutations and apply them to A_0
       for(Index i=k; i<k+bs; ++i)
       {
-        Index piv = (row_transpositions[i] += k);
+        Index piv = (row_transpositions[i] += internal::convert_index<PivIndex>(k));
         A_0.row(i).swap(A_0.row(piv));
       }
 
@@ -377,45 +499,44 @@
 /** \internal performs the LU decomposition with partial pivoting in-place.
   */
 template<typename MatrixType, typename TranspositionType>
-void partial_lu_inplace(MatrixType& lu, TranspositionType& row_transpositions, typename TranspositionType::Index& nb_transpositions)
+void partial_lu_inplace(MatrixType& lu, TranspositionType& row_transpositions, typename TranspositionType::StorageIndex& nb_transpositions)
 {
   eigen_assert(lu.cols() == row_transpositions.size());
   eigen_assert((&row_transpositions.coeffRef(1)-&row_transpositions.coeffRef(0)) == 1);
 
   partial_lu_impl
-    <typename MatrixType::Scalar, MatrixType::Flags&RowMajorBit?RowMajor:ColMajor, typename TranspositionType::Index>
+    <typename MatrixType::Scalar, MatrixType::Flags&RowMajorBit?RowMajor:ColMajor, typename TranspositionType::StorageIndex>
     ::blocked_lu(lu.rows(), lu.cols(), &lu.coeffRef(0,0), lu.outerStride(), &row_transpositions.coeffRef(0), nb_transpositions);
 }
 
 } // end namespace internal
 
 template<typename MatrixType>
-PartialPivLU<MatrixType>& PartialPivLU<MatrixType>::compute(const MatrixType& matrix)
+void PartialPivLU<MatrixType>::compute()
 {
   check_template_parameters();
-  
-  // the row permutation is stored as int indices, so just to be sure:
-  eigen_assert(matrix.rows()<NumTraits<int>::highest());
-  
-  m_lu = matrix;
 
-  eigen_assert(matrix.rows() == matrix.cols() && "PartialPivLU is only for square (and moreover invertible) matrices");
-  const Index size = matrix.rows();
+  // the row permutation is stored as int indices, so just to be sure:
+  eigen_assert(m_lu.rows()<NumTraits<int>::highest());
+
+  m_l1_norm = m_lu.cwiseAbs().colwise().sum().maxCoeff();
+
+  eigen_assert(m_lu.rows() == m_lu.cols() && "PartialPivLU is only for square (and moreover invertible) matrices");
+  const Index size = m_lu.rows();
 
   m_rowsTranspositions.resize(size);
 
-  typename TranspositionType::Index nb_transpositions;
+  typename TranspositionType::StorageIndex nb_transpositions;
   internal::partial_lu_inplace(m_lu, m_rowsTranspositions, nb_transpositions);
   m_det_p = (nb_transpositions%2) ? -1 : 1;
 
   m_p = m_rowsTranspositions;
 
   m_isInitialized = true;
-  return *this;
 }
 
 template<typename MatrixType>
-typename internal::traits<MatrixType>::Scalar PartialPivLU<MatrixType>::determinant() const
+typename PartialPivLU<MatrixType>::Scalar PartialPivLU<MatrixType>::determinant() const
 {
   eigen_assert(m_isInitialized && "PartialPivLU is not initialized.");
   return Scalar(m_det_p) * m_lu.diagonal().prod();
@@ -438,38 +559,21 @@
   return res;
 }
 
-/***** Implementation of solve() *****************************************************/
+/***** Implementation details *****************************************************/
 
 namespace internal {
 
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<PartialPivLU<_MatrixType>, Rhs>
-  : solve_retval_base<PartialPivLU<_MatrixType>, Rhs>
+/***** Implementation of inverse() *****************************************************/
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<PartialPivLU<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename PartialPivLU<MatrixType>::Scalar>, Dense2Dense>
 {
-  EIGEN_MAKE_SOLVE_HELPERS(PartialPivLU<_MatrixType>,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
+  typedef PartialPivLU<MatrixType> LuType;
+  typedef Inverse<LuType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename LuType::Scalar> &)
   {
-    /* The decomposition PA = LU can be rewritten as A = P^{-1} L U.
-    * So we proceed as follows:
-    * Step 1: compute c = Pb.
-    * Step 2: replace c by the solution x to Lx = c.
-    * Step 3: replace c by the solution x to Ux = c.
-    */
-
-    eigen_assert(rhs().rows() == dec().matrixLU().rows());
-
-    // Step 1
-    dst = dec().permutationP() * rhs();
-
-    // Step 2
-    dec().matrixLU().template triangularView<UnitLower>().solveInPlace(dst);
-
-    // Step 3
-    dec().matrixLU().template triangularView<Upper>().solveInPlace(dst);
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
-
 } // end namespace internal
 
 /******** MatrixBase methods *******/
@@ -487,7 +591,6 @@
   return PartialPivLU<PlainObject>(eval());
 }
 
-#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS
 /** \lu_module
   *
   * Synonym of partialPivLu().
@@ -502,7 +605,6 @@
 {
   return PartialPivLU<PlainObject>(eval());
 }
-#endif
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/LU/PartialPivLU_MKL.h b/Eigen/src/LU/PartialPivLU_LAPACKE.h
similarity index 76%
rename from Eigen/src/LU/PartialPivLU_MKL.h
rename to Eigen/src/LU/PartialPivLU_LAPACKE.h
index 9035953..755168a 100644
--- a/Eigen/src/LU/PartialPivLU_MKL.h
+++ b/Eigen/src/LU/PartialPivLU_LAPACKE.h
@@ -25,7 +25,7 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to LAPACKe
  *     LU decomposition with partial pivoting based on LAPACKE_?getrf function.
  ********************************************************************************
 */
@@ -33,20 +33,18 @@
 #ifndef EIGEN_PARTIALLU_LAPACK_H
 #define EIGEN_PARTIALLU_LAPACK_H
 
-#include "Eigen/src/Core/util/MKL_support.h"
-
 namespace Eigen { 
 
 namespace internal {
 
-/** \internal Specialization for the data types supported by MKL */
+/** \internal Specialization for the data types supported by LAPACKe */
 
-#define EIGEN_MKL_LU_PARTPIV(EIGTYPE, MKLTYPE, MKLPREFIX) \
+#define EIGEN_LAPACKE_LU_PARTPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX) \
 template<int StorageOrder> \
 struct partial_lu_impl<EIGTYPE, StorageOrder, lapack_int> \
 { \
   /* \internal performs the LU decomposition in-place of the matrix represented */ \
-  static lapack_int blocked_lu(lapack_int rows, lapack_int cols, EIGTYPE* lu_data, lapack_int luStride, lapack_int* row_transpositions, lapack_int& nb_transpositions, lapack_int maxBlockSize=256) \
+  static lapack_int blocked_lu(Index rows, Index cols, EIGTYPE* lu_data, Index luStride, lapack_int* row_transpositions, lapack_int& nb_transpositions, lapack_int maxBlockSize=256) \
   { \
     EIGEN_UNUSED_VARIABLE(maxBlockSize);\
     lapack_int matrix_order, first_zero_pivot; \
@@ -54,14 +52,14 @@
     EIGTYPE* a; \
 /* Set up parameters for ?getrf */ \
     matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
-    lda = luStride; \
+    lda = convert_index<lapack_int>(luStride); \
     a = lu_data; \
     ipiv = row_transpositions; \
-    m = rows; \
-    n = cols; \
+    m = convert_index<lapack_int>(rows); \
+    n = convert_index<lapack_int>(cols); \
     nb_transpositions = 0; \
 \
-    info = LAPACKE_##MKLPREFIX##getrf( matrix_order, m, n, (MKLTYPE*)a, lda, ipiv ); \
+    info = LAPACKE_##LAPACKE_PREFIX##getrf( matrix_order, m, n, (LAPACKE_TYPE*)a, lda, ipiv ); \
 \
     for(int i=0;i<m;i++) { ipiv[i]--; if (ipiv[i]!=i) nb_transpositions++; } \
 \
@@ -73,10 +71,10 @@
   } \
 };
 
-EIGEN_MKL_LU_PARTPIV(double, double, d)
-EIGEN_MKL_LU_PARTPIV(float, float, s)
-EIGEN_MKL_LU_PARTPIV(dcomplex, MKL_Complex16, z)
-EIGEN_MKL_LU_PARTPIV(scomplex, MKL_Complex8, c)
+EIGEN_LAPACKE_LU_PARTPIV(double, double, d)
+EIGEN_LAPACKE_LU_PARTPIV(float, float, s)
+EIGEN_LAPACKE_LU_PARTPIV(dcomplex, lapack_complex_double, z)
+EIGEN_LAPACKE_LU_PARTPIV(scomplex, lapack_complex_float,  c)
 
 } // end namespace internal
 
diff --git a/Eigen/src/LU/arch/CMakeLists.txt b/Eigen/src/LU/arch/CMakeLists.txt
deleted file mode 100644
index f6b7ed9..0000000
--- a/Eigen/src/LU/arch/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_LU_arch_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_LU_arch_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/LU/arch COMPONENT Devel
-  )
diff --git a/Eigen/src/LU/arch/Inverse_SSE.h b/Eigen/src/LU/arch/Inverse_SSE.h
index 60b7a23..ebb64a6 100644
--- a/Eigen/src/LU/arch/Inverse_SSE.h
+++ b/Eigen/src/LU/arch/Inverse_SSE.h
@@ -35,13 +35,15 @@
 struct compute_inverse_size4<Architecture::SSE, float, MatrixType, ResultType>
 {
   enum {
-    MatrixAlignment     = bool(MatrixType::Flags&AlignedBit),
-    ResultAlignment     = bool(ResultType::Flags&AlignedBit),
+    MatrixAlignment     = traits<MatrixType>::Alignment,
+    ResultAlignment     = traits<ResultType>::Alignment,
     StorageOrdersMatch  = (MatrixType::Flags&RowMajorBit) == (ResultType::Flags&RowMajorBit)
   };
+  typedef typename conditional<(MatrixType::Flags&LinearAccessBit),MatrixType const &,typename MatrixType::PlainObject>::type ActualMatrixType;
   
-  static void run(const MatrixType& matrix, ResultType& result)
+  static void run(const MatrixType& mat, ResultType& result)
   {
+    ActualMatrixType matrix(mat);
     EIGEN_ALIGN16 const unsigned int _Sign_PNNP[4] = { 0x00000000, 0x80000000, 0x80000000, 0x00000000 };
 
     // Load the full matrix into registers
@@ -151,10 +153,12 @@
     iC = _mm_mul_ps(rd,iC);
     iD = _mm_mul_ps(rd,iD);
 
-    result.template writePacket<ResultAlignment>( 0, _mm_shuffle_ps(iA,iB,0x77));
-    result.template writePacket<ResultAlignment>( 4, _mm_shuffle_ps(iA,iB,0x22));
-    result.template writePacket<ResultAlignment>( 8, _mm_shuffle_ps(iC,iD,0x77));
-    result.template writePacket<ResultAlignment>(12, _mm_shuffle_ps(iC,iD,0x22));
+    Index res_stride = result.outerStride();
+    float* res = result.data();
+    pstoret<float, Packet4f, ResultAlignment>(res+0,            _mm_shuffle_ps(iA,iB,0x77));
+    pstoret<float, Packet4f, ResultAlignment>(res+res_stride,   _mm_shuffle_ps(iA,iB,0x22));
+    pstoret<float, Packet4f, ResultAlignment>(res+2*res_stride, _mm_shuffle_ps(iC,iD,0x77));
+    pstoret<float, Packet4f, ResultAlignment>(res+3*res_stride, _mm_shuffle_ps(iC,iD,0x22));
   }
 
 };
@@ -163,18 +167,21 @@
 struct compute_inverse_size4<Architecture::SSE, double, MatrixType, ResultType>
 {
   enum {
-    MatrixAlignment = bool(MatrixType::Flags&AlignedBit),
-    ResultAlignment = bool(ResultType::Flags&AlignedBit),
+    MatrixAlignment     = traits<MatrixType>::Alignment,
+    ResultAlignment     = traits<ResultType>::Alignment,
     StorageOrdersMatch  = (MatrixType::Flags&RowMajorBit) == (ResultType::Flags&RowMajorBit)
   };
-  static void run(const MatrixType& matrix, ResultType& result)
+  typedef typename conditional<(MatrixType::Flags&LinearAccessBit),MatrixType const &,typename MatrixType::PlainObject>::type ActualMatrixType;
+  
+  static void run(const MatrixType& mat, ResultType& result)
   {
+    ActualMatrixType matrix(mat);
     const __m128d _Sign_NP = _mm_castsi128_pd(_mm_set_epi32(0x0,0x0,0x80000000,0x0));
     const __m128d _Sign_PN = _mm_castsi128_pd(_mm_set_epi32(0x80000000,0x0,0x0,0x0));
 
     // The inverse is calculated using "Divide and Conquer" technique. The
     // original matrix is divide into four 2x2 sub-matrices. Since each
-    // register of the matrix holds two element, the smaller matrices are
+    // register of the matrix holds two elements, the smaller matrices are
     // consisted of two registers. Hence we get a better locality of the
     // calculations.
 
@@ -311,14 +318,16 @@
     iC1 = _mm_sub_pd(_mm_mul_pd(B1, dC), iC1);
     iC2 = _mm_sub_pd(_mm_mul_pd(B2, dC), iC2);
 
-    result.template writePacket<ResultAlignment>( 0, _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 3), d1));     // iA# / det
-    result.template writePacket<ResultAlignment>( 4, _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 0), d2));
-    result.template writePacket<ResultAlignment>( 2, _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 3), d1));     // iB# / det
-    result.template writePacket<ResultAlignment>( 6, _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 0), d2));
-    result.template writePacket<ResultAlignment>( 8, _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 3), d1));     // iC# / det
-    result.template writePacket<ResultAlignment>(12, _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 0), d2));
-    result.template writePacket<ResultAlignment>(10, _mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 3), d1));     // iD# / det
-    result.template writePacket<ResultAlignment>(14, _mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 0), d2));
+    Index res_stride = result.outerStride();
+    double* res = result.data();
+    pstoret<double, Packet2d, ResultAlignment>(res+0,             _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+res_stride,    _mm_mul_pd(_mm_shuffle_pd(iA2, iA1, 0), d2));
+    pstoret<double, Packet2d, ResultAlignment>(res+2,             _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+res_stride+2,  _mm_mul_pd(_mm_shuffle_pd(iB2, iB1, 0), d2));
+    pstoret<double, Packet2d, ResultAlignment>(res+2*res_stride,  _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+3*res_stride,  _mm_mul_pd(_mm_shuffle_pd(iC2, iC1, 0), d2));
+    pstoret<double, Packet2d, ResultAlignment>(res+2*res_stride+2,_mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 3), d1));
+    pstoret<double, Packet2d, ResultAlignment>(res+3*res_stride+2,_mm_mul_pd(_mm_shuffle_pd(iD2, iD1, 0), d2));
   }
 };
 
diff --git a/Eigen/src/MetisSupport/CMakeLists.txt b/Eigen/src/MetisSupport/CMakeLists.txt
deleted file mode 100644
index 2bad314..0000000
--- a/Eigen/src/MetisSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_MetisSupport_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_MetisSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/MetisSupport COMPONENT Devel
-  )
diff --git a/Eigen/src/MetisSupport/MetisSupport.h b/Eigen/src/MetisSupport/MetisSupport.h
index f2bbef2..4c15304 100644
--- a/Eigen/src/MetisSupport/MetisSupport.h
+++ b/Eigen/src/MetisSupport/MetisSupport.h
@@ -18,12 +18,12 @@
  * Row (column) i of A is the matperm(i) row (column) of Ap. 
  * WARNING: As computed by METIS, this corresponds to the vector iperm (instead of perm)
  */
-template <typename Index>
+template <typename StorageIndex>
 class MetisOrdering
 {
 public:
-  typedef PermutationMatrix<Dynamic,Dynamic,Index> PermutationType;
-  typedef Matrix<Index,Dynamic,1> IndexVector; 
+  typedef PermutationMatrix<Dynamic,Dynamic,StorageIndex> PermutationType;
+  typedef Matrix<StorageIndex,Dynamic,1> IndexVector; 
   
   template <typename MatrixType>
   void get_symmetrized_graph(const MatrixType& A)
@@ -36,7 +36,7 @@
     Index TotNz = 0; 
     IndexVector visited(m); 
     visited.setConstant(-1); 
-    for (int j = 0; j < m; j++)
+    for (StorageIndex j = 0; j < m; j++)
     {
       // Compute the union structure of of A(j,:) and At(j,:)
       visited(j) = j; // Do not include the diagonal element
@@ -67,8 +67,8 @@
 
     // Now compute the real adjacency list of each column/row 
     visited.setConstant(-1); 
-    Index CurNz = 0; 
-    for (int j = 0; j < m; j++)
+    StorageIndex CurNz = 0; 
+    for (StorageIndex j = 0; j < m; j++)
     {
       m_indexPtr(j) = CurNz; 
       
@@ -76,7 +76,7 @@
       // Add the pattern of row/column j of A to A+At
       for (typename MatrixType::InnerIterator it(A,j); it; ++it)
       {
-        Index idx = it.index(); // Get the row index (for column major) or column index (for row major)
+        StorageIndex idx = it.index(); // Get the row index (for column major) or column index (for row major)
         if (visited(idx) != j ) 
         {
           visited(idx) = j; 
@@ -87,7 +87,7 @@
       //Add the pattern of row/column j of At to A+At
       for (typename MatrixType::InnerIterator it(At, j); it; ++it)
       {
-        Index idx = it.index(); 
+        StorageIndex idx = it.index(); 
         if(visited(idx) != j)
         {
           visited(idx) = j; 
@@ -102,7 +102,7 @@
   template <typename MatrixType>
   void operator() (const MatrixType& A, PermutationType& matperm)
   {
-     Index m = A.cols();
+     StorageIndex m = internal::convert_index<StorageIndex>(A.cols()); // must be StorageIndex, because it is passed by address to METIS
      IndexVector perm(m),iperm(m); 
     // First, symmetrize the matrix graph. 
      get_symmetrized_graph(A); 
diff --git a/Eigen/src/OrderingMethods/Amd.h b/Eigen/src/OrderingMethods/Amd.h
index 1c28bdf..f91ecb2 100644
--- a/Eigen/src/OrderingMethods/Amd.h
+++ b/Eigen/src/OrderingMethods/Amd.h
@@ -8,7 +8,7 @@
 NOTE: this routine has been adapted from the CSparse library:
 
 Copyright (c) 2006, Timothy A. Davis.
-http://www.cise.ufl.edu/research/sparse/CSparse
+http://www.suitesparse.com
 
 CSparse is free software; you can redistribute it and/or
 modify it under the terms of the GNU Lesser General Public
@@ -41,10 +41,10 @@
 template<typename T0, typename T1> inline void amd_mark(const T0* w, const T1& j) { return w[j] = amd_flip(w[j]); }
 
 /* clear w */
-template<typename Index>
-static int cs_wclear (Index mark, Index lemax, Index *w, Index n)
+template<typename StorageIndex>
+static StorageIndex cs_wclear (StorageIndex mark, StorageIndex lemax, StorageIndex *w, StorageIndex n)
 {
-  Index k;
+  StorageIndex k;
   if(mark < 2 || (mark + lemax < 0))
   {
     for(k = 0; k < n; k++)
@@ -56,10 +56,10 @@
 }
 
 /* depth-first search and postorder of a tree rooted at node j */
-template<typename Index>
-Index cs_tdfs(Index j, Index k, Index *head, const Index *next, Index *post, Index *stack)
+template<typename StorageIndex>
+StorageIndex cs_tdfs(StorageIndex j, StorageIndex k, StorageIndex *head, const StorageIndex *next, StorageIndex *post, StorageIndex *stack)
 {
-  int i, p, top = 0;
+  StorageIndex i, p, top = 0;
   if(!head || !next || !post || !stack) return (-1);    /* check inputs */
   stack[0] = j;                 /* place j on the stack */
   while (top >= 0)                /* while (stack is not empty) */
@@ -84,42 +84,45 @@
 /** \internal
   * \ingroup OrderingMethods_Module 
   * Approximate minimum degree ordering algorithm.
-  * \returns the permutation P reducing the fill-in of the input matrix \a C
-  * The input matrix \a C must be a selfadjoint compressed column major SparseMatrix object. Both the upper and lower parts have to be stored, but the diagonal entries are optional.
+  *
+  * \param[in] C the input selfadjoint matrix stored in compressed column major format.
+  * \param[out] perm the permutation P reducing the fill-in of the input matrix \a C
+  *
+  * Note that the input matrix \a C must be complete, that is both the upper and lower parts have to be stored, as well as the diagonal entries.
   * On exit the values of C are destroyed */
-template<typename Scalar, typename Index>
-void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,Index>& C, PermutationMatrix<Dynamic,Dynamic,Index>& perm)
+template<typename Scalar, typename StorageIndex>
+void minimum_degree_ordering(SparseMatrix<Scalar,ColMajor,StorageIndex>& C, PermutationMatrix<Dynamic,Dynamic,StorageIndex>& perm)
 {
   using std::sqrt;
   
-  int d, dk, dext, lemax = 0, e, elenk, eln, i, j, k, k1,
-      k2, k3, jlast, ln, dense, nzmax, mindeg = 0, nvi, nvj, nvk, mark, wnvi,
-      ok, nel = 0, p, p1, p2, p3, p4, pj, pk, pk1, pk2, pn, q, t;
-  unsigned int h;
+  StorageIndex d, dk, dext, lemax = 0, e, elenk, eln, i, j, k, k1,
+                k2, k3, jlast, ln, dense, nzmax, mindeg = 0, nvi, nvj, nvk, mark, wnvi,
+                ok, nel = 0, p, p1, p2, p3, p4, pj, pk, pk1, pk2, pn, q, t, h;
   
-  Index n = C.cols();
-  dense = std::max<Index> (16, Index(10 * sqrt(double(n))));   /* find dense threshold */
-  dense = std::min<Index> (n-2, dense);
+  StorageIndex n = StorageIndex(C.cols());
+  dense = std::max<StorageIndex> (16, StorageIndex(10 * sqrt(double(n))));   /* find dense threshold */
+  dense = (std::min)(n-2, dense);
   
-  Index cnz = C.nonZeros();
+  StorageIndex cnz = StorageIndex(C.nonZeros());
   perm.resize(n+1);
   t = cnz + cnz/5 + 2*n;                 /* add elbow room to C */
   C.resizeNonZeros(t);
   
-  Index* W       = new Index[8*(n+1)]; /* get workspace */
-  Index* len     = W;
-  Index* nv      = W +   (n+1);
-  Index* next    = W + 2*(n+1);
-  Index* head    = W + 3*(n+1);
-  Index* elen    = W + 4*(n+1);
-  Index* degree  = W + 5*(n+1);
-  Index* w       = W + 6*(n+1);
-  Index* hhead   = W + 7*(n+1);
-  Index* last    = perm.indices().data();                              /* use P as workspace for last */
+  // get workspace
+  ei_declare_aligned_stack_constructed_variable(StorageIndex,W,8*(n+1),0);
+  StorageIndex* len     = W;
+  StorageIndex* nv      = W +   (n+1);
+  StorageIndex* next    = W + 2*(n+1);
+  StorageIndex* head    = W + 3*(n+1);
+  StorageIndex* elen    = W + 4*(n+1);
+  StorageIndex* degree  = W + 5*(n+1);
+  StorageIndex* w       = W + 6*(n+1);
+  StorageIndex* hhead   = W + 7*(n+1);
+  StorageIndex* last    = perm.indices().data();                              /* use P as workspace for last */
   
   /* --- Initialize quotient graph ---------------------------------------- */
-  Index* Cp = C.outerIndexPtr();
-  Index* Ci = C.innerIndexPtr();
+  StorageIndex* Cp = C.outerIndexPtr();
+  StorageIndex* Ci = C.innerIndexPtr();
   for(k = 0; k < n; k++)
     len[k] = Cp[k+1] - Cp[k];
   len[n] = 0;
@@ -136,10 +139,7 @@
     elen[i]   = 0;                      // Ek of node i is empty
     degree[i] = len[i];                 // degree of node i
   }
-  mark = internal::cs_wclear<Index>(0, 0, w, n);         /* clear w */
-  elen[n] = -2;                         /* n is a dead element */
-  Cp[n] = -1;                           /* n is a root of assembly tree */
-  w[n] = 0;                             /* n is a dead element */
+  mark = internal::cs_wclear<StorageIndex>(0, 0, w, n);         /* clear w */
   
   /* --- Initialize degree lists ------------------------------------------ */
   for(i = 0; i < n; i++)
@@ -153,7 +153,7 @@
       }
    
     d = degree[i];
-    if(d == 1)                      /* node i is empty */
+    if(d == 1 && has_diag)           /* node i is empty */
     {
       elen[i] = -2;                 /* element i is dead */
       nel++;
@@ -263,7 +263,7 @@
     elen[k] = -2;                     /* k is now an element */
     
     /* --- Find set differences ----------------------------------------- */
-    mark = internal::cs_wclear<Index>(mark, lemax, w, n);  /* clear w if necessary */
+    mark = internal::cs_wclear<StorageIndex>(mark, lemax, w, n);  /* clear w if necessary */
     for(pk = pk1; pk < pk2; pk++)    /* scan 1: find |Le\Lk| */
     {
       i = Ci[pk];
@@ -333,7 +333,7 @@
       }
       else
       {
-        degree[i] = std::min<Index> (degree[i], d);   /* update degree(i) */
+        degree[i] = std::min<StorageIndex> (degree[i], d);   /* update degree(i) */
         Ci[pn] = Ci[p3];         /* move first node to end */
         Ci[p3] = Ci[p1];         /* move 1st el. to end of Ei */
         Ci[p1] = k;               /* add k as 1st element in of Ei */
@@ -341,12 +341,12 @@
         h %= n;                    /* finalize hash of i */
         next[i] = hhead[h];      /* place i in hash bucket */
         hhead[h] = i;
-        last[i] = h;              /* save hash of i in last[i] */
+        last[i] = h;      /* save hash of i in last[i] */
       }
     }                                   /* scan2 is done */
     degree[k] = dk;                   /* finalize |Lk| */
-    lemax = std::max<Index>(lemax, dk);
-    mark = internal::cs_wclear<Index>(mark+lemax, lemax, w, n);    /* clear w */
+    lemax = std::max<StorageIndex>(lemax, dk);
+    mark = internal::cs_wclear<StorageIndex>(mark+lemax, lemax, w, n);    /* clear w */
     
     /* --- Supernode detection ------------------------------------------ */
     for(pk = pk1; pk < pk2; pk++)
@@ -394,12 +394,12 @@
       if((nvi = -nv[i]) <= 0) continue;/* skip if i is dead */
       nv[i] = nvi;                      /* restore nv[i] */
       d = degree[i] + dk - nvi;         /* compute external degree(i) */
-      d = std::min<Index> (d, n - nel - nvi);
+      d = std::min<StorageIndex> (d, n - nel - nvi);
       if(head[d] != -1) last[head[d]] = i;
       next[i] = head[d];               /* put i back in degree list */
       last[i] = -1;
       head[d] = i;
-      mindeg = std::min<Index> (mindeg, d);       /* find new minimum degree */
+      mindeg = std::min<StorageIndex> (mindeg, d);       /* find new minimum degree */
       degree[i] = d;
       Ci[p++] = i;                      /* place i in Lk */
     }
@@ -432,12 +432,10 @@
   }
   for(k = 0, i = 0; i <= n; i++)       /* postorder the assembly tree */
   {
-    if(Cp[i] == -1) k = internal::cs_tdfs<Index>(i, k, head, next, perm.indices().data(), w);
+    if(Cp[i] == -1) k = internal::cs_tdfs<StorageIndex>(i, k, head, next, perm.indices().data(), w);
   }
   
   perm.indices().conservativeResize(n);
-
-  delete[] W;
 }
 
 } // namespace internal
diff --git a/Eigen/src/OrderingMethods/CMakeLists.txt b/Eigen/src/OrderingMethods/CMakeLists.txt
deleted file mode 100644
index 9f4bb27..0000000
--- a/Eigen/src/OrderingMethods/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_OrderingMethods_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_OrderingMethods_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/OrderingMethods COMPONENT Devel
-  )
diff --git a/Eigen/src/OrderingMethods/Eigen_Colamd.h b/Eigen/src/OrderingMethods/Eigen_Colamd.h
index 44548f6..da85b4d 100644
--- a/Eigen/src/OrderingMethods/Eigen_Colamd.h
+++ b/Eigen/src/OrderingMethods/Eigen_Colamd.h
@@ -41,12 +41,8 @@
 // 
 //   The colamd/symamd library is available at
 // 
-//       http://www.cise.ufl.edu/research/sparse/colamd/
+//       http://www.suitesparse.com
 
-//   This is the http://www.cise.ufl.edu/research/sparse/colamd/colamd.h
-//   file.  It is required by the colamd.c, colamdmex.c, and symamdmex.c
-//   files, and by any C code that calls the routines whose prototypes are
-//   listed below, or that uses the colamd/symamd definitions listed below.
   
 #ifndef EIGEN_COLAMD_H
 #define EIGEN_COLAMD_H
@@ -102,9 +98,6 @@
 /* === Definitions ========================================================== */
 /* ========================================================================== */
 
-#define COLAMD_MAX(a,b) (((a) > (b)) ? (a) : (b))
-#define COLAMD_MIN(a,b) (((a) < (b)) ? (a) : (b))
-
 #define ONES_COMPLEMENT(r) (-(r)-1)
 
 /* -------------------------------------------------------------------------- */
@@ -135,54 +128,54 @@
 /* ========================================================================== */
 
 // == Row and Column structures ==
-template <typename Index>
+template <typename IndexType>
 struct colamd_col
 {
-  Index start ;   /* index for A of first row in this column, or DEAD */
+  IndexType start ;   /* index for A of first row in this column, or DEAD */
   /* if column is dead */
-  Index length ;  /* number of rows in this column */
+  IndexType length ;  /* number of rows in this column */
   union
   {
-    Index thickness ; /* number of original columns represented by this */
+    IndexType thickness ; /* number of original columns represented by this */
     /* col, if the column is alive */
-    Index parent ;  /* parent in parent tree super-column structure, if */
+    IndexType parent ;  /* parent in parent tree super-column structure, if */
     /* the column is dead */
   } shared1 ;
   union
   {
-    Index score ; /* the score used to maintain heap, if col is alive */
-    Index order ; /* pivot ordering of this column, if col is dead */
+    IndexType score ; /* the score used to maintain heap, if col is alive */
+    IndexType order ; /* pivot ordering of this column, if col is dead */
   } shared2 ;
   union
   {
-    Index headhash ;  /* head of a hash bucket, if col is at the head of */
+    IndexType headhash ;  /* head of a hash bucket, if col is at the head of */
     /* a degree list */
-    Index hash ;  /* hash value, if col is not in a degree list */
-    Index prev ;  /* previous column in degree list, if col is in a */
+    IndexType hash ;  /* hash value, if col is not in a degree list */
+    IndexType prev ;  /* previous column in degree list, if col is in a */
     /* degree list (but not at the head of a degree list) */
   } shared3 ;
   union
   {
-    Index degree_next ; /* next column, if col is in a degree list */
-    Index hash_next ;   /* next column, if col is in a hash list */
+    IndexType degree_next ; /* next column, if col is in a degree list */
+    IndexType hash_next ;   /* next column, if col is in a hash list */
   } shared4 ;
   
 };
  
-template <typename Index>
+template <typename IndexType>
 struct Colamd_Row
 {
-  Index start ;   /* index for A of first col in this row */
-  Index length ;  /* number of principal columns in this row */
+  IndexType start ;   /* index for A of first col in this row */
+  IndexType length ;  /* number of principal columns in this row */
   union
   {
-    Index degree ;  /* number of principal & non-principal columns in row */
-    Index p ;   /* used as a row pointer in init_rows_cols () */
+    IndexType degree ;  /* number of principal & non-principal columns in row */
+    IndexType p ;   /* used as a row pointer in init_rows_cols () */
   } shared1 ;
   union
   {
-    Index mark ;  /* for computing set differences and marking dead rows*/
-    Index first_column ;/* first column in row (used in garbage collection) */
+    IndexType mark ;  /* for computing set differences and marking dead rows*/
+    IndexType first_column ;/* first column in row (used in garbage collection) */
   } shared2 ;
   
 };
@@ -202,38 +195,38 @@
   
   This macro is not needed when using symamd.
   
-  Explicit typecast to Index added Sept. 23, 2002, COLAMD version 2.2, to avoid
+  Explicit typecast to IndexType added Sept. 23, 2002, COLAMD version 2.2, to avoid
   gcc -pedantic warning messages.
 */
-template <typename Index>
-inline Index colamd_c(Index n_col) 
-{ return Index( ((n_col) + 1) * sizeof (colamd_col<Index>) / sizeof (Index) ) ; }
+template <typename IndexType>
+inline IndexType colamd_c(IndexType n_col) 
+{ return IndexType( ((n_col) + 1) * sizeof (colamd_col<IndexType>) / sizeof (IndexType) ) ; }
 
-template <typename Index>
-inline Index  colamd_r(Index n_row)
-{ return Index(((n_row) + 1) * sizeof (Colamd_Row<Index>) / sizeof (Index)); }
+template <typename IndexType>
+inline IndexType  colamd_r(IndexType n_row)
+{ return IndexType(((n_row) + 1) * sizeof (Colamd_Row<IndexType>) / sizeof (IndexType)); }
 
 // Prototypes of non-user callable routines
-template <typename Index>
-static Index init_rows_cols (Index n_row, Index n_col, Colamd_Row<Index> Row [], colamd_col<Index> col [], Index A [], Index p [], Index stats[COLAMD_STATS] ); 
+template <typename IndexType>
+static IndexType init_rows_cols (IndexType n_row, IndexType n_col, Colamd_Row<IndexType> Row [], colamd_col<IndexType> col [], IndexType A [], IndexType p [], IndexType stats[COLAMD_STATS] ); 
 
-template <typename Index>
-static void init_scoring (Index n_row, Index n_col, Colamd_Row<Index> Row [], colamd_col<Index> Col [], Index A [], Index head [], double knobs[COLAMD_KNOBS], Index *p_n_row2, Index *p_n_col2, Index *p_max_deg);
+template <typename IndexType>
+static void init_scoring (IndexType n_row, IndexType n_col, Colamd_Row<IndexType> Row [], colamd_col<IndexType> Col [], IndexType A [], IndexType head [], double knobs[COLAMD_KNOBS], IndexType *p_n_row2, IndexType *p_n_col2, IndexType *p_max_deg);
 
-template <typename Index>
-static Index find_ordering (Index n_row, Index n_col, Index Alen, Colamd_Row<Index> Row [], colamd_col<Index> Col [], Index A [], Index head [], Index n_col2, Index max_deg, Index pfree);
+template <typename IndexType>
+static IndexType find_ordering (IndexType n_row, IndexType n_col, IndexType Alen, Colamd_Row<IndexType> Row [], colamd_col<IndexType> Col [], IndexType A [], IndexType head [], IndexType n_col2, IndexType max_deg, IndexType pfree);
 
-template <typename Index>
-static void order_children (Index n_col, colamd_col<Index> Col [], Index p []);
+template <typename IndexType>
+static void order_children (IndexType n_col, colamd_col<IndexType> Col [], IndexType p []);
 
-template <typename Index>
-static void detect_super_cols (colamd_col<Index> Col [], Index A [], Index head [], Index row_start, Index row_length ) ;
+template <typename IndexType>
+static void detect_super_cols (colamd_col<IndexType> Col [], IndexType A [], IndexType head [], IndexType row_start, IndexType row_length ) ;
 
-template <typename Index>
-static Index garbage_collection (Index n_row, Index n_col, Colamd_Row<Index> Row [], colamd_col<Index> Col [], Index A [], Index *pfree) ;
+template <typename IndexType>
+static IndexType garbage_collection (IndexType n_row, IndexType n_col, Colamd_Row<IndexType> Row [], colamd_col<IndexType> Col [], IndexType A [], IndexType *pfree) ;
 
-template <typename Index>
-static inline  Index clear_mark (Index n_row, Colamd_Row<Index> Row [] ) ;
+template <typename IndexType>
+static inline  IndexType clear_mark (IndexType n_row, Colamd_Row<IndexType> Row [] ) ;
 
 /* === No debugging ========================================================= */
 
@@ -260,8 +253,8 @@
  * \param n_col number of columns in A
  * \return recommended value of Alen for use by colamd
  */
-template <typename Index>
-inline Index colamd_recommended ( Index nnz, Index n_row, Index n_col)
+template <typename IndexType>
+inline IndexType colamd_recommended ( IndexType nnz, IndexType n_row, IndexType n_col)
 {
   if ((nnz) < 0 || (n_row) < 0 || (n_col) < 0)
     return (-1);
@@ -325,22 +318,22 @@
  * \param knobs parameter settings for colamd
  * \param stats colamd output statistics and error codes
  */
-template <typename Index>
-static bool colamd(Index n_row, Index n_col, Index Alen, Index *A, Index *p, double knobs[COLAMD_KNOBS], Index stats[COLAMD_STATS])
+template <typename IndexType>
+static bool colamd(IndexType n_row, IndexType n_col, IndexType Alen, IndexType *A, IndexType *p, double knobs[COLAMD_KNOBS], IndexType stats[COLAMD_STATS])
 {
   /* === Local variables ================================================== */
   
-  Index i ;     /* loop index */
-  Index nnz ;     /* nonzeros in A */
-  Index Row_size ;    /* size of Row [], in integers */
-  Index Col_size ;    /* size of Col [], in integers */
-  Index need ;      /* minimum required length of A */
-  Colamd_Row<Index> *Row ;   /* pointer into A of Row [0..n_row] array */
-  colamd_col<Index> *Col ;   /* pointer into A of Col [0..n_col] array */
-  Index n_col2 ;    /* number of non-dense, non-empty columns */
-  Index n_row2 ;    /* number of non-dense, non-empty rows */
-  Index ngarbage ;    /* number of garbage collections performed */
-  Index max_deg ;   /* maximum row degree */
+  IndexType i ;     /* loop index */
+  IndexType nnz ;     /* nonzeros in A */
+  IndexType Row_size ;    /* size of Row [], in integers */
+  IndexType Col_size ;    /* size of Col [], in integers */
+  IndexType need ;      /* minimum required length of A */
+  Colamd_Row<IndexType> *Row ;   /* pointer into A of Row [0..n_row] array */
+  colamd_col<IndexType> *Col ;   /* pointer into A of Col [0..n_col] array */
+  IndexType n_col2 ;    /* number of non-dense, non-empty columns */
+  IndexType n_row2 ;    /* number of non-dense, non-empty rows */
+  IndexType ngarbage ;    /* number of garbage collections performed */
+  IndexType max_deg ;   /* maximum row degree */
   double default_knobs [COLAMD_KNOBS] ; /* default knobs array */
   
   
@@ -431,8 +424,8 @@
   }
   
   Alen -= Col_size + Row_size ;
-  Col = (colamd_col<Index> *) &A [Alen] ;
-  Row = (Colamd_Row<Index> *) &A [Alen + Col_size] ;
+  Col = (colamd_col<IndexType> *) &A [Alen] ;
+  Row = (Colamd_Row<IndexType> *) &A [Alen + Col_size] ;
 
   /* === Construct the row and column data structures ===================== */
   
@@ -485,29 +478,29 @@
   column form of the matrix.  Returns false if the matrix is invalid,
   true otherwise.  Not user-callable.
 */
-template <typename Index>
-static Index init_rows_cols  /* returns true if OK, or false otherwise */
+template <typename IndexType>
+static IndexType init_rows_cols  /* returns true if OK, or false otherwise */
   (
     /* === Parameters ======================================================= */
 
-    Index n_row,      /* number of rows of A */
-    Index n_col,      /* number of columns of A */
-    Colamd_Row<Index> Row [],    /* of size n_row+1 */
-    colamd_col<Index> Col [],    /* of size n_col+1 */
-    Index A [],     /* row indices of A, of size Alen */
-    Index p [],     /* pointers to columns in A, of size n_col+1 */
-    Index stats [COLAMD_STATS]  /* colamd statistics */ 
+    IndexType n_row,      /* number of rows of A */
+    IndexType n_col,      /* number of columns of A */
+    Colamd_Row<IndexType> Row [],    /* of size n_row+1 */
+    colamd_col<IndexType> Col [],    /* of size n_col+1 */
+    IndexType A [],     /* row indices of A, of size Alen */
+    IndexType p [],     /* pointers to columns in A, of size n_col+1 */
+    IndexType stats [COLAMD_STATS]  /* colamd statistics */ 
     )
 {
   /* === Local variables ================================================== */
 
-  Index col ;     /* a column index */
-  Index row ;     /* a row index */
-  Index *cp ;     /* a column pointer */
-  Index *cp_end ;   /* a pointer to the end of a column */
-  Index *rp ;     /* a row pointer */
-  Index *rp_end ;   /* a pointer to the end of a row */
-  Index last_row ;    /* previous row */
+  IndexType col ;     /* a column index */
+  IndexType row ;     /* a row index */
+  IndexType *cp ;     /* a column pointer */
+  IndexType *cp_end ;   /* a pointer to the end of a column */
+  IndexType *rp ;     /* a row pointer */
+  IndexType *rp_end ;   /* a pointer to the end of a row */
+  IndexType last_row ;    /* previous row */
 
   /* === Initialize columns, and check column pointers ==================== */
 
@@ -516,7 +509,7 @@
     Col [col].start = p [col] ;
     Col [col].length = p [col+1] - p [col] ;
 
-    if (Col [col].length < 0)
+    if ((Col [col].length) < 0) // extra parentheses to work-around gcc bug 10200
     {
       /* column pointers must be non-decreasing */
       stats [COLAMD_STATUS] = COLAMD_ERROR_col_length_negative ;
@@ -701,46 +694,46 @@
   Kills dense or empty columns and rows, calculates an initial score for
   each column, and places all columns in the degree lists.  Not user-callable.
 */
-template <typename Index>
+template <typename IndexType>
 static void init_scoring
   (
     /* === Parameters ======================================================= */
 
-    Index n_row,      /* number of rows of A */
-    Index n_col,      /* number of columns of A */
-    Colamd_Row<Index> Row [],    /* of size n_row+1 */
-    colamd_col<Index> Col [],    /* of size n_col+1 */
-    Index A [],     /* column form and row form of A */
-    Index head [],    /* of size n_col+1 */
+    IndexType n_row,      /* number of rows of A */
+    IndexType n_col,      /* number of columns of A */
+    Colamd_Row<IndexType> Row [],    /* of size n_row+1 */
+    colamd_col<IndexType> Col [],    /* of size n_col+1 */
+    IndexType A [],     /* column form and row form of A */
+    IndexType head [],    /* of size n_col+1 */
     double knobs [COLAMD_KNOBS],/* parameters */
-    Index *p_n_row2,    /* number of non-dense, non-empty rows */
-    Index *p_n_col2,    /* number of non-dense, non-empty columns */
-    Index *p_max_deg    /* maximum row degree */
+    IndexType *p_n_row2,    /* number of non-dense, non-empty rows */
+    IndexType *p_n_col2,    /* number of non-dense, non-empty columns */
+    IndexType *p_max_deg    /* maximum row degree */
     )
 {
   /* === Local variables ================================================== */
 
-  Index c ;     /* a column index */
-  Index r, row ;    /* a row index */
-  Index *cp ;     /* a column pointer */
-  Index deg ;     /* degree of a row or column */
-  Index *cp_end ;   /* a pointer to the end of a column */
-  Index *new_cp ;   /* new column pointer */
-  Index col_length ;    /* length of pruned column */
-  Index score ;     /* current column score */
-  Index n_col2 ;    /* number of non-dense, non-empty columns */
-  Index n_row2 ;    /* number of non-dense, non-empty rows */
-  Index dense_row_count ; /* remove rows with more entries than this */
-  Index dense_col_count ; /* remove cols with more entries than this */
-  Index min_score ;   /* smallest column score */
-  Index max_deg ;   /* maximum row degree */
-  Index next_col ;    /* Used to add to degree list.*/
+  IndexType c ;     /* a column index */
+  IndexType r, row ;    /* a row index */
+  IndexType *cp ;     /* a column pointer */
+  IndexType deg ;     /* degree of a row or column */
+  IndexType *cp_end ;   /* a pointer to the end of a column */
+  IndexType *new_cp ;   /* new column pointer */
+  IndexType col_length ;    /* length of pruned column */
+  IndexType score ;     /* current column score */
+  IndexType n_col2 ;    /* number of non-dense, non-empty columns */
+  IndexType n_row2 ;    /* number of non-dense, non-empty rows */
+  IndexType dense_row_count ; /* remove rows with more entries than this */
+  IndexType dense_col_count ; /* remove cols with more entries than this */
+  IndexType min_score ;   /* smallest column score */
+  IndexType max_deg ;   /* maximum row degree */
+  IndexType next_col ;    /* Used to add to degree list.*/
 
 
   /* === Extract knobs ==================================================== */
 
-  dense_row_count = COLAMD_MAX (0, COLAMD_MIN (knobs [COLAMD_DENSE_ROW] * n_col, n_col)) ;
-  dense_col_count = COLAMD_MAX (0, COLAMD_MIN (knobs [COLAMD_DENSE_COL] * n_row, n_row)) ;
+  dense_row_count = numext::maxi(IndexType(0), numext::mini(IndexType(knobs [COLAMD_DENSE_ROW] * n_col), n_col)) ;
+  dense_col_count = numext::maxi(IndexType(0), numext::mini(IndexType(knobs [COLAMD_DENSE_COL] * n_row), n_row)) ;
   COLAMD_DEBUG1 (("colamd: densecount: %d %d\n", dense_row_count, dense_col_count)) ;
   max_deg = 0 ;
   n_col2 = n_col ;
@@ -804,7 +797,7 @@
     else
     {
       /* keep track of max degree of remaining rows */
-      max_deg = COLAMD_MAX (max_deg, deg) ;
+      max_deg = numext::maxi(max_deg, deg) ;
     }
   }
   COLAMD_DEBUG1 (("colamd: Dense and null rows killed: %d\n", n_row - n_row2)) ;
@@ -842,10 +835,10 @@
       /* add row's external degree */
       score += Row [row].shared1.degree - 1 ;
       /* guard against integer overflow */
-      score = COLAMD_MIN (score, n_col) ;
+      score = numext::mini(score, n_col) ;
     }
     /* determine pruned column length */
-    col_length = (Index) (new_cp - &A [Col [c].start]) ;
+    col_length = (IndexType) (new_cp - &A [Col [c].start]) ;
     if (col_length == 0)
     {
       /* a newly-made null column (all rows in this col are "dense" */
@@ -914,7 +907,7 @@
       head [score] = c ;
 
       /* see if this score is less than current min */
-      min_score = COLAMD_MIN (min_score, score) ;
+      min_score = numext::mini(min_score, score) ;
 
 
     }
@@ -938,56 +931,56 @@
   (no supercolumns on input).  Uses a minimum approximate column minimum
   degree ordering method.  Not user-callable.
 */
-template <typename Index>
-static Index find_ordering /* return the number of garbage collections */
+template <typename IndexType>
+static IndexType find_ordering /* return the number of garbage collections */
   (
     /* === Parameters ======================================================= */
 
-    Index n_row,      /* number of rows of A */
-    Index n_col,      /* number of columns of A */
-    Index Alen,     /* size of A, 2*nnz + n_col or larger */
-    Colamd_Row<Index> Row [],    /* of size n_row+1 */
-    colamd_col<Index> Col [],    /* of size n_col+1 */
-    Index A [],     /* column form and row form of A */
-    Index head [],    /* of size n_col+1 */
-    Index n_col2,     /* Remaining columns to order */
-    Index max_deg,    /* Maximum row degree */
-    Index pfree     /* index of first free slot (2*nnz on entry) */
+    IndexType n_row,      /* number of rows of A */
+    IndexType n_col,      /* number of columns of A */
+    IndexType Alen,     /* size of A, 2*nnz + n_col or larger */
+    Colamd_Row<IndexType> Row [],    /* of size n_row+1 */
+    colamd_col<IndexType> Col [],    /* of size n_col+1 */
+    IndexType A [],     /* column form and row form of A */
+    IndexType head [],    /* of size n_col+1 */
+    IndexType n_col2,     /* Remaining columns to order */
+    IndexType max_deg,    /* Maximum row degree */
+    IndexType pfree     /* index of first free slot (2*nnz on entry) */
     )
 {
   /* === Local variables ================================================== */
 
-  Index k ;     /* current pivot ordering step */
-  Index pivot_col ;   /* current pivot column */
-  Index *cp ;     /* a column pointer */
-  Index *rp ;     /* a row pointer */
-  Index pivot_row ;   /* current pivot row */
-  Index *new_cp ;   /* modified column pointer */
-  Index *new_rp ;   /* modified row pointer */
-  Index pivot_row_start ; /* pointer to start of pivot row */
-  Index pivot_row_degree ;  /* number of columns in pivot row */
-  Index pivot_row_length ;  /* number of supercolumns in pivot row */
-  Index pivot_col_score ; /* score of pivot column */
-  Index needed_memory ;   /* free space needed for pivot row */
-  Index *cp_end ;   /* pointer to the end of a column */
-  Index *rp_end ;   /* pointer to the end of a row */
-  Index row ;     /* a row index */
-  Index col ;     /* a column index */
-  Index max_score ;   /* maximum possible score */
-  Index cur_score ;   /* score of current column */
+  IndexType k ;     /* current pivot ordering step */
+  IndexType pivot_col ;   /* current pivot column */
+  IndexType *cp ;     /* a column pointer */
+  IndexType *rp ;     /* a row pointer */
+  IndexType pivot_row ;   /* current pivot row */
+  IndexType *new_cp ;   /* modified column pointer */
+  IndexType *new_rp ;   /* modified row pointer */
+  IndexType pivot_row_start ; /* pointer to start of pivot row */
+  IndexType pivot_row_degree ;  /* number of columns in pivot row */
+  IndexType pivot_row_length ;  /* number of supercolumns in pivot row */
+  IndexType pivot_col_score ; /* score of pivot column */
+  IndexType needed_memory ;   /* free space needed for pivot row */
+  IndexType *cp_end ;   /* pointer to the end of a column */
+  IndexType *rp_end ;   /* pointer to the end of a row */
+  IndexType row ;     /* a row index */
+  IndexType col ;     /* a column index */
+  IndexType max_score ;   /* maximum possible score */
+  IndexType cur_score ;   /* score of current column */
   unsigned int hash ;   /* hash value for supernode detection */
-  Index head_column ;   /* head of hash bucket */
-  Index first_col ;   /* first column in hash bucket */
-  Index tag_mark ;    /* marker value for mark array */
-  Index row_mark ;    /* Row [row].shared2.mark */
-  Index set_difference ;  /* set difference size of row with pivot row */
-  Index min_score ;   /* smallest column score */
-  Index col_thickness ;   /* "thickness" (no. of columns in a supercol) */
-  Index max_mark ;    /* maximum value of tag_mark */
-  Index pivot_col_thickness ; /* number of columns represented by pivot col */
-  Index prev_col ;    /* Used by Dlist operations. */
-  Index next_col ;    /* Used by Dlist operations. */
-  Index ngarbage ;    /* number of garbage collections performed */
+  IndexType head_column ;   /* head of hash bucket */
+  IndexType first_col ;   /* first column in hash bucket */
+  IndexType tag_mark ;    /* marker value for mark array */
+  IndexType row_mark ;    /* Row [row].shared2.mark */
+  IndexType set_difference ;  /* set difference size of row with pivot row */
+  IndexType min_score ;   /* smallest column score */
+  IndexType col_thickness ;   /* "thickness" (no. of columns in a supercol) */
+  IndexType max_mark ;    /* maximum value of tag_mark */
+  IndexType pivot_col_thickness ; /* number of columns represented by pivot col */
+  IndexType prev_col ;    /* Used by Dlist operations. */
+  IndexType next_col ;    /* Used by Dlist operations. */
+  IndexType ngarbage ;    /* number of garbage collections performed */
 
 
   /* === Initialization and clear mark ==================================== */
@@ -1011,7 +1004,7 @@
     COLAMD_ASSERT (head [min_score] >= COLAMD_EMPTY) ;
 
     /* get pivot column from head of minimum degree list */
-    while (head [min_score] == COLAMD_EMPTY && min_score < n_col)
+    while (min_score < n_col && head [min_score] == COLAMD_EMPTY)
     {
       min_score++ ;
     }
@@ -1040,7 +1033,7 @@
 
     /* === Garbage_collection, if necessary ============================= */
 
-    needed_memory = COLAMD_MIN (pivot_col_score, n_col - k) ;
+    needed_memory = numext::mini(pivot_col_score, n_col - k) ;
     if (pfree + needed_memory >= Alen)
     {
       pfree = Eigen::internal::garbage_collection (n_row, n_col, Row, Col, A, &A [pfree]) ;
@@ -1099,7 +1092,7 @@
 
     /* clear tag on pivot column */
     Col [pivot_col].shared1.thickness = pivot_col_thickness ;
-    max_deg = COLAMD_MAX (max_deg, pivot_row_degree) ;
+    max_deg = numext::maxi(max_deg, pivot_row_degree) ;
 
 
     /* === Kill all rows used to construct pivot row ==================== */
@@ -1273,11 +1266,11 @@
 	/* add set difference */
 	cur_score += row_mark - tag_mark ;
 	/* integer overflow... */
-	cur_score = COLAMD_MIN (cur_score, n_col) ;
+	cur_score = numext::mini(cur_score, n_col) ;
       }
 
       /* recompute the column's length */
-      Col [col].length = (Index) (new_cp - &A [Col [col].start]) ;
+      Col [col].length = (IndexType) (new_cp - &A [Col [col].start]) ;
 
       /* === Further mass elimination ================================= */
 
@@ -1325,7 +1318,7 @@
 	Col [col].shared4.hash_next = first_col ;
 
 	/* save hash function in Col [col].shared3.hash */
-	Col [col].shared3.hash = (Index) hash ;
+	Col [col].shared3.hash = (IndexType) hash ;
 	COLAMD_ASSERT (COL_IS_ALIVE (col)) ;
       }
     }
@@ -1386,7 +1379,7 @@
       cur_score -= Col [col].shared1.thickness ;
 
       /* make sure score is less or equal than the max score */
-      cur_score = COLAMD_MIN (cur_score, max_score) ;
+      cur_score = numext::mini(cur_score, max_score) ;
       COLAMD_ASSERT (cur_score >= 0) ;
 
       /* store updated score */
@@ -1409,7 +1402,7 @@
       head [cur_score] = col ;
 
       /* see if this score is less than current min */
-      min_score = COLAMD_MIN (min_score, cur_score) ;
+      min_score = numext::mini(min_score, cur_score) ;
 
     }
 
@@ -1420,7 +1413,7 @@
       /* update pivot row length to reflect any cols that were killed */
       /* during super-col detection and mass elimination */
       Row [pivot_row].start  = pivot_row_start ;
-      Row [pivot_row].length = (Index) (new_rp - &A[pivot_row_start]) ;
+      Row [pivot_row].length = (IndexType) (new_rp - &A[pivot_row_start]) ;
       Row [pivot_row].shared1.degree = pivot_row_degree ;
       Row [pivot_row].shared2.mark = 0 ;
       /* pivot row is no longer dead */
@@ -1449,22 +1442,22 @@
   taken by this routine is O (n_col), that is, linear in the number of
   columns.  Not user-callable.
 */
-template <typename Index>
+template <typename IndexType>
 static inline  void order_children
 (
   /* === Parameters ======================================================= */
 
-  Index n_col,      /* number of columns of A */
-  colamd_col<Index> Col [],    /* of size n_col+1 */
-  Index p []      /* p [0 ... n_col-1] is the column permutation*/
+  IndexType n_col,      /* number of columns of A */
+  colamd_col<IndexType> Col [],    /* of size n_col+1 */
+  IndexType p []      /* p [0 ... n_col-1] is the column permutation*/
   )
 {
   /* === Local variables ================================================== */
 
-  Index i ;     /* loop counter for all columns */
-  Index c ;     /* column index */
-  Index parent ;    /* index of column's parent */
-  Index order ;     /* column's order */
+  IndexType i ;     /* loop counter for all columns */
+  IndexType c ;     /* column index */
+  IndexType parent ;    /* index of column's parent */
+  IndexType order ;     /* column's order */
 
   /* === Order each non-principal column ================================== */
 
@@ -1550,33 +1543,33 @@
   just been computed in the approximate degree computation.
   Not user-callable.
 */
-template <typename Index>
+template <typename IndexType>
 static void detect_super_cols
 (
   /* === Parameters ======================================================= */
   
-  colamd_col<Index> Col [],    /* of size n_col+1 */
-  Index A [],     /* row indices of A */
-  Index head [],    /* head of degree lists and hash buckets */
-  Index row_start,    /* pointer to set of columns to check */
-  Index row_length    /* number of columns to check */
+  colamd_col<IndexType> Col [],    /* of size n_col+1 */
+  IndexType A [],     /* row indices of A */
+  IndexType head [],    /* head of degree lists and hash buckets */
+  IndexType row_start,    /* pointer to set of columns to check */
+  IndexType row_length    /* number of columns to check */
 )
 {
   /* === Local variables ================================================== */
 
-  Index hash ;      /* hash value for a column */
-  Index *rp ;     /* pointer to a row */
-  Index c ;     /* a column index */
-  Index super_c ;   /* column index of the column to absorb into */
-  Index *cp1 ;      /* column pointer for column super_c */
-  Index *cp2 ;      /* column pointer for column c */
-  Index length ;    /* length of column super_c */
-  Index prev_c ;    /* column preceding c in hash bucket */
-  Index i ;     /* loop counter */
-  Index *rp_end ;   /* pointer to the end of the row */
-  Index col ;     /* a column index in the row to check */
-  Index head_column ;   /* first column in hash bucket or degree list */
-  Index first_col ;   /* first column in hash bucket */
+  IndexType hash ;      /* hash value for a column */
+  IndexType *rp ;     /* pointer to a row */
+  IndexType c ;     /* a column index */
+  IndexType super_c ;   /* column index of the column to absorb into */
+  IndexType *cp1 ;      /* column pointer for column super_c */
+  IndexType *cp2 ;      /* column pointer for column c */
+  IndexType length ;    /* length of column super_c */
+  IndexType prev_c ;    /* column preceding c in hash bucket */
+  IndexType i ;     /* loop counter */
+  IndexType *rp_end ;   /* pointer to the end of the row */
+  IndexType col ;     /* a column index in the row to check */
+  IndexType head_column ;   /* first column in hash bucket or degree list */
+  IndexType first_col ;   /* first column in hash bucket */
 
   /* === Consider each column in the row ================================== */
 
@@ -1701,27 +1694,27 @@
   itself linear in the number of nonzeros in the input matrix.
   Not user-callable.
 */
-template <typename Index>
-static Index garbage_collection  /* returns the new value of pfree */
+template <typename IndexType>
+static IndexType garbage_collection  /* returns the new value of pfree */
   (
     /* === Parameters ======================================================= */
     
-    Index n_row,      /* number of rows */
-    Index n_col,      /* number of columns */
-    Colamd_Row<Index> Row [],    /* row info */
-    colamd_col<Index> Col [],    /* column info */
-    Index A [],     /* A [0 ... Alen-1] holds the matrix */
-    Index *pfree      /* &A [0] ... pfree is in use */
+    IndexType n_row,      /* number of rows */
+    IndexType n_col,      /* number of columns */
+    Colamd_Row<IndexType> Row [],    /* row info */
+    colamd_col<IndexType> Col [],    /* column info */
+    IndexType A [],     /* A [0 ... Alen-1] holds the matrix */
+    IndexType *pfree      /* &A [0] ... pfree is in use */
     )
 {
   /* === Local variables ================================================== */
 
-  Index *psrc ;     /* source pointer */
-  Index *pdest ;    /* destination pointer */
-  Index j ;     /* counter */
-  Index r ;     /* a row index */
-  Index c ;     /* a column index */
-  Index length ;    /* length of a row or column */
+  IndexType *psrc ;     /* source pointer */
+  IndexType *pdest ;    /* destination pointer */
+  IndexType j ;     /* counter */
+  IndexType r ;     /* a row index */
+  IndexType c ;     /* a column index */
+  IndexType length ;    /* length of a row or column */
 
   /* === Defragment the columns =========================================== */
 
@@ -1734,7 +1727,7 @@
 
       /* move and compact the column */
       COLAMD_ASSERT (pdest <= psrc) ;
-      Col [c].start = (Index) (pdest - &A [0]) ;
+      Col [c].start = (IndexType) (pdest - &A [0]) ;
       length = Col [c].length ;
       for (j = 0 ; j < length ; j++)
       {
@@ -1744,7 +1737,7 @@
 	  *pdest++ = r ;
 	}
       }
-      Col [c].length = (Index) (pdest - &A [Col [c].start]) ;
+      Col [c].length = (IndexType) (pdest - &A [Col [c].start]) ;
     }
   }
 
@@ -1791,7 +1784,7 @@
 
       /* move and compact the row */
       COLAMD_ASSERT (pdest <= psrc) ;
-      Row [r].start = (Index) (pdest - &A [0]) ;
+      Row [r].start = (IndexType) (pdest - &A [0]) ;
       length = Row [r].length ;
       for (j = 0 ; j < length ; j++)
       {
@@ -1801,7 +1794,7 @@
 	  *pdest++ = c ;
 	}
       }
-      Row [r].length = (Index) (pdest - &A [Row [r].start]) ;
+      Row [r].length = (IndexType) (pdest - &A [Row [r].start]) ;
 
     }
   }
@@ -1810,7 +1803,7 @@
 
   /* === Return the new value of pfree ==================================== */
 
-  return ((Index) (pdest - &A [0])) ;
+  return ((IndexType) (pdest - &A [0])) ;
 }
 
 
@@ -1822,18 +1815,18 @@
   Clears the Row [].shared2.mark array, and returns the new tag_mark.
   Return value is the new tag_mark.  Not user-callable.
 */
-template <typename Index>
-static inline  Index clear_mark  /* return the new value for tag_mark */
+template <typename IndexType>
+static inline  IndexType clear_mark  /* return the new value for tag_mark */
   (
       /* === Parameters ======================================================= */
 
-    Index n_row,    /* number of rows in A */
-    Colamd_Row<Index> Row [] /* Row [0 ... n_row-1].shared2.mark is set to zero */
+    IndexType n_row,    /* number of rows in A */
+    Colamd_Row<IndexType> Row [] /* Row [0 ... n_row-1].shared2.mark is set to zero */
     )
 {
   /* === Local variables ================================================== */
 
-  Index r ;
+  IndexType r ;
 
   for (r = 0 ; r < n_row ; r++)
   {
diff --git a/Eigen/src/OrderingMethods/Ordering.h b/Eigen/src/OrderingMethods/Ordering.h
index f3c31f9..7ea9b14 100644
--- a/Eigen/src/OrderingMethods/Ordering.h
+++ b/Eigen/src/OrderingMethods/Ordering.h
@@ -19,20 +19,21 @@
     
 /** \internal
   * \ingroup OrderingMethods_Module
-  * \returns the symmetric pattern A^T+A from the input matrix A. 
+  * \param[in] A the input non-symmetric matrix
+  * \param[out] symmat the symmetric pattern A^T+A from the input matrix \a A.
   * FIXME: The values should not be considered here
   */
 template<typename MatrixType> 
-void ordering_helper_at_plus_a(const MatrixType& mat, MatrixType& symmat)
+void ordering_helper_at_plus_a(const MatrixType& A, MatrixType& symmat)
 {
   MatrixType C;
-  C = mat.transpose(); // NOTE: Could be  costly
+  C = A.transpose(); // NOTE: Could be  costly
   for (int i = 0; i < C.rows(); i++) 
   {
       for (typename MatrixType::InnerIterator it(C, i); it; ++it)
         it.valueRef() = 0.0;
   }
-  symmat = C + mat; 
+  symmat = C + A;
 }
     
 }
@@ -44,14 +45,14 @@
   *
   * Functor computing the \em approximate \em minimum \em degree ordering
   * If the matrix is not structurally symmetric, an ordering of A^T+A is computed
-  * \tparam  Index The type of indices of the matrix 
+  * \tparam  StorageIndex The type of indices of the matrix 
   * \sa COLAMDOrdering
   */
-template <typename Index>
+template <typename StorageIndex>
 class AMDOrdering
 {
   public:
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
     
     /** Compute the permutation vector from a sparse matrix
      * This routine is much faster if the input matrix is column-major     
@@ -60,7 +61,7 @@
     void operator()(const MatrixType& mat, PermutationType& perm)
     {
       // Compute the symmetric pattern
-      SparseMatrix<typename MatrixType::Scalar, ColMajor, Index> symm;
+      SparseMatrix<typename MatrixType::Scalar, ColMajor, StorageIndex> symm;
       internal::ordering_helper_at_plus_a(mat,symm); 
     
       // Call the AMD routine 
@@ -72,7 +73,7 @@
     template <typename SrcType, unsigned int SrcUpLo> 
     void operator()(const SparseSelfAdjointView<SrcType, SrcUpLo>& mat, PermutationType& perm)
     { 
-      SparseMatrix<typename SrcType::Scalar, ColMajor, Index> C; C = mat;
+      SparseMatrix<typename SrcType::Scalar, ColMajor, StorageIndex> C; C = mat;
       
       // Call the AMD routine 
       // m_mat.prune(keep_diag()); //Remove the diagonal elements 
@@ -88,13 +89,13 @@
   * Functor computing the natural ordering (identity)
   * 
   * \note Returns an empty permutation matrix
-  * \tparam  Index The type of indices of the matrix 
+  * \tparam  StorageIndex The type of indices of the matrix 
   */
-template <typename Index>
+template <typename StorageIndex>
 class NaturalOrdering
 {
   public:
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
     
     /** Compute the permutation vector from a column-major sparse matrix */
     template <typename MatrixType>
@@ -108,15 +109,17 @@
 /** \ingroup OrderingMethods_Module
   * \class COLAMDOrdering
   *
+  * \tparam  StorageIndex The type of indices of the matrix 
+  * 
   * Functor computing the \em column \em approximate \em minimum \em degree ordering 
   * The matrix should be in column-major and \b compressed format (see SparseMatrix::makeCompressed()).
   */
-template<typename Index>
+template<typename StorageIndex>
 class COLAMDOrdering
 {
   public:
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType; 
-    typedef Matrix<Index, Dynamic, 1> IndexVector;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType; 
+    typedef Matrix<StorageIndex, Dynamic, 1> IndexVector;
     
     /** Compute the permutation vector \a perm form the sparse matrix \a mat
       * \warning The input sparse matrix \a mat must be in compressed mode (see SparseMatrix::makeCompressed()).
@@ -126,26 +129,26 @@
     {
       eigen_assert(mat.isCompressed() && "COLAMDOrdering requires a sparse matrix in compressed mode. Call .makeCompressed() before passing it to COLAMDOrdering");
       
-      Index m = mat.rows();
-      Index n = mat.cols();
-      Index nnz = mat.nonZeros();
+      StorageIndex m = StorageIndex(mat.rows());
+      StorageIndex n = StorageIndex(mat.cols());
+      StorageIndex nnz = StorageIndex(mat.nonZeros());
       // Get the recommended value of Alen to be used by colamd
-      Index Alen = internal::colamd_recommended(nnz, m, n); 
+      StorageIndex Alen = internal::colamd_recommended(nnz, m, n); 
       // Set the default parameters
       double knobs [COLAMD_KNOBS]; 
-      Index stats [COLAMD_STATS];
+      StorageIndex stats [COLAMD_STATS];
       internal::colamd_set_defaults(knobs);
       
       IndexVector p(n+1), A(Alen); 
-      for(Index i=0; i <= n; i++)   p(i) = mat.outerIndexPtr()[i];
-      for(Index i=0; i < nnz; i++)  A(i) = mat.innerIndexPtr()[i];
+      for(StorageIndex i=0; i <= n; i++)   p(i) = mat.outerIndexPtr()[i];
+      for(StorageIndex i=0; i < nnz; i++)  A(i) = mat.innerIndexPtr()[i];
       // Call Colamd routine to compute the ordering 
-      Index info = internal::colamd(m, n, Alen, A.data(), p.data(), knobs, stats); 
+      StorageIndex info = internal::colamd(m, n, Alen, A.data(), p.data(), knobs, stats); 
       EIGEN_UNUSED_VARIABLE(info);
       eigen_assert( info && "COLAMD failed " );
       
       perm.resize(n);
-      for (Index i = 0; i < n; i++) perm.indices()(p(i)) = i;
+      for (StorageIndex i = 0; i < n; i++) perm.indices()(p(i)) = i;
     }
 };
 
diff --git a/Eigen/src/PaStiXSupport/CMakeLists.txt b/Eigen/src/PaStiXSupport/CMakeLists.txt
deleted file mode 100644
index 28c657e..0000000
--- a/Eigen/src/PaStiXSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_PastixSupport_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_PastixSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/PaStiXSupport COMPONENT Devel
-  )
diff --git a/Eigen/src/PaStiXSupport/PaStiXSupport.h b/Eigen/src/PaStiXSupport/PaStiXSupport.h
index a955287..160d8a5 100644
--- a/Eigen/src/PaStiXSupport/PaStiXSupport.h
+++ b/Eigen/src/PaStiXSupport/PaStiXSupport.h
@@ -12,6 +12,14 @@
 
 namespace Eigen { 
 
+#if defined(DCOMPLEX)
+  #define PASTIX_COMPLEX  COMPLEX
+  #define PASTIX_DCOMPLEX DCOMPLEX
+#else
+  #define PASTIX_COMPLEX  std::complex<float>
+  #define PASTIX_DCOMPLEX std::complex<double>
+#endif
+
 /** \ingroup PaStiXSupport_Module
   * \brief Interface to the PaStix solver
   * 
@@ -35,7 +43,7 @@
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -44,7 +52,7 @@
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -53,35 +61,35 @@
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, float *vals, int *perm, int * invp, float *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, float *vals, int *perm, int * invp, float *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
     s_pastix(pastix_data, pastix_comm, n, ptr, idx, vals, perm, invp, x, nbrhs, iparm, dparm); 
   }
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, double *vals, int *perm, int * invp, double *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, double *vals, int *perm, int * invp, double *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
     d_pastix(pastix_data, pastix_comm, n, ptr, idx, vals, perm, invp, x, nbrhs, iparm, dparm); 
   }
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<float> *vals, int *perm, int * invp, std::complex<float> *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<float> *vals, int *perm, int * invp, std::complex<float> *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
-    c_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<COMPLEX*>(vals), perm, invp, reinterpret_cast<COMPLEX*>(x), nbrhs, iparm, dparm); 
+    c_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<PASTIX_COMPLEX*>(vals), perm, invp, reinterpret_cast<PASTIX_COMPLEX*>(x), nbrhs, iparm, dparm); 
   }
   
-  void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<double> *vals, int *perm, int * invp, std::complex<double> *x, int nbrhs, int *iparm, double *dparm)
+  inline void eigen_pastix(pastix_data_t **pastix_data, int pastix_comm, int n, int *ptr, int *idx, std::complex<double> *vals, int *perm, int * invp, std::complex<double> *x, int nbrhs, int *iparm, double *dparm)
   {
     if (n == 0) { ptr = NULL; idx = NULL; vals = NULL; }
     if (nbrhs == 0) {x = NULL; nbrhs=1;}
-    z_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<DCOMPLEX*>(vals), perm, invp, reinterpret_cast<DCOMPLEX*>(x), nbrhs, iparm, dparm); 
+    z_pastix(pastix_data, pastix_comm, n, ptr, idx, reinterpret_cast<PASTIX_DCOMPLEX*>(vals), perm, invp, reinterpret_cast<PASTIX_DCOMPLEX*>(x), nbrhs, iparm, dparm); 
   }
 
   // Convert the matrix  to Fortran-style Numbering
@@ -117,20 +125,30 @@
 // This is the base class to interface with PaStiX functions. 
 // Users should not used this class directly. 
 template <class Derived>
-class PastixBase : internal::noncopyable
+class PastixBase : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
   public:
+    using Base::_solve_impl;
+    
     typedef typename internal::pastix_traits<Derived>::MatrixType _MatrixType;
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar,Dynamic,1> Vector;
     typedef SparseMatrix<Scalar, ColMajor> ColSpMatrix;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
     
   public:
     
-    PastixBase() : m_initisOk(false), m_analysisIsOk(false), m_factorizationIsOk(false), m_isInitialized(false), m_pastixdata(0), m_size(0)
+    PastixBase() : m_initisOk(false), m_analysisIsOk(false), m_factorizationIsOk(false), m_pastixdata(0), m_size(0)
     {
       init();
     }
@@ -139,39 +157,16 @@
     {
       clean();
     }
-
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<PastixBase, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Pastix solver is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PastixBase::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<PastixBase, Rhs>(*this, b.derived());
-    }
     
     template<typename Rhs,typename Dest>
-    bool _solve (const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const;
+    bool _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const;
     
-    Derived& derived()
-    {
-      return *static_cast<Derived*>(this);
-    }
-    const Derived& derived() const
-    {
-      return *static_cast<const Derived*>(this);
-    }
-
     /** Returns a reference to the integer vector IPARM of PaStiX parameters
       * to modify the default parameters. 
       * The statistics related to the different phases of factorization and solve are saved here as well
       * \sa analyzePattern() factorize()
       */
-    Array<Index,IPARM_SIZE,1>& iparm()
+    Array<StorageIndex,IPARM_SIZE,1>& iparm()
     {
       return m_iparm; 
     }
@@ -189,7 +184,7 @@
       * The statistics related to the different phases of factorization and solve are saved here as well
       * \sa analyzePattern() factorize()
       */
-    Array<RealScalar,IPARM_SIZE,1>& dparm()
+    Array<double,DPARM_SIZE,1>& dparm()
     {
       return m_dparm; 
     }
@@ -220,20 +215,6 @@
       return m_info;
     }
     
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<PastixBase, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Pastix LU, LLT or LDLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PastixBase::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<PastixBase, Rhs>(*this, b.derived());
-    }
-    
   protected:
 
     // Initialize the Pastix data structure, check the matrix
@@ -260,14 +241,13 @@
     int m_initisOk; 
     int m_analysisIsOk;
     int m_factorizationIsOk;
-    bool m_isInitialized;
     mutable ComputationInfo m_info; 
     mutable pastix_data_t *m_pastixdata; // Data structure for pastix
     mutable int m_comm; // The MPI communicator identifier
-    mutable Matrix<int,IPARM_SIZE,1> m_iparm; // integer vector for the input parameters
-    mutable Matrix<double,DPARM_SIZE,1> m_dparm; // Scalar vector for the input parameters
-    mutable Matrix<Index,Dynamic,1> m_perm;  // Permutation vector
-    mutable Matrix<Index,Dynamic,1> m_invp;  // Inverse permutation vector
+    mutable Array<int,IPARM_SIZE,1> m_iparm; // integer vector for the input parameters
+    mutable Array<double,DPARM_SIZE,1> m_dparm; // Scalar vector for the input parameters
+    mutable Matrix<StorageIndex,Dynamic,1> m_perm;  // Permutation vector
+    mutable Matrix<StorageIndex,Dynamic,1> m_invp;  // Inverse permutation vector
     mutable int m_size; // Size of the matrix 
 }; 
 
@@ -288,7 +268,7 @@
          0, 0, 0, 1, m_iparm.data(), m_dparm.data());
   
   m_iparm[IPARM_MATRIX_VERIFICATION] = API_NO;
-  m_iparm[IPARM_VERBOSE]             = 2;
+  m_iparm[IPARM_VERBOSE]             = API_VERBOSE_NOT;
   m_iparm[IPARM_ORDERING]            = API_ORDER_SCOTCH;
   m_iparm[IPARM_INCOMPLETE]          = API_NO;
   m_iparm[IPARM_OOC_LIMIT]           = 2000;
@@ -320,7 +300,6 @@
   factorize(mat);
   
   m_iparm(IPARM_MATRIX_VERIFICATION) = API_NO;
-  m_isInitialized = m_factorizationIsOk;
 }
 
 
@@ -333,7 +312,7 @@
   if(m_size>0)
     clean();
   
-  m_size = mat.rows();
+  m_size = internal::convert_index<int>(mat.rows());
   m_perm.resize(m_size);
   m_invp.resize(m_size);
   
@@ -362,7 +341,7 @@
   eigen_assert(m_analysisIsOk && "The analysis phase should be called before the factorization phase");
   m_iparm(IPARM_START_TASK) = API_TASK_NUMFACT;
   m_iparm(IPARM_END_TASK) = API_TASK_NUMFACT;
-  m_size = mat.rows();
+  m_size = internal::convert_index<int>(mat.rows());
   
   internal::eigen_pastix(&m_pastixdata, MPI_COMM_WORLD, m_size, mat.outerIndexPtr(), mat.innerIndexPtr(),
                mat.valuePtr(), m_perm.data(), m_invp.data(), 0, 0, m_iparm.data(), m_dparm.data());
@@ -385,7 +364,7 @@
 /* Solve the system */
 template<typename Base>
 template<typename Rhs,typename Dest>
-bool PastixBase<Base>::_solve (const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const
+bool PastixBase<Base>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &x) const
 {
   eigen_assert(m_isInitialized && "The matrix should be factorized first");
   EIGEN_STATIC_ASSERT((Dest::Flags&RowMajorBit)==0,
@@ -398,7 +377,7 @@
     m_iparm[IPARM_START_TASK]          = API_TASK_SOLVE;
     m_iparm[IPARM_END_TASK]            = API_TASK_REFINE;
   
-    internal::eigen_pastix(&m_pastixdata, MPI_COMM_WORLD, x.rows(), 0, 0, 0,
+    internal::eigen_pastix(&m_pastixdata, MPI_COMM_WORLD, internal::convert_index<int>(x.rows()), 0, 0, 0,
                            m_perm.data(), m_invp.data(), &x(0, i), rhs, m_iparm.data(), m_dparm.data());
   }
   
@@ -423,8 +402,10 @@
   * NOTE : Note that if the analysis and factorization phase are called separately, 
   * the input matrix will be symmetrized at each call, hence it is advised to 
   * symmetrize the matrix in a end-user program and set \p IsStrSym to true
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   * 
   */
 template<typename _MatrixType, bool IsStrSym>
@@ -434,7 +415,7 @@
     typedef _MatrixType MatrixType;
     typedef PastixBase<PastixLU<MatrixType> > Base;
     typedef typename Base::ColSpMatrix ColSpMatrix;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     
   public:
     PastixLU() : Base()
@@ -442,7 +423,7 @@
       init();
     }
     
-    PastixLU(const MatrixType& matrix):Base()
+    explicit PastixLU(const MatrixType& matrix):Base()
     {
       init();
       compute(matrix);
@@ -534,8 +515,10 @@
   * 
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam UpLo The part of the matrix to use : Lower or Upper. The default is Lower as required by PaStiX
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLLT
   */
 template<typename _MatrixType, int _UpLo>
 class PastixLLT : public PastixBase< PastixLLT<_MatrixType, _UpLo> >
@@ -552,7 +535,7 @@
       init();
     }
     
-    PastixLLT(const MatrixType& matrix):Base()
+    explicit PastixLLT(const MatrixType& matrix):Base()
     {
       init();
       compute(matrix);
@@ -598,6 +581,7 @@
     
     void grabMatrix(const MatrixType& matrix, ColSpMatrix& out)
     {
+      out.resize(matrix.rows(), matrix.cols());
       // Pastix supports only lower, column-major matrices 
       out.template selfadjointView<Lower>() = matrix.template selfadjointView<UpLo>();
       internal::c_to_fortran_numbering(out);
@@ -615,8 +599,10 @@
   * 
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam UpLo The part of the matrix to use : Lower or Upper. The default is Lower as required by PaStiX
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLDLT
   */
 template<typename _MatrixType, int _UpLo>
 class PastixLDLT : public PastixBase< PastixLDLT<_MatrixType, _UpLo> >
@@ -633,7 +619,7 @@
       init();
     }
     
-    PastixLDLT(const MatrixType& matrix):Base()
+    explicit PastixLDLT(const MatrixType& matrix):Base()
     {
       init();
       compute(matrix);
@@ -681,41 +667,12 @@
     void grabMatrix(const MatrixType& matrix, ColSpMatrix& out)
     {
       // Pastix supports only lower, column-major matrices 
+      out.resize(matrix.rows(), matrix.cols());
       out.template selfadjointView<Lower>() = matrix.template selfadjointView<UpLo>();
       internal::c_to_fortran_numbering(out);
     }
 };
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<PastixBase<_MatrixType>, Rhs>
-  : solve_retval_base<PastixBase<_MatrixType>, Rhs>
-{
-  typedef PastixBase<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Rhs>
-struct sparse_solve_retval<PastixBase<_MatrixType>, Rhs>
-  : sparse_solve_retval_base<PastixBase<_MatrixType>, Rhs>
-{
-  typedef PastixBase<_MatrixType> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif
diff --git a/Eigen/src/PardisoSupport/CMakeLists.txt b/Eigen/src/PardisoSupport/CMakeLists.txt
deleted file mode 100644
index a097ab4..0000000
--- a/Eigen/src/PardisoSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_PardisoSupport_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_PardisoSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/PardisoSupport COMPONENT Devel
-  )
diff --git a/Eigen/src/PardisoSupport/PardisoSupport.h b/Eigen/src/PardisoSupport/PardisoSupport.h
index 18cd7d8..091c397 100644
--- a/Eigen/src/PardisoSupport/PardisoSupport.h
+++ b/Eigen/src/PardisoSupport/PardisoSupport.h
@@ -40,13 +40,13 @@
 
 namespace internal
 {
-  template<typename Index>
+  template<typename IndexType>
   struct pardiso_run_selector
   {
-    static Index run( _MKL_DSS_HANDLE_t pt, Index maxfct, Index mnum, Index type, Index phase, Index n, void *a,
-                      Index *ia, Index *ja, Index *perm, Index nrhs, Index *iparm, Index msglvl, void *b, void *x)
+    static IndexType run( _MKL_DSS_HANDLE_t pt, IndexType maxfct, IndexType mnum, IndexType type, IndexType phase, IndexType n, void *a,
+                      IndexType *ia, IndexType *ja, IndexType *perm, IndexType nrhs, IndexType *iparm, IndexType msglvl, void *b, void *x)
     {
-      Index error = 0;
+      IndexType error = 0;
       ::pardiso(pt, &maxfct, &mnum, &type, &phase, &n, a, ia, ja, perm, &nrhs, iparm, &msglvl, b, x, &error);
       return error;
     }
@@ -54,11 +54,11 @@
   template<>
   struct pardiso_run_selector<long long int>
   {
-    typedef long long int Index;
-    static Index run( _MKL_DSS_HANDLE_t pt, Index maxfct, Index mnum, Index type, Index phase, Index n, void *a,
-                      Index *ia, Index *ja, Index *perm, Index nrhs, Index *iparm, Index msglvl, void *b, void *x)
+    typedef long long int IndexType;
+    static IndexType run( _MKL_DSS_HANDLE_t pt, IndexType maxfct, IndexType mnum, IndexType type, IndexType phase, IndexType n, void *a,
+                      IndexType *ia, IndexType *ja, IndexType *perm, IndexType nrhs, IndexType *iparm, IndexType msglvl, void *b, void *x)
     {
-      Index error = 0;
+      IndexType error = 0;
       ::pardiso_64(pt, &maxfct, &mnum, &type, &phase, &n, a, ia, ja, perm, &nrhs, iparm, &msglvl, b, x, &error);
       return error;
     }
@@ -72,7 +72,7 @@
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -81,7 +81,7 @@
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;
+    typedef typename _MatrixType::StorageIndex StorageIndex;
   };
 
   template<typename _MatrixType, int Options>
@@ -90,35 +90,44 @@
     typedef _MatrixType MatrixType;
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef typename _MatrixType::Index Index;    
+    typedef typename _MatrixType::StorageIndex StorageIndex;    
   };
 
-}
+} // end namespace internal
 
 template<class Derived>
-class PardisoImpl
+class PardisoImpl : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
+    
     typedef internal::pardiso_traits<Derived> Traits;
   public:
+    using Base::_solve_impl;
+    
     typedef typename Traits::MatrixType MatrixType;
     typedef typename Traits::Scalar Scalar;
     typedef typename Traits::RealScalar RealScalar;
-    typedef typename Traits::Index Index;
-    typedef SparseMatrix<Scalar,RowMajor,Index> SparseMatrixType;
+    typedef typename Traits::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,RowMajor,StorageIndex> SparseMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
-    typedef Matrix<Index, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
-    typedef Matrix<Index, MatrixType::RowsAtCompileTime, 1> IntColVectorType;
-    typedef Array<Index,64,1,DontAlign> ParameterType;
+    typedef Matrix<StorageIndex, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
+    typedef Matrix<StorageIndex, MatrixType::RowsAtCompileTime, 1> IntColVectorType;
+    typedef Array<StorageIndex,64,1,DontAlign> ParameterType;
     enum {
-      ScalarIsComplex = NumTraits<Scalar>::IsComplex
+      ScalarIsComplex = NumTraits<Scalar>::IsComplex,
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
     };
 
     PardisoImpl()
     {
-      eigen_assert((sizeof(Index) >= sizeof(_INTEGER_t) && sizeof(Index) <= 8) && "Non-supported index type");
+      eigen_assert((sizeof(StorageIndex) >= sizeof(_INTEGER_t) && sizeof(StorageIndex) <= 8) && "Non-supported index type");
       m_iparm.setZero();
       m_msglvl = 0; // No output
-      m_initialized = false;
+      m_isInitialized = false;
     }
 
     ~PardisoImpl()
@@ -136,7 +145,7 @@
       */
     ComputationInfo info() const
     {
-      eigen_assert(m_initialized && "Decomposition is not initialized.");
+      eigen_assert(m_isInitialized && "Decomposition is not initialized.");
       return m_info;
     }
 
@@ -165,54 +174,18 @@
     Derived& factorize(const MatrixType& matrix);
 
     Derived& compute(const MatrixType& matrix);
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<PardisoImpl, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_initialized && "Pardiso solver is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PardisoImpl::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<PardisoImpl, Rhs>(*this, b.derived());
-    }
 
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<PardisoImpl, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_initialized && "Pardiso solver is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "PardisoImpl::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<PardisoImpl, Rhs>(*this, b.derived());
-    }
-
-    Derived& derived()
-    {
-      return *static_cast<Derived*>(this);
-    }
-    const Derived& derived() const
-    {
-      return *static_cast<const Derived*>(this);
-    }
-
-    template<typename BDerived, typename XDerived>
-    bool _solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived>& x) const;
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
 
   protected:
     void pardisoRelease()
     {
-      if(m_initialized) // Factorization ran at least once
+      if(m_isInitialized) // Factorization ran at least once
       {
-        internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, -1, m_size, 0, 0, 0, m_perm.data(), 0,
-                                                   m_iparm.data(), m_msglvl, 0, 0);
+        internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, -1, internal::convert_index<StorageIndex>(m_size),0, 0, 0, m_perm.data(), 0,
+                                                          m_iparm.data(), m_msglvl, NULL, NULL);
+        m_isInitialized = false;
       }
     }
 
@@ -221,11 +194,11 @@
       m_type = type;
       bool symmetric = std::abs(m_type) < 10;
       m_iparm[0] = 1;   // No solver default
-      m_iparm[1] = 3;   // use Metis for the ordering
-      m_iparm[2] = 1;   // Numbers of processors, value of OMP_NUM_THREADS
+      m_iparm[1] = 2;   // use Metis for the ordering
+      m_iparm[2] = 0;   // Reserved. Set to zero. (??Numbers of processors, value of OMP_NUM_THREADS??)
       m_iparm[3] = 0;   // No iterative-direct algorithm
       m_iparm[4] = 0;   // No user fill-in reducing permutation
-      m_iparm[5] = 0;   // Write solution into x
+      m_iparm[5] = 0;   // Write solution into x, b is left unchanged
       m_iparm[6] = 0;   // Not in use
       m_iparm[7] = 2;   // Max numbers of iterative refinement steps
       m_iparm[8] = 0;   // Not in use
@@ -246,13 +219,16 @@
       m_iparm[26] = 0;  // No matrix checker
       m_iparm[27] = (sizeof(RealScalar) == 4) ? 1 : 0;
       m_iparm[34] = 1;  // C indexing
-      m_iparm[59] = 1;  // Automatic switch between In-Core and Out-of-Core modes
+      m_iparm[36] = 0;  // CSR
+      m_iparm[59] = 0;  // 0 - In-Core ; 1 - Automatic switch between In-Core and Out-of-Core modes ; 2 - Out-of-Core
+      
+      memset(m_pt, 0, sizeof(m_pt));
     }
 
   protected:
     // cached data to reduce reallocation, etc.
     
-    void manageErrorCode(Index error)
+    void manageErrorCode(Index error) const
     {
       switch(error)
       {
@@ -269,16 +245,14 @@
     }
 
     mutable SparseMatrixType m_matrix;
-    ComputationInfo m_info;
-    bool m_initialized, m_analysisIsOk, m_factorizationIsOk;
-    Index m_type, m_msglvl;
+    mutable ComputationInfo m_info;
+    bool m_analysisIsOk, m_factorizationIsOk;
+    StorageIndex m_type, m_msglvl;
     mutable void *m_pt[64];
     mutable ParameterType m_iparm;
     mutable IntColVectorType m_perm;
     Index m_size;
     
-  private:
-    PardisoImpl(PardisoImpl &) {}
 };
 
 template<class Derived>
@@ -288,19 +262,17 @@
   eigen_assert(a.rows() == a.cols());
 
   pardisoRelease();
-  memset(m_pt, 0, sizeof(m_pt));
   m_perm.setZero(m_size);
   derived().getMatrix(a);
   
   Index error;
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 12, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
-
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 12, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
   manageErrorCode(error);
   m_analysisIsOk = true;
   m_factorizationIsOk = true;
-  m_initialized = true;
+  m_isInitialized = true;
   return derived();
 }
 
@@ -311,19 +283,18 @@
   eigen_assert(m_size == a.cols());
 
   pardisoRelease();
-  memset(m_pt, 0, sizeof(m_pt));
   m_perm.setZero(m_size);
   derived().getMatrix(a);
   
   Index error;
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 11, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 11, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
   
   manageErrorCode(error);
   m_analysisIsOk = true;
   m_factorizationIsOk = false;
-  m_initialized = true;
+  m_isInitialized = true;
   return derived();
 }
 
@@ -335,22 +306,25 @@
   
   derived().getMatrix(a);
 
-  Index error;  
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 22, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
+  Index error;
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 22, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), 0, m_iparm.data(), m_msglvl, NULL, NULL);
   
   manageErrorCode(error);
   m_factorizationIsOk = true;
   return derived();
 }
 
-template<class Base>
+template<class Derived>
 template<typename BDerived,typename XDerived>
-bool PardisoImpl<Base>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived>& x) const
+void PardisoImpl<Derived>::_solve_impl(const MatrixBase<BDerived> &b, MatrixBase<XDerived>& x) const
 {
   if(m_iparm[0] == 0) // Factorization was not computed
-    return false;
+  {
+    m_info = InvalidInput;
+    return;
+  }
 
   //Index n = m_matrix.rows();
   Index nrhs = Index(b.cols());
@@ -380,12 +354,12 @@
   }
   
   Index error;
-  error = internal::pardiso_run_selector<Index>::run(m_pt, 1, 1, m_type, 33, m_size,
-                                                     m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
-                                                     m_perm.data(), nrhs, m_iparm.data(), m_msglvl,
-                                                     rhs_ptr, x.derived().data());
+  error = internal::pardiso_run_selector<StorageIndex>::run(m_pt, 1, 1, m_type, 33, internal::convert_index<StorageIndex>(m_size),
+                                                            m_matrix.valuePtr(), m_matrix.outerIndexPtr(), m_matrix.innerIndexPtr(),
+                                                            m_perm.data(), internal::convert_index<StorageIndex>(nrhs), m_iparm.data(), m_msglvl,
+                                                            rhs_ptr, x.derived().data());
 
-  return error==0;
+  manageErrorCode(error);
 }
 
 
@@ -397,15 +371,20 @@
   * using the Intel MKL PARDISO library. The sparse matrix A must be squared and invertible.
   * The vectors or matrices X and B can be either dense or sparse.
   *
+  * By default, it runs in in-core mode. To enable PARDISO's out-of-core feature, set:
+  * \code solver.pardisoParameterArray()[59] = 1; \endcode
+  *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   */
 template<typename MatrixType>
 class PardisoLU : public PardisoImpl< PardisoLU<MatrixType> >
 {
   protected:
-    typedef PardisoImpl< PardisoLU<MatrixType> > Base;
+    typedef PardisoImpl<PardisoLU> Base;
     typedef typename Base::Scalar Scalar;
     typedef typename Base::RealScalar RealScalar;
     using Base::pardisoInit;
@@ -423,7 +402,7 @@
       pardisoInit(Base::ScalarIsComplex ? 13 : 11);
     }
 
-    PardisoLU(const MatrixType& matrix)
+    explicit PardisoLU(const MatrixType& matrix)
       : Base()
     {
       pardisoInit(Base::ScalarIsComplex ? 13 : 11);
@@ -433,10 +412,8 @@
     void getMatrix(const MatrixType& matrix)
     {
       m_matrix = matrix;
+      m_matrix.makeCompressed();
     }
-    
-  private:
-    PardisoLU(PardisoLU& ) {}
 };
 
 /** \ingroup PardisoSupport_Module
@@ -447,11 +424,16 @@
   * using the Intel MKL PARDISO library. The sparse matrix A must be selfajoint and positive definite.
   * The vectors or matrices X and B can be either dense or sparse.
   *
+  * By default, it runs in in-core mode. To enable PARDISO's out-of-core feature, set:
+  * \code solver.pardisoParameterArray()[59] = 1; \endcode
+  *
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam UpLo can be any bitwise combination of Upper, Lower. The default is Upper, meaning only the upper triangular part has to be used.
   *         Upper|Lower can be used to tell both triangular parts can be used as input.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLLT
   */
 template<typename MatrixType, int _UpLo>
 class PardisoLLT : public PardisoImpl< PardisoLLT<MatrixType,_UpLo> >
@@ -459,7 +441,6 @@
   protected:
     typedef PardisoImpl< PardisoLLT<MatrixType,_UpLo> > Base;
     typedef typename Base::Scalar Scalar;
-    typedef typename Base::Index Index;
     typedef typename Base::RealScalar RealScalar;
     using Base::pardisoInit;
     using Base::m_matrix;
@@ -467,9 +448,9 @@
 
   public:
 
+    typedef typename Base::StorageIndex StorageIndex;
     enum { UpLo = _UpLo };
     using Base::compute;
-    using Base::solve;
 
     PardisoLLT()
       : Base()
@@ -477,7 +458,7 @@
       pardisoInit(Base::ScalarIsComplex ? 4 : 2);
     }
 
-    PardisoLLT(const MatrixType& matrix)
+    explicit PardisoLLT(const MatrixType& matrix)
       : Base()
     {
       pardisoInit(Base::ScalarIsComplex ? 4 : 2);
@@ -489,13 +470,11 @@
     void getMatrix(const MatrixType& matrix)
     {
       // PARDISO supports only upper, row-major matrices
-      PermutationMatrix<Dynamic,Dynamic,Index> p_null;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> p_null;
       m_matrix.resize(matrix.rows(), matrix.cols());
       m_matrix.template selfadjointView<Upper>() = matrix.template selfadjointView<UpLo>().twistedBy(p_null);
+      m_matrix.makeCompressed();
     }
-    
-  private:
-    PardisoLLT(PardisoLLT& ) {}
 };
 
 /** \ingroup PardisoSupport_Module
@@ -507,12 +486,17 @@
   * For complex matrices, A can also be symmetric only, see the \a Options template parameter.
   * The vectors or matrices X and B can be either dense or sparse.
   *
+  * By default, it runs in in-core mode. To enable PARDISO's out-of-core feature, set:
+  * \code solver.pardisoParameterArray()[59] = 1; \endcode
+  *
   * \tparam MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   * \tparam Options can be any bitwise combination of Upper, Lower, and Symmetric. The default is Upper, meaning only the upper triangular part has to be used.
   *         Symmetric can be used for symmetric, non-selfadjoint complex matrices, the default being to assume a selfadjoint matrix.
   *         Upper|Lower can be used to tell both triangular parts can be used as input.
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SimplicialLDLT
   */
 template<typename MatrixType, int Options>
 class PardisoLDLT : public PardisoImpl< PardisoLDLT<MatrixType,Options> >
@@ -520,7 +504,6 @@
   protected:
     typedef PardisoImpl< PardisoLDLT<MatrixType,Options> > Base;
     typedef typename Base::Scalar Scalar;
-    typedef typename Base::Index Index;
     typedef typename Base::RealScalar RealScalar;
     using Base::pardisoInit;
     using Base::m_matrix;
@@ -528,8 +511,8 @@
 
   public:
 
+    typedef typename Base::StorageIndex StorageIndex;
     using Base::compute;
-    using Base::solve;
     enum { UpLo = Options&(Upper|Lower) };
 
     PardisoLDLT()
@@ -538,7 +521,7 @@
       pardisoInit(Base::ScalarIsComplex ? ( bool(Options&Symmetric) ? 6 : -4 ) : -2);
     }
 
-    PardisoLDLT(const MatrixType& matrix)
+    explicit PardisoLDLT(const MatrixType& matrix)
       : Base()
     {
       pardisoInit(Base::ScalarIsComplex ? ( bool(Options&Symmetric) ? 6 : -4 ) : -2);
@@ -548,45 +531,13 @@
     void getMatrix(const MatrixType& matrix)
     {
       // PARDISO supports only upper, row-major matrices
-      PermutationMatrix<Dynamic,Dynamic,Index> p_null;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> p_null;
       m_matrix.resize(matrix.rows(), matrix.cols());
       m_matrix.template selfadjointView<Upper>() = matrix.template selfadjointView<UpLo>().twistedBy(p_null);
+      m_matrix.makeCompressed();
     }
-    
-  private:
-    PardisoLDLT(PardisoLDLT& ) {}
 };
 
-namespace internal {
-  
-template<typename _Derived, typename Rhs>
-struct solve_retval<PardisoImpl<_Derived>, Rhs>
-  : solve_retval_base<PardisoImpl<_Derived>, Rhs>
-{
-  typedef PardisoImpl<_Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename Derived, typename Rhs>
-struct sparse_solve_retval<PardisoImpl<Derived>, Rhs>
-  : sparse_solve_retval_base<PardisoImpl<Derived>, Rhs>
-{
-  typedef PardisoImpl<Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_PARDISOSUPPORT_H
diff --git a/Eigen/src/QR/CMakeLists.txt b/Eigen/src/QR/CMakeLists.txt
deleted file mode 100644
index 96f43d7..0000000
--- a/Eigen/src/QR/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_QR_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_QR_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/QR COMPONENT Devel
-  )
diff --git a/Eigen/src/QR/ColPivHouseholderQR.h b/Eigen/src/QR/ColPivHouseholderQR.h
index 567eab7..a7b47d5 100644
--- a/Eigen/src/QR/ColPivHouseholderQR.h
+++ b/Eigen/src/QR/ColPivHouseholderQR.h
@@ -11,7 +11,16 @@
 #ifndef EIGEN_COLPIVOTINGHOUSEHOLDERQR_H
 #define EIGEN_COLPIVOTINGHOUSEHOLDERQR_H
 
-namespace Eigen { 
+namespace Eigen {
+
+namespace internal {
+template<typename _MatrixType> struct traits<ColPivHouseholderQR<_MatrixType> >
+ : traits<_MatrixType>
+{
+  enum { Flags = 0 };
+};
+
+} // end namespace internal
 
 /** \ingroup QR_Module
   *
@@ -19,19 +28,21 @@
   *
   * \brief Householder rank-revealing QR decomposition of a matrix with column-pivoting
   *
-  * \param MatrixType the type of the matrix of which we are computing the QR decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the QR decomposition
   *
   * This class performs a rank-revealing QR decomposition of a matrix \b A into matrices \b P, \b Q and \b R
-  * such that 
+  * such that
   * \f[
   *  \mathbf{A} \, \mathbf{P} = \mathbf{Q} \, \mathbf{R}
   * \f]
-  * by using Householder transformations. Here, \b P is a permutation matrix, \b Q a unitary matrix and \b R an 
+  * by using Householder transformations. Here, \b P is a permutation matrix, \b Q a unitary matrix and \b R an
   * upper triangular matrix.
   *
   * This decomposition performs column pivoting in order to be rank-revealing and improve
   * numerical stability. It is slower than HouseholderQR, and faster than FullPivHouseholderQR.
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::colPivHouseholderQr()
   */
 template<typename _MatrixType> class ColPivHouseholderQR
@@ -42,25 +53,25 @@
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime, Options, MaxRowsAtCompileTime, MaxRowsAtCompileTime> MatrixQType;
+    // FIXME should be int
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
     typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationType;
     typedef typename internal::plain_row_type<MatrixType, Index>::type IntRowVectorType;
     typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
     typedef typename internal::plain_row_type<MatrixType, RealScalar>::type RealRowVectorType;
     typedef HouseholderSequence<MatrixType,typename internal::remove_all<typename HCoeffsType::ConjugateReturnType>::type> HouseholderSequenceType;
-    
+    typedef typename MatrixType::PlainObject PlainObject;
+
   private:
-    
-    typedef typename PermutationType::Index PermIndexType;
-    
+
+    typedef typename PermutationType::StorageIndex PermIndexType;
+
   public:
 
     /**
@@ -75,7 +86,8 @@
         m_colsPermutation(),
         m_colsTranspositions(),
         m_temp(),
-        m_colSqNorms(),
+        m_colNormsUpdated(),
+        m_colNormsDirect(),
         m_isInitialized(false),
         m_usePrescribedThreshold(false) {}
 
@@ -91,7 +103,8 @@
         m_colsPermutation(PermIndexType(cols)),
         m_colsTranspositions(cols),
         m_temp(cols),
-        m_colSqNorms(cols),
+        m_colNormsUpdated(cols),
+        m_colNormsDirect(cols),
         m_isInitialized(false),
         m_usePrescribedThreshold(false) {}
 
@@ -99,25 +112,48 @@
       *
       * This constructor computes the QR factorization of the matrix \a matrix by calling
       * the method compute(). It is a short cut for:
-      * 
+      *
       * \code
       * ColPivHouseholderQR<MatrixType> qr(matrix.rows(), matrix.cols());
       * qr.compute(matrix);
       * \endcode
-      * 
+      *
       * \sa compute()
       */
-    ColPivHouseholderQR(const MatrixType& matrix)
+    template<typename InputType>
+    explicit ColPivHouseholderQR(const EigenBase<InputType>& matrix)
       : m_qr(matrix.rows(), matrix.cols()),
         m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
         m_colsPermutation(PermIndexType(matrix.cols())),
         m_colsTranspositions(matrix.cols()),
         m_temp(matrix.cols()),
-        m_colSqNorms(matrix.cols()),
+        m_colNormsUpdated(matrix.cols()),
+        m_colNormsDirect(matrix.cols()),
         m_isInitialized(false),
         m_usePrescribedThreshold(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a QR factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa ColPivHouseholderQR(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit ColPivHouseholderQR(EigenBase<InputType>& matrix)
+      : m_qr(matrix.derived()),
+        m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
+        m_colsPermutation(PermIndexType(matrix.cols())),
+        m_colsTranspositions(matrix.cols()),
+        m_temp(matrix.cols()),
+        m_colNormsUpdated(matrix.cols()),
+        m_colNormsDirect(matrix.cols()),
+        m_isInitialized(false),
+        m_usePrescribedThreshold(false)
+    {
+      computeInPlace();
     }
 
     /** This method finds a solution x to the equation Ax=b, where A is the matrix of which
@@ -127,9 +163,6 @@
       *
       * \returns a solution.
       *
-      * \note The case where b is a matrix is not yet implemented. Also, this
-      *       code is space inefficient.
-      *
       * \note_about_checking_solutions
       *
       * \note_about_arbitrary_choice_of_solution
@@ -138,17 +171,17 @@
       * Output: \verbinclude ColPivHouseholderQR_solve.out
       */
     template<typename Rhs>
-    inline const internal::solve_retval<ColPivHouseholderQR, Rhs>
+    inline const Solve<ColPivHouseholderQR, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
-      return internal::solve_retval<ColPivHouseholderQR, Rhs>(*this, b.derived());
+      return Solve<ColPivHouseholderQR, Rhs>(*this, b.derived());
     }
 
-    HouseholderSequenceType householderQ(void) const;
-    HouseholderSequenceType matrixQ(void) const
+    HouseholderSequenceType householderQ() const;
+    HouseholderSequenceType matrixQ() const
     {
-      return householderQ(); 
+      return householderQ();
     }
 
     /** \returns a reference to the matrix where the Householder QR decomposition is stored
@@ -158,14 +191,14 @@
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
       return m_qr;
     }
-    
-    /** \returns a reference to the matrix where the result Householder QR is stored 
-     * \warning The strict lower part of this matrix contains internal values. 
+
+    /** \returns a reference to the matrix where the result Householder QR is stored
+     * \warning The strict lower part of this matrix contains internal values.
      * Only the upper triangular part should be referenced. To get it, use
      * \code matrixR().template triangularView<Upper>() \endcode
-     * For rank-deficient matrices, use 
-     * \code 
-     * matrixR().topLeftCorner(rank(), rank()).template triangularView<Upper>() 
+     * For rank-deficient matrices, use
+     * \code
+     * matrixR().topLeftCorner(rank(), rank()).template triangularView<Upper>()
      * \endcode
      */
     const MatrixType& matrixR() const
@@ -173,8 +206,9 @@
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
       return m_qr;
     }
-    
-    ColPivHouseholderQR& compute(const MatrixType& matrix);
+
+    template<typename InputType>
+    ColPivHouseholderQR& compute(const EigenBase<InputType>& matrix);
 
     /** \returns a const reference to the column permutation matrix */
     const PermutationType& colsPermutation() const
@@ -284,20 +318,17 @@
       * \note If this matrix is not invertible, the returned matrix has undefined coefficients.
       *       Use isInvertible() to first determine whether this matrix is invertible.
       */
-    inline const
-    internal::solve_retval<ColPivHouseholderQR, typename MatrixType::IdentityReturnType>
-    inverse() const
+    inline const Inverse<ColPivHouseholderQR> inverse() const
     {
       eigen_assert(m_isInitialized && "ColPivHouseholderQR is not initialized.");
-      return internal::solve_retval<ColPivHouseholderQR,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_qr.rows(), m_qr.cols()));
+      return Inverse<ColPivHouseholderQR>(*this);
     }
 
     inline Index rows() const { return m_qr.rows(); }
     inline Index cols() const { return m_qr.cols(); }
-    
+
     /** \returns a const reference to the vector of Householder coefficients used to represent the factor \c Q.
-      * 
+      *
       * For advanced uses only.
       */
     const HCoeffsType& hCoeffs() const { return m_hCoeffs; }
@@ -370,12 +401,12 @@
       *          diagonal coefficient of R.
       */
     RealScalar maxPivot() const { return m_maxpivot; }
-    
+
     /** \brief Reports whether the QR factorization was succesful.
       *
-      * \note This function always returns \c Success. It is provided for compatibility 
+      * \note This function always returns \c Success. It is provided for compatibility
       * with other factorization routines.
-      * \returns \c Success 
+      * \returns \c Success
       */
     ComputationInfo info() const
     {
@@ -383,19 +414,30 @@
       return Success;
     }
 
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
+
   protected:
-    
+
+    friend class CompleteOrthogonalDecomposition<MatrixType>;
+
     static void check_template_parameters()
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
-    
+
+    void computeInPlace();
+
     MatrixType m_qr;
     HCoeffsType m_hCoeffs;
     PermutationType m_colsPermutation;
     IntRowVectorType m_colsTranspositions;
     RowVectorType m_temp;
-    RealRowVectorType m_colSqNorms;
+    RealRowVectorType m_colNormsUpdated;
+    RealRowVectorType m_colNormsDirect;
     bool m_isInitialized, m_usePrescribedThreshold;
     RealScalar m_prescribedThreshold, m_maxpivot;
     Index m_nonzero_pivots;
@@ -426,51 +468,57 @@
   * \sa class ColPivHouseholderQR, ColPivHouseholderQR(const MatrixType&)
   */
 template<typename MatrixType>
-ColPivHouseholderQR<MatrixType>& ColPivHouseholderQR<MatrixType>::compute(const MatrixType& matrix)
+template<typename InputType>
+ColPivHouseholderQR<MatrixType>& ColPivHouseholderQR<MatrixType>::compute(const EigenBase<InputType>& matrix)
+{
+  m_qr = matrix.derived();
+  computeInPlace();
+  return *this;
+}
+
+template<typename MatrixType>
+void ColPivHouseholderQR<MatrixType>::computeInPlace()
 {
   check_template_parameters();
-  
-  using std::abs;
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
-  Index size = matrix.diagonalSize();
-  
-  // the column permutation is stored as int indices, so just to be sure:
-  eigen_assert(cols<=NumTraits<int>::highest());
 
-  m_qr = matrix;
+  // the column permutation is stored as int indices, so just to be sure:
+  eigen_assert(m_qr.cols()<=NumTraits<int>::highest());
+
+  using std::abs;
+
+  Index rows = m_qr.rows();
+  Index cols = m_qr.cols();
+  Index size = m_qr.diagonalSize();
+
   m_hCoeffs.resize(size);
 
   m_temp.resize(cols);
 
-  m_colsTranspositions.resize(matrix.cols());
+  m_colsTranspositions.resize(m_qr.cols());
   Index number_of_transpositions = 0;
 
-  m_colSqNorms.resize(cols);
-  for(Index k = 0; k < cols; ++k)
-    m_colSqNorms.coeffRef(k) = m_qr.col(k).squaredNorm();
+  m_colNormsUpdated.resize(cols);
+  m_colNormsDirect.resize(cols);
+  for (Index k = 0; k < cols; ++k) {
+    // colNormsDirect(k) caches the most recent directly computed norm of
+    // column k.
+    m_colNormsDirect.coeffRef(k) = m_qr.col(k).norm();
+    m_colNormsUpdated.coeffRef(k) = m_colNormsDirect.coeffRef(k);
+  }
 
-  RealScalar threshold_helper = m_colSqNorms.maxCoeff() * numext::abs2(NumTraits<Scalar>::epsilon()) / RealScalar(rows);
+  RealScalar threshold_helper =  numext::abs2<RealScalar>(m_colNormsUpdated.maxCoeff() * NumTraits<RealScalar>::epsilon()) / RealScalar(rows);
+  RealScalar norm_downdate_threshold = numext::sqrt(NumTraits<RealScalar>::epsilon());
 
   m_nonzero_pivots = size; // the generic case is that in which all pivots are nonzero (invertible case)
   m_maxpivot = RealScalar(0);
 
   for(Index k = 0; k < size; ++k)
   {
-    // first, we look up in our table m_colSqNorms which column has the biggest squared norm
+    // first, we look up in our table m_colNormsUpdated which column has the biggest norm
     Index biggest_col_index;
-    RealScalar biggest_col_sq_norm = m_colSqNorms.tail(cols-k).maxCoeff(&biggest_col_index);
+    RealScalar biggest_col_sq_norm = numext::abs2(m_colNormsUpdated.tail(cols-k).maxCoeff(&biggest_col_index));
     biggest_col_index += k;
 
-    // since our table m_colSqNorms accumulates imprecision at every step, we must now recompute
-    // the actual squared norm of the selected column.
-    // Note that not doing so does result in solve() sometimes returning inf/nan values
-    // when running the unit test with 1000 repetitions.
-    biggest_col_sq_norm = m_qr.col(biggest_col_index).tail(rows-k).squaredNorm();
-
-    // we store that back into our table: it can't hurt to correct our table.
-    m_colSqNorms.coeffRef(biggest_col_index) = biggest_col_sq_norm;
-
     // Track the number of meaningful pivots but do not stop the decomposition to make
     // sure that the initial matrix is properly reproduced. See bug 941.
     if(m_nonzero_pivots==size && biggest_col_sq_norm < threshold_helper * RealScalar(rows-k))
@@ -480,7 +528,8 @@
     m_colsTranspositions.coeffRef(k) = biggest_col_index;
     if(k != biggest_col_index) {
       m_qr.col(k).swap(m_qr.col(biggest_col_index));
-      std::swap(m_colSqNorms.coeffRef(k), m_colSqNorms.coeffRef(biggest_col_index));
+      std::swap(m_colNormsUpdated.coeffRef(k), m_colNormsUpdated.coeffRef(biggest_col_index));
+      std::swap(m_colNormsDirect.coeffRef(k), m_colNormsDirect.coeffRef(biggest_col_index));
       ++number_of_transpositions;
     }
 
@@ -498,8 +547,28 @@
     m_qr.bottomRightCorner(rows-k, cols-k-1)
         .applyHouseholderOnTheLeft(m_qr.col(k).tail(rows-k-1), m_hCoeffs.coeffRef(k), &m_temp.coeffRef(k+1));
 
-    // update our table of squared norms of the columns
-    m_colSqNorms.tail(cols-k-1) -= m_qr.row(k).tail(cols-k-1).cwiseAbs2();
+    // update our table of norms of the columns
+    for (Index j = k + 1; j < cols; ++j) {
+      // The following implements the stable norm downgrade step discussed in
+      // http://www.netlib.org/lapack/lawnspdf/lawn176.pdf
+      // and used in LAPACK routines xGEQPF and xGEQP3.
+      // See lines 278-297 in http://www.netlib.org/lapack/explore-html/dc/df4/sgeqpf_8f_source.html
+      if (m_colNormsUpdated.coeffRef(j) != RealScalar(0)) {
+        RealScalar temp = abs(m_qr.coeffRef(k, j)) / m_colNormsUpdated.coeffRef(j);
+        temp = (RealScalar(1) + temp) * (RealScalar(1) - temp);
+        temp = temp <  RealScalar(0) ? RealScalar(0) : temp;
+        RealScalar temp2 = temp * numext::abs2<RealScalar>(m_colNormsUpdated.coeffRef(j) /
+                                                           m_colNormsDirect.coeffRef(j));
+        if (temp2 <= norm_downdate_threshold) {
+          // The updated norm has become too inaccurate so re-compute the column
+          // norm directly.
+          m_colNormsDirect.coeffRef(j) = m_qr.col(j).tail(rows - k - 1).norm();
+          m_colNormsUpdated.coeffRef(j) = m_colNormsDirect.coeffRef(j);
+        } else {
+          m_colNormsUpdated.coeffRef(j) *= numext::sqrt(temp);
+        }
+      }
+    }
   }
 
   m_colsPermutation.setIdentity(PermIndexType(cols));
@@ -508,46 +577,50 @@
 
   m_det_pq = (number_of_transpositions%2) ? -1 : 1;
   m_isInitialized = true;
-
-  return *this;
 }
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void ColPivHouseholderQR<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  eigen_assert(rhs.rows() == rows());
+
+  const Index nonzero_pivots = nonzeroPivots();
+
+  if(nonzero_pivots == 0)
+  {
+    dst.setZero();
+    return;
+  }
+
+  typename RhsType::PlainObject c(rhs);
+
+  // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
+  c.applyOnTheLeft(householderSequence(m_qr, m_hCoeffs)
+                    .setLength(nonzero_pivots)
+                    .transpose()
+    );
+
+  m_qr.topLeftCorner(nonzero_pivots, nonzero_pivots)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(nonzero_pivots));
+
+  for(Index i = 0; i < nonzero_pivots; ++i) dst.row(m_colsPermutation.indices().coeff(i)) = c.row(i);
+  for(Index i = nonzero_pivots; i < cols(); ++i) dst.row(m_colsPermutation.indices().coeff(i)).setZero();
+}
+#endif
+
 namespace internal {
 
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<ColPivHouseholderQR<_MatrixType>, Rhs>
-  : solve_retval_base<ColPivHouseholderQR<_MatrixType>, Rhs>
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<ColPivHouseholderQR<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename ColPivHouseholderQR<MatrixType>::Scalar>, Dense2Dense>
 {
-  EIGEN_MAKE_SOLVE_HELPERS(ColPivHouseholderQR<_MatrixType>,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
+  typedef ColPivHouseholderQR<MatrixType> QrType;
+  typedef Inverse<QrType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename QrType::Scalar> &)
   {
-    eigen_assert(rhs().rows() == dec().rows());
-
-    const Index cols = dec().cols(),
-				nonzero_pivots = dec().nonzeroPivots();
-
-    if(nonzero_pivots == 0)
-    {
-      dst.setZero();
-      return;
-    }
-
-    typename Rhs::PlainObject c(rhs());
-
-    // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
-    c.applyOnTheLeft(householderSequence(dec().matrixQR(), dec().hCoeffs())
-                     .setLength(dec().nonzeroPivots())
-		     .transpose()
-      );
-
-    dec().matrixR()
-       .topLeftCorner(nonzero_pivots, nonzero_pivots)
-       .template triangularView<Upper>()
-       .solveInPlace(c.topRows(nonzero_pivots));
-
-    for(Index i = 0; i < nonzero_pivots; ++i) dst.row(dec().colsPermutation().indices().coeff(i)) = c.row(i);
-    for(Index i = nonzero_pivots; i < cols; ++i) dst.row(dec().colsPermutation().indices().coeff(i)).setZero();
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
 
diff --git a/Eigen/src/QR/ColPivHouseholderQR_MKL.h b/Eigen/src/QR/ColPivHouseholderQR_LAPACKE.h
similarity index 64%
rename from Eigen/src/QR/ColPivHouseholderQR_MKL.h
rename to Eigen/src/QR/ColPivHouseholderQR_LAPACKE.h
index b5b1983..4e9651f 100644
--- a/Eigen/src/QR/ColPivHouseholderQR_MKL.h
+++ b/Eigen/src/QR/ColPivHouseholderQR_LAPACKE.h
@@ -25,37 +25,34 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to LAPACKe
  *    Householder QR decomposition of a matrix with column pivoting based on
  *    LAPACKE_?geqp3 function.
  ********************************************************************************
 */
 
-#ifndef EIGEN_COLPIVOTINGHOUSEHOLDERQR_MKL_H
-#define EIGEN_COLPIVOTINGHOUSEHOLDERQR_MKL_H
-
-#include "Eigen/src/Core/util/MKL_support.h"
+#ifndef EIGEN_COLPIVOTINGHOUSEHOLDERQR_LAPACKE_H
+#define EIGEN_COLPIVOTINGHOUSEHOLDERQR_LAPACKE_H
 
 namespace Eigen { 
 
-/** \internal Specialization for the data types supported by MKL */
+/** \internal Specialization for the data types supported by LAPACKe */
 
-#define EIGEN_MKL_QR_COLPIV(EIGTYPE, MKLTYPE, MKLPREFIX, EIGCOLROW, MKLCOLROW) \
-template<> inline \
+#define EIGEN_LAPACKE_QR_COLPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX, EIGCOLROW, LAPACKE_COLROW) \
+template<> template<typename InputType> inline \
 ColPivHouseholderQR<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> >& \
 ColPivHouseholderQR<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> >::compute( \
-              const Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>& matrix) \
+              const EigenBase<InputType>& matrix) \
 \
 { \
   using std::abs; \
   typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> MatrixType; \
-  typedef MatrixType::Scalar Scalar; \
   typedef MatrixType::RealScalar RealScalar; \
   Index rows = matrix.rows();\
   Index cols = matrix.cols();\
-  Index size = matrix.diagonalSize();\
 \
   m_qr = matrix;\
+  Index size = m_qr.diagonalSize();\
   m_hCoeffs.resize(size);\
 \
   m_colsTranspositions.resize(cols);\
@@ -66,34 +63,35 @@
   m_colsPermutation.resize(cols); \
   m_colsPermutation.indices().setZero(); \
 \
-  lapack_int lda = m_qr.outerStride(), i; \
-  lapack_int matrix_order = MKLCOLROW; \
-  LAPACKE_##MKLPREFIX##geqp3( matrix_order, rows, cols, (MKLTYPE*)m_qr.data(), lda, (lapack_int*)m_colsPermutation.indices().data(), (MKLTYPE*)m_hCoeffs.data()); \
+  lapack_int lda = internal::convert_index<lapack_int,Index>(m_qr.outerStride()); \
+  lapack_int matrix_order = LAPACKE_COLROW; \
+  LAPACKE_##LAPACKE_PREFIX##geqp3( matrix_order, internal::convert_index<lapack_int,Index>(rows), internal::convert_index<lapack_int,Index>(cols), \
+                              (LAPACKE_TYPE*)m_qr.data(), lda, (lapack_int*)m_colsPermutation.indices().data(), (LAPACKE_TYPE*)m_hCoeffs.data()); \
   m_isInitialized = true; \
   m_maxpivot=m_qr.diagonal().cwiseAbs().maxCoeff(); \
   m_hCoeffs.adjointInPlace(); \
   RealScalar premultiplied_threshold = abs(m_maxpivot) * threshold(); \
   lapack_int *perm = m_colsPermutation.indices().data(); \
-  for(i=0;i<size;i++) { \
+  for(Index i=0;i<size;i++) { \
     m_nonzero_pivots += (abs(m_qr.coeff(i,i)) > premultiplied_threshold);\
   } \
-  for(i=0;i<cols;i++) perm[i]--;\
+  for(Index i=0;i<cols;i++) perm[i]--;\
 \
   /*m_det_pq = (number_of_transpositions%2) ? -1 : 1;  // TODO: It's not needed now; fix upon availability in Eigen */ \
 \
   return *this; \
 }
 
-EIGEN_MKL_QR_COLPIV(double,   double,        d, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_QR_COLPIV(float,    float,         s, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_QR_COLPIV(dcomplex, MKL_Complex16, z, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_QR_COLPIV(scomplex, MKL_Complex8,  c, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(double,   double,        d, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(float,    float,         s, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(dcomplex, lapack_complex_double, z, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(scomplex, lapack_complex_float,  c, ColMajor, LAPACK_COL_MAJOR)
 
-EIGEN_MKL_QR_COLPIV(double,   double,        d, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_QR_COLPIV(float,    float,         s, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_QR_COLPIV(dcomplex, MKL_Complex16, z, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_QR_COLPIV(scomplex, MKL_Complex8,  c, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(double,   double,        d, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(float,    float,         s, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(dcomplex, lapack_complex_double, z, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_QR_COLPIV(scomplex, lapack_complex_float,  c, RowMajor, LAPACK_ROW_MAJOR)
 
 } // end namespace Eigen
 
-#endif // EIGEN_COLPIVOTINGHOUSEHOLDERQR_MKL_H
+#endif // EIGEN_COLPIVOTINGHOUSEHOLDERQR_LAPACKE_H
diff --git a/Eigen/src/QR/CompleteOrthogonalDecomposition.h b/Eigen/src/QR/CompleteOrthogonalDecomposition.h
new file mode 100644
index 0000000..34c637b
--- /dev/null
+++ b/Eigen/src/QR/CompleteOrthogonalDecomposition.h
@@ -0,0 +1,562 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2016 Rasmus Munk Larsen <rmlarsen@google.com>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_COMPLETEORTHOGONALDECOMPOSITION_H
+#define EIGEN_COMPLETEORTHOGONALDECOMPOSITION_H
+
+namespace Eigen {
+
+namespace internal {
+template <typename _MatrixType>
+struct traits<CompleteOrthogonalDecomposition<_MatrixType> >
+    : traits<_MatrixType> {
+  enum { Flags = 0 };
+};
+
+}  // end namespace internal
+
+/** \ingroup QR_Module
+  *
+  * \class CompleteOrthogonalDecomposition
+  *
+  * \brief Complete orthogonal decomposition (COD) of a matrix.
+  *
+  * \param MatrixType the type of the matrix of which we are computing the COD.
+  *
+  * This class performs a rank-revealing complete orthogonal decomposition of a
+  * matrix  \b A into matrices \b P, \b Q, \b T, and \b Z such that
+  * \f[
+  *  \mathbf{A} \, \mathbf{P} = \mathbf{Q} \,
+  *                     \begin{bmatrix} \mathbf{T} &  \mathbf{0} \\
+  *                                     \mathbf{0} & \mathbf{0} \end{bmatrix} \, \mathbf{Z}
+  * \f]
+  * by using Householder transformations. Here, \b P is a permutation matrix,
+  * \b Q and \b Z are unitary matrices and \b T an upper triangular matrix of
+  * size rank-by-rank. \b A may be rank deficient.
+  *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
+  * \sa MatrixBase::completeOrthogonalDecomposition()
+  */
+template <typename _MatrixType>
+class CompleteOrthogonalDecomposition {
+ public:
+  typedef _MatrixType MatrixType;
+  enum {
+    RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+  };
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef typename MatrixType::StorageIndex StorageIndex;
+  typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
+  typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime>
+      PermutationType;
+  typedef typename internal::plain_row_type<MatrixType, Index>::type
+      IntRowVectorType;
+  typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
+  typedef typename internal::plain_row_type<MatrixType, RealScalar>::type
+      RealRowVectorType;
+  typedef HouseholderSequence<
+      MatrixType, typename internal::remove_all<
+                      typename HCoeffsType::ConjugateReturnType>::type>
+      HouseholderSequenceType;
+  typedef typename MatrixType::PlainObject PlainObject;
+
+ private:
+  typedef typename PermutationType::Index PermIndexType;
+
+ public:
+  /**
+   * \brief Default Constructor.
+   *
+   * The default constructor is useful in cases in which the user intends to
+   * perform decompositions via
+   * \c CompleteOrthogonalDecomposition::compute(const* MatrixType&).
+   */
+  CompleteOrthogonalDecomposition() : m_cpqr(), m_zCoeffs(), m_temp() {}
+
+  /** \brief Default Constructor with memory preallocation
+   *
+   * Like the default constructor but with preallocation of the internal data
+   * according to the specified problem \a size.
+   * \sa CompleteOrthogonalDecomposition()
+   */
+  CompleteOrthogonalDecomposition(Index rows, Index cols)
+      : m_cpqr(rows, cols), m_zCoeffs((std::min)(rows, cols)), m_temp(cols) {}
+
+  /** \brief Constructs a complete orthogonal decomposition from a given
+   * matrix.
+   *
+   * This constructor computes the complete orthogonal decomposition of the
+   * matrix \a matrix by calling the method compute(). The default
+   * threshold for rank determination will be used. It is a short cut for:
+   *
+   * \code
+   * CompleteOrthogonalDecomposition<MatrixType> cod(matrix.rows(),
+   *                                                 matrix.cols());
+   * cod.setThreshold(Default);
+   * cod.compute(matrix);
+   * \endcode
+   *
+   * \sa compute()
+   */
+  template <typename InputType>
+  explicit CompleteOrthogonalDecomposition(const EigenBase<InputType>& matrix)
+      : m_cpqr(matrix.rows(), matrix.cols()),
+        m_zCoeffs((std::min)(matrix.rows(), matrix.cols())),
+        m_temp(matrix.cols())
+  {
+    compute(matrix.derived());
+  }
+
+  /** \brief Constructs a complete orthogonal decomposition from a given matrix
+    *
+    * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+    *
+    * \sa CompleteOrthogonalDecomposition(const EigenBase&)
+    */
+  template<typename InputType>
+  explicit CompleteOrthogonalDecomposition(EigenBase<InputType>& matrix)
+    : m_cpqr(matrix.derived()),
+      m_zCoeffs((std::min)(matrix.rows(), matrix.cols())),
+      m_temp(matrix.cols())
+  {
+    computeInPlace();
+  }
+
+
+  /** This method computes the minimum-norm solution X to a least squares
+   * problem \f[\mathrm{minimize} \|A X - B\|, \f] where \b A is the matrix of
+   * which \c *this is the complete orthogonal decomposition.
+   *
+   * \param b the right-hand sides of the problem to solve.
+   *
+   * \returns a solution.
+   *
+   */
+  template <typename Rhs>
+  inline const Solve<CompleteOrthogonalDecomposition, Rhs> solve(
+      const MatrixBase<Rhs>& b) const {
+    eigen_assert(m_cpqr.m_isInitialized &&
+                 "CompleteOrthogonalDecomposition is not initialized.");
+    return Solve<CompleteOrthogonalDecomposition, Rhs>(*this, b.derived());
+  }
+
+  HouseholderSequenceType householderQ(void) const;
+  HouseholderSequenceType matrixQ(void) const { return m_cpqr.householderQ(); }
+
+  /** \returns the matrix \b Z.
+   */
+  MatrixType matrixZ() const {
+    MatrixType Z = MatrixType::Identity(m_cpqr.cols(), m_cpqr.cols());
+    applyZAdjointOnTheLeftInPlace(Z);
+    return Z.adjoint();
+  }
+
+  /** \returns a reference to the matrix where the complete orthogonal
+   * decomposition is stored
+   */
+  const MatrixType& matrixQTZ() const { return m_cpqr.matrixQR(); }
+
+  /** \returns a reference to the matrix where the complete orthogonal
+   * decomposition is stored.
+   * \warning The strict lower part and \code cols() - rank() \endcode right
+   * columns of this matrix contains internal values.
+   * Only the upper triangular part should be referenced. To get it, use
+   * \code matrixT().template triangularView<Upper>() \endcode
+   * For rank-deficient matrices, use
+   * \code
+   * matrixR().topLeftCorner(rank(), rank()).template triangularView<Upper>()
+   * \endcode
+   */
+  const MatrixType& matrixT() const { return m_cpqr.matrixQR(); }
+
+  template <typename InputType>
+  CompleteOrthogonalDecomposition& compute(const EigenBase<InputType>& matrix) {
+    // Compute the column pivoted QR factorization A P = Q R.
+    m_cpqr.compute(matrix);
+    computeInPlace();
+    return *this;
+  }
+
+  /** \returns a const reference to the column permutation matrix */
+  const PermutationType& colsPermutation() const {
+    return m_cpqr.colsPermutation();
+  }
+
+  /** \returns the absolute value of the determinant of the matrix of which
+   * *this is the complete orthogonal decomposition. It has only linear
+   * complexity (that is, O(n) where n is the dimension of the square matrix)
+   * as the complete orthogonal decomposition has already been computed.
+   *
+   * \note This is only for square matrices.
+   *
+   * \warning a determinant can be very big or small, so for matrices
+   * of large enough dimension, there is a risk of overflow/underflow.
+   * One way to work around that is to use logAbsDeterminant() instead.
+   *
+   * \sa logAbsDeterminant(), MatrixBase::determinant()
+   */
+  typename MatrixType::RealScalar absDeterminant() const;
+
+  /** \returns the natural log of the absolute value of the determinant of the
+   * matrix of which *this is the complete orthogonal decomposition. It has
+   * only linear complexity (that is, O(n) where n is the dimension of the
+   * square matrix) as the complete orthogonal decomposition has already been
+   * computed.
+   *
+   * \note This is only for square matrices.
+   *
+   * \note This method is useful to work around the risk of overflow/underflow
+   * that's inherent to determinant computation.
+   *
+   * \sa absDeterminant(), MatrixBase::determinant()
+   */
+  typename MatrixType::RealScalar logAbsDeterminant() const;
+
+  /** \returns the rank of the matrix of which *this is the complete orthogonal
+   * decomposition.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline Index rank() const { return m_cpqr.rank(); }
+
+  /** \returns the dimension of the kernel of the matrix of which *this is the
+   * complete orthogonal decomposition.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline Index dimensionOfKernel() const { return m_cpqr.dimensionOfKernel(); }
+
+  /** \returns true if the matrix of which *this is the decomposition represents
+   * an injective linear map, i.e. has trivial kernel; false otherwise.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline bool isInjective() const { return m_cpqr.isInjective(); }
+
+  /** \returns true if the matrix of which *this is the decomposition represents
+   * a surjective linear map; false otherwise.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline bool isSurjective() const { return m_cpqr.isSurjective(); }
+
+  /** \returns true if the matrix of which *this is the complete orthogonal
+   * decomposition is invertible.
+   *
+   * \note This method has to determine which pivots should be considered
+   * nonzero. For that, it uses the threshold value that you can control by
+   * calling setThreshold(const RealScalar&).
+   */
+  inline bool isInvertible() const { return m_cpqr.isInvertible(); }
+
+  /** \returns the pseudo-inverse of the matrix of which *this is the complete
+   * orthogonal decomposition.
+   * \warning: Do not compute \c this->pseudoInverse()*rhs to solve a linear systems.
+   * It is more efficient and numerically stable to call \c this->solve(rhs).
+   */
+  inline const Inverse<CompleteOrthogonalDecomposition> pseudoInverse() const
+  {
+    return Inverse<CompleteOrthogonalDecomposition>(*this);
+  }
+
+  inline Index rows() const { return m_cpqr.rows(); }
+  inline Index cols() const { return m_cpqr.cols(); }
+
+  /** \returns a const reference to the vector of Householder coefficients used
+   * to represent the factor \c Q.
+   *
+   * For advanced uses only.
+   */
+  inline const HCoeffsType& hCoeffs() const { return m_cpqr.hCoeffs(); }
+
+  /** \returns a const reference to the vector of Householder coefficients
+   * used to represent the factor \c Z.
+   *
+   * For advanced uses only.
+   */
+  const HCoeffsType& zCoeffs() const { return m_zCoeffs; }
+
+  /** Allows to prescribe a threshold to be used by certain methods, such as
+   * rank(), who need to determine when pivots are to be considered nonzero.
+   * Most be called before calling compute().
+   *
+   * When it needs to get the threshold value, Eigen calls threshold(). By
+   * default, this uses a formula to automatically determine a reasonable
+   * threshold. Once you have called the present method
+   * setThreshold(const RealScalar&), your value is used instead.
+   *
+   * \param threshold The new value to use as the threshold.
+   *
+   * A pivot will be considered nonzero if its absolute value is strictly
+   * greater than
+   *  \f$ \vert pivot \vert \leqslant threshold \times \vert maxpivot \vert \f$
+   * where maxpivot is the biggest pivot.
+   *
+   * If you want to come back to the default behavior, call
+   * setThreshold(Default_t)
+   */
+  CompleteOrthogonalDecomposition& setThreshold(const RealScalar& threshold) {
+    m_cpqr.setThreshold(threshold);
+    return *this;
+  }
+
+  /** Allows to come back to the default behavior, letting Eigen use its default
+   * formula for determining the threshold.
+   *
+   * You should pass the special object Eigen::Default as parameter here.
+   * \code qr.setThreshold(Eigen::Default); \endcode
+   *
+   * See the documentation of setThreshold(const RealScalar&).
+   */
+  CompleteOrthogonalDecomposition& setThreshold(Default_t) {
+    m_cpqr.setThreshold(Default);
+    return *this;
+  }
+
+  /** Returns the threshold that will be used by certain methods such as rank().
+   *
+   * See the documentation of setThreshold(const RealScalar&).
+   */
+  RealScalar threshold() const { return m_cpqr.threshold(); }
+
+  /** \returns the number of nonzero pivots in the complete orthogonal
+   * decomposition. Here nonzero is meant in the exact sense, not in a
+   * fuzzy sense. So that notion isn't really intrinsically interesting,
+   * but it is still useful when implementing algorithms.
+   *
+   * \sa rank()
+   */
+  inline Index nonzeroPivots() const { return m_cpqr.nonzeroPivots(); }
+
+  /** \returns the absolute value of the biggest pivot, i.e. the biggest
+   *          diagonal coefficient of R.
+   */
+  inline RealScalar maxPivot() const { return m_cpqr.maxPivot(); }
+
+  /** \brief Reports whether the complete orthogonal decomposition was
+   * succesful.
+   *
+   * \note This function always returns \c Success. It is provided for
+   * compatibility
+   * with other factorization routines.
+   * \returns \c Success
+   */
+  ComputationInfo info() const {
+    eigen_assert(m_cpqr.m_isInitialized && "Decomposition is not initialized.");
+    return Success;
+  }
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  template <typename RhsType, typename DstType>
+  EIGEN_DEVICE_FUNC void _solve_impl(const RhsType& rhs, DstType& dst) const;
+#endif
+
+ protected:
+  static void check_template_parameters() {
+    EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
+  }
+
+  void computeInPlace();
+
+  /** Overwrites \b rhs with \f$ \mathbf{Z}^* * \mathbf{rhs} \f$.
+   */
+  template <typename Rhs>
+  void applyZAdjointOnTheLeftInPlace(Rhs& rhs) const;
+
+  ColPivHouseholderQR<MatrixType> m_cpqr;
+  HCoeffsType m_zCoeffs;
+  RowVectorType m_temp;
+};
+
+template <typename MatrixType>
+typename MatrixType::RealScalar
+CompleteOrthogonalDecomposition<MatrixType>::absDeterminant() const {
+  return m_cpqr.absDeterminant();
+}
+
+template <typename MatrixType>
+typename MatrixType::RealScalar
+CompleteOrthogonalDecomposition<MatrixType>::logAbsDeterminant() const {
+  return m_cpqr.logAbsDeterminant();
+}
+
+/** Performs the complete orthogonal decomposition of the given matrix \a
+ * matrix. The result of the factorization is stored into \c *this, and a
+ * reference to \c *this is returned.
+ *
+ * \sa class CompleteOrthogonalDecomposition,
+ * CompleteOrthogonalDecomposition(const MatrixType&)
+ */
+template <typename MatrixType>
+void CompleteOrthogonalDecomposition<MatrixType>::computeInPlace()
+{
+  check_template_parameters();
+
+  // the column permutation is stored as int indices, so just to be sure:
+  eigen_assert(m_cpqr.cols() <= NumTraits<int>::highest());
+
+  const Index rank = m_cpqr.rank();
+  const Index cols = m_cpqr.cols();
+  const Index rows = m_cpqr.rows();
+  m_zCoeffs.resize((std::min)(rows, cols));
+  m_temp.resize(cols);
+
+  if (rank < cols) {
+    // We have reduced the (permuted) matrix to the form
+    //   [R11 R12]
+    //   [ 0  R22]
+    // where R11 is r-by-r (r = rank) upper triangular, R12 is
+    // r-by-(n-r), and R22 is empty or the norm of R22 is negligible.
+    // We now compute the complete orthogonal decomposition by applying
+    // Householder transformations from the right to the upper trapezoidal
+    // matrix X = [R11 R12] to zero out R12 and obtain the factorization
+    // [R11 R12] = [T11 0] * Z, where T11 is r-by-r upper triangular and
+    // Z = Z(0) * Z(1) ... Z(r-1) is an n-by-n orthogonal matrix.
+    // We store the data representing Z in R12 and m_zCoeffs.
+    for (Index k = rank - 1; k >= 0; --k) {
+      if (k != rank - 1) {
+        // Given the API for Householder reflectors, it is more convenient if
+        // we swap the leading parts of columns k and r-1 (zero-based) to form
+        // the matrix X_k = [X(0:k, k), X(0:k, r:n)]
+        m_cpqr.m_qr.col(k).head(k + 1).swap(
+            m_cpqr.m_qr.col(rank - 1).head(k + 1));
+      }
+      // Construct Householder reflector Z(k) to zero out the last row of X_k,
+      // i.e. choose Z(k) such that
+      // [X(k, k), X(k, r:n)] * Z(k) = [beta, 0, .., 0].
+      RealScalar beta;
+      m_cpqr.m_qr.row(k)
+          .tail(cols - rank + 1)
+          .makeHouseholderInPlace(m_zCoeffs(k), beta);
+      m_cpqr.m_qr(k, rank - 1) = beta;
+      if (k > 0) {
+        // Apply Z(k) to the first k rows of X_k
+        m_cpqr.m_qr.topRightCorner(k, cols - rank + 1)
+            .applyHouseholderOnTheRight(
+                m_cpqr.m_qr.row(k).tail(cols - rank).transpose(), m_zCoeffs(k),
+                &m_temp(0));
+      }
+      if (k != rank - 1) {
+        // Swap X(0:k,k) back to its proper location.
+        m_cpqr.m_qr.col(k).head(k + 1).swap(
+            m_cpqr.m_qr.col(rank - 1).head(k + 1));
+      }
+    }
+  }
+}
+
+template <typename MatrixType>
+template <typename Rhs>
+void CompleteOrthogonalDecomposition<MatrixType>::applyZAdjointOnTheLeftInPlace(
+    Rhs& rhs) const {
+  const Index cols = this->cols();
+  const Index nrhs = rhs.cols();
+  const Index rank = this->rank();
+  Matrix<typename MatrixType::Scalar, Dynamic, 1> temp((std::max)(cols, nrhs));
+  for (Index k = 0; k < rank; ++k) {
+    if (k != rank - 1) {
+      rhs.row(k).swap(rhs.row(rank - 1));
+    }
+    rhs.middleRows(rank - 1, cols - rank + 1)
+        .applyHouseholderOnTheLeft(
+            matrixQTZ().row(k).tail(cols - rank).adjoint(), zCoeffs()(k),
+            &temp(0));
+    if (k != rank - 1) {
+      rhs.row(k).swap(rhs.row(rank - 1));
+    }
+  }
+}
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template <typename _MatrixType>
+template <typename RhsType, typename DstType>
+void CompleteOrthogonalDecomposition<_MatrixType>::_solve_impl(
+    const RhsType& rhs, DstType& dst) const {
+  eigen_assert(rhs.rows() == this->rows());
+
+  const Index rank = this->rank();
+  if (rank == 0) {
+    dst.setZero();
+    return;
+  }
+
+  // Compute c = Q^* * rhs
+  // Note that the matrix Q = H_0^* H_1^*... so its inverse is
+  // Q^* = (H_0 H_1 ...)^T
+  typename RhsType::PlainObject c(rhs);
+  c.applyOnTheLeft(
+      householderSequence(matrixQTZ(), hCoeffs()).setLength(rank).transpose());
+
+  // Solve T z = c(1:rank, :)
+  dst.topRows(rank) = matrixT()
+                          .topLeftCorner(rank, rank)
+                          .template triangularView<Upper>()
+                          .solve(c.topRows(rank));
+
+  const Index cols = this->cols();
+  if (rank < cols) {
+    // Compute y = Z^* * [ z ]
+    //                   [ 0 ]
+    dst.bottomRows(cols - rank).setZero();
+    applyZAdjointOnTheLeftInPlace(dst);
+  }
+
+  // Undo permutation to get x = P^{-1} * y.
+  dst = colsPermutation() * dst;
+}
+#endif
+
+namespace internal {
+
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<CompleteOrthogonalDecomposition<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename CompleteOrthogonalDecomposition<MatrixType>::Scalar>, Dense2Dense>
+{
+  typedef CompleteOrthogonalDecomposition<MatrixType> CodType;
+  typedef Inverse<CodType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename CodType::Scalar> &)
+  {
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.rows()));
+  }
+};
+
+} // end namespace internal
+
+/** \returns the matrix Q as a sequence of householder transformations */
+template <typename MatrixType>
+typename CompleteOrthogonalDecomposition<MatrixType>::HouseholderSequenceType
+CompleteOrthogonalDecomposition<MatrixType>::householderQ() const {
+  return m_cpqr.householderQ();
+}
+
+/** \return the complete orthogonal decomposition of \c *this.
+  *
+  * \sa class CompleteOrthogonalDecomposition
+  */
+template <typename Derived>
+const CompleteOrthogonalDecomposition<typename MatrixBase<Derived>::PlainObject>
+MatrixBase<Derived>::completeOrthogonalDecomposition() const {
+  return CompleteOrthogonalDecomposition<PlainObject>(eval());
+}
+
+}  // end namespace Eigen
+
+#endif  // EIGEN_COMPLETEORTHOGONALDECOMPOSITION_H
diff --git a/Eigen/src/QR/FullPivHouseholderQR.h b/Eigen/src/QR/FullPivHouseholderQR.h
index 0b39966..e489bdd 100644
--- a/Eigen/src/QR/FullPivHouseholderQR.h
+++ b/Eigen/src/QR/FullPivHouseholderQR.h
@@ -15,6 +15,12 @@
 
 namespace internal {
 
+template<typename _MatrixType> struct traits<FullPivHouseholderQR<_MatrixType> >
+ : traits<_MatrixType>
+{
+  enum { Flags = 0 };
+};
+
 template<typename MatrixType> struct FullPivHouseholderQRMatrixQReturnType;
 
 template<typename MatrixType>
@@ -23,7 +29,7 @@
   typedef typename MatrixType::PlainObject ReturnType;
 };
 
-}
+} // end namespace internal
 
 /** \ingroup QR_Module
   *
@@ -31,19 +37,21 @@
   *
   * \brief Householder rank-revealing QR decomposition of a matrix with full pivoting
   *
-  * \param MatrixType the type of the matrix of which we are computing the QR decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the QR decomposition
   *
-  * This class performs a rank-revealing QR decomposition of a matrix \b A into matrices \b P, \b Q and \b R
+  * This class performs a rank-revealing QR decomposition of a matrix \b A into matrices \b P, \b P', \b Q and \b R
   * such that 
   * \f[
-  *  \mathbf{A} \, \mathbf{P} = \mathbf{Q} \, \mathbf{R}
+  *  \mathbf{P} \, \mathbf{A} \, \mathbf{P}' = \mathbf{Q} \, \mathbf{R}
   * \f]
-  * by using Householder transformations. Here, \b P is a permutation matrix, \b Q a unitary matrix and \b R an 
-  * upper triangular matrix.
+  * by using Householder transformations. Here, \b P and \b P' are permutation matrices, \b Q a unitary matrix 
+  * and \b R an upper triangular matrix.
   *
   * This decomposition performs a very prudent full pivoting in order to be rank-revealing and achieve optimal
   * numerical stability. The trade-off is that it is slower than HouseholderQR and ColPivHouseholderQR.
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  * 
   * \sa MatrixBase::fullPivHouseholderQr()
   */
 template<typename _MatrixType> class FullPivHouseholderQR
@@ -54,21 +62,22 @@
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    // FIXME should be int
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef internal::FullPivHouseholderQRMatrixQReturnType<MatrixType> MatrixQReturnType;
     typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
-    typedef Matrix<Index, 1,
+    typedef Matrix<StorageIndex, 1,
                    EIGEN_SIZE_MIN_PREFER_DYNAMIC(ColsAtCompileTime,RowsAtCompileTime), RowMajor, 1,
                    EIGEN_SIZE_MIN_PREFER_FIXED(MaxColsAtCompileTime,MaxRowsAtCompileTime)> IntDiagSizeVectorType;
     typedef PermutationMatrix<ColsAtCompileTime, MaxColsAtCompileTime> PermutationType;
     typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
     typedef typename internal::plain_col_type<MatrixType>::type ColVectorType;
+    typedef typename MatrixType::PlainObject PlainObject;
 
     /** \brief Default Constructor.
       *
@@ -113,7 +122,8 @@
       * 
       * \sa compute()
       */
-    FullPivHouseholderQR(const MatrixType& matrix)
+    template<typename InputType>
+    explicit FullPivHouseholderQR(const EigenBase<InputType>& matrix)
       : m_qr(matrix.rows(), matrix.cols()),
         m_hCoeffs((std::min)(matrix.rows(), matrix.cols())),
         m_rows_transpositions((std::min)(matrix.rows(), matrix.cols())),
@@ -123,7 +133,27 @@
         m_isInitialized(false),
         m_usePrescribedThreshold(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+    /** \brief Constructs a QR factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref.
+      *
+      * \sa FullPivHouseholderQR(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit FullPivHouseholderQR(EigenBase<InputType>& matrix)
+      : m_qr(matrix.derived()),
+        m_hCoeffs((std::min)(matrix.rows(), matrix.cols())),
+        m_rows_transpositions((std::min)(matrix.rows(), matrix.cols())),
+        m_cols_transpositions((std::min)(matrix.rows(), matrix.cols())),
+        m_cols_permutation(matrix.cols()),
+        m_temp(matrix.cols()),
+        m_isInitialized(false),
+        m_usePrescribedThreshold(false)
+    {
+      computeInPlace();
     }
 
     /** This method finds a solution x to the equation Ax=b, where A is the matrix of which
@@ -134,9 +164,6 @@
       * \returns the exact or least-square solution if the rank is greater or equal to the number of columns of A,
       * and an arbitrary solution otherwise.
       *
-      * \note The case where b is a matrix is not yet implemented. Also, this
-      *       code is space inefficient.
-      *
       * \note_about_checking_solutions
       *
       * \note_about_arbitrary_choice_of_solution
@@ -145,11 +172,11 @@
       * Output: \verbinclude FullPivHouseholderQR_solve.out
       */
     template<typename Rhs>
-    inline const internal::solve_retval<FullPivHouseholderQR, Rhs>
+    inline const Solve<FullPivHouseholderQR, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "FullPivHouseholderQR is not initialized.");
-      return internal::solve_retval<FullPivHouseholderQR, Rhs>(*this, b.derived());
+      return Solve<FullPivHouseholderQR, Rhs>(*this, b.derived());
     }
 
     /** \returns Expression object representing the matrix Q
@@ -164,7 +191,8 @@
       return m_qr;
     }
 
-    FullPivHouseholderQR& compute(const MatrixType& matrix);
+    template<typename InputType>
+    FullPivHouseholderQR& compute(const EigenBase<InputType>& matrix);
 
     /** \returns a const reference to the column permutation matrix */
     const PermutationType& colsPermutation() const
@@ -280,13 +308,11 @@
       *
       * \note If this matrix is not invertible, the returned matrix has undefined coefficients.
       *       Use isInvertible() to first determine whether this matrix is invertible.
-      */    inline const
-    internal::solve_retval<FullPivHouseholderQR, typename MatrixType::IdentityReturnType>
-    inverse() const
+      */
+    inline const Inverse<FullPivHouseholderQR> inverse() const
     {
       eigen_assert(m_isInitialized && "FullPivHouseholderQR is not initialized.");
-      return internal::solve_retval<FullPivHouseholderQR,typename MatrixType::IdentityReturnType>
-               (*this, MatrixType::Identity(m_qr.rows(), m_qr.cols()));
+      return Inverse<FullPivHouseholderQR>(*this);
     }
 
     inline Index rows() const { return m_qr.rows(); }
@@ -366,6 +392,12 @@
       *          diagonal coefficient of U.
       */
     RealScalar maxPivot() const { return m_maxpivot; }
+    
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
     
@@ -374,6 +406,8 @@
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
     
+    void computeInPlace();
+    
     MatrixType m_qr;
     HCoeffsType m_hCoeffs;
     IntDiagSizeVectorType m_rows_transpositions;
@@ -411,16 +445,25 @@
   * \sa class FullPivHouseholderQR, FullPivHouseholderQR(const MatrixType&)
   */
 template<typename MatrixType>
-FullPivHouseholderQR<MatrixType>& FullPivHouseholderQR<MatrixType>::compute(const MatrixType& matrix)
+template<typename InputType>
+FullPivHouseholderQR<MatrixType>& FullPivHouseholderQR<MatrixType>::compute(const EigenBase<InputType>& matrix)
+{
+  m_qr = matrix.derived();
+  computeInPlace();
+  return *this;
+}
+
+template<typename MatrixType>
+void FullPivHouseholderQR<MatrixType>::computeInPlace()
 {
   check_template_parameters();
-  
+
   using std::abs;
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
+  Index rows = m_qr.rows();
+  Index cols = m_qr.cols();
   Index size = (std::min)(rows,cols);
 
-  m_qr = matrix;
+  
   m_hCoeffs.resize(size);
 
   m_temp.resize(cols);
@@ -439,13 +482,15 @@
   for (Index k = 0; k < size; ++k)
   {
     Index row_of_biggest_in_corner, col_of_biggest_in_corner;
-    RealScalar biggest_in_corner;
+    typedef internal::scalar_score_coeff_op<Scalar> Scoring;
+    typedef typename Scoring::result_type Score;
 
-    biggest_in_corner = m_qr.bottomRightCorner(rows-k, cols-k)
-                            .cwiseAbs()
-                            .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
+    Score score = m_qr.bottomRightCorner(rows-k, cols-k)
+                      .unaryExpr(Scoring())
+                      .maxCoeff(&row_of_biggest_in_corner, &col_of_biggest_in_corner);
     row_of_biggest_in_corner += k;
     col_of_biggest_in_corner += k;
+    RealScalar biggest_in_corner = internal::abs_knowing_score<Scalar>()(m_qr(row_of_biggest_in_corner, col_of_biggest_in_corner), score);
     if(k==0) biggest = biggest_in_corner;
 
     // if the corner is negligible, then we have less than full rank, and we can finish early
@@ -489,50 +534,55 @@
 
   m_det_pq = (number_of_transpositions%2) ? -1 : 1;
   m_isInitialized = true;
-
-  return *this;
 }
 
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<FullPivHouseholderQR<_MatrixType>, Rhs>
-  : solve_retval_base<FullPivHouseholderQR<_MatrixType>, Rhs>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void FullPivHouseholderQR<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
 {
-  EIGEN_MAKE_SOLVE_HELPERS(FullPivHouseholderQR<_MatrixType>,Rhs)
+  eigen_assert(rhs.rows() == rows());
+  const Index l_rank = rank();
 
-  template<typename Dest> void evalTo(Dest& dst) const
+  // FIXME introduce nonzeroPivots() and use it here. and more generally,
+  // make the same improvements in this dec as in FullPivLU.
+  if(l_rank==0)
   {
-    const Index rows = dec().rows(), cols = dec().cols();
-    eigen_assert(rhs().rows() == rows);
+    dst.setZero();
+    return;
+  }
 
-    // FIXME introduce nonzeroPivots() and use it here. and more generally,
-    // make the same improvements in this dec as in FullPivLU.
-    if(dec().rank()==0)
-    {
-      dst.setZero();
-      return;
-    }
+  typename RhsType::PlainObject c(rhs);
 
-    typename Rhs::PlainObject c(rhs());
+  Matrix<Scalar,1,RhsType::ColsAtCompileTime> temp(rhs.cols());
+  for (Index k = 0; k < l_rank; ++k)
+  {
+    Index remainingSize = rows()-k;
+    c.row(k).swap(c.row(m_rows_transpositions.coeff(k)));
+    c.bottomRightCorner(remainingSize, rhs.cols())
+      .applyHouseholderOnTheLeft(m_qr.col(k).tail(remainingSize-1),
+                               m_hCoeffs.coeff(k), &temp.coeffRef(0));
+  }
 
-    Matrix<Scalar,1,Rhs::ColsAtCompileTime> temp(rhs().cols());
-    for (Index k = 0; k < dec().rank(); ++k)
-    {
-      Index remainingSize = rows-k;
-      c.row(k).swap(c.row(dec().rowsTranspositions().coeff(k)));
-      c.bottomRightCorner(remainingSize, rhs().cols())
-       .applyHouseholderOnTheLeft(dec().matrixQR().col(k).tail(remainingSize-1),
-                                  dec().hCoeffs().coeff(k), &temp.coeffRef(0));
-    }
+  m_qr.topLeftCorner(l_rank, l_rank)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(l_rank));
 
-    dec().matrixQR()
-       .topLeftCorner(dec().rank(), dec().rank())
-       .template triangularView<Upper>()
-       .solveInPlace(c.topRows(dec().rank()));
+  for(Index i = 0; i < l_rank; ++i) dst.row(m_cols_permutation.indices().coeff(i)) = c.row(i);
+  for(Index i = l_rank; i < cols(); ++i) dst.row(m_cols_permutation.indices().coeff(i)).setZero();
+}
+#endif
 
-    for(Index i = 0; i < dec().rank(); ++i) dst.row(dec().colsPermutation().indices().coeff(i)) = c.row(i);
-    for(Index i = dec().rank(); i < cols; ++i) dst.row(dec().colsPermutation().indices().coeff(i)).setZero();
+namespace internal {
+  
+template<typename DstXprType, typename MatrixType>
+struct Assignment<DstXprType, Inverse<FullPivHouseholderQR<MatrixType> >, internal::assign_op<typename DstXprType::Scalar,typename FullPivHouseholderQR<MatrixType>::Scalar>, Dense2Dense>
+{
+  typedef FullPivHouseholderQR<MatrixType> QrType;
+  typedef Inverse<QrType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename QrType::Scalar> &)
+  {    
+    dst = src.nestedExpression().solve(MatrixType::Identity(src.rows(), src.cols()));
   }
 };
 
@@ -546,7 +596,6 @@
   : public ReturnByValue<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename FullPivHouseholderQR<MatrixType>::IntDiagSizeVectorType IntDiagSizeVectorType;
   typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
   typedef Matrix<typename MatrixType::Scalar, 1, MatrixType::RowsAtCompileTime, RowMajor, 1,
@@ -558,7 +607,7 @@
     : m_qr(qr),
       m_hCoeffs(hCoeffs),
       m_rowsTranspositions(rowsTranspositions)
-      {}
+  {}
 
   template <typename ResultType>
   void evalTo(ResultType& result) const
@@ -588,8 +637,8 @@
     }
   }
 
-    Index rows() const { return m_qr.rows(); }
-    Index cols() const { return m_qr.rows(); }
+  Index rows() const { return m_qr.rows(); }
+  Index cols() const { return m_qr.rows(); }
 
 protected:
   typename MatrixType::Nested m_qr;
@@ -597,6 +646,11 @@
   typename IntDiagSizeVectorType::Nested m_rowsTranspositions;
 };
 
+// template<typename MatrixType>
+// struct evaluator<FullPivHouseholderQRMatrixQReturnType<MatrixType> >
+//  : public evaluator<ReturnByValue<FullPivHouseholderQRMatrixQReturnType<MatrixType> > >
+// {};
+
 } // end namespace internal
 
 template<typename MatrixType>
diff --git a/Eigen/src/QR/HouseholderQR.h b/Eigen/src/QR/HouseholderQR.h
index 343a664..3513d99 100644
--- a/Eigen/src/QR/HouseholderQR.h
+++ b/Eigen/src/QR/HouseholderQR.h
@@ -21,7 +21,7 @@
   *
   * \brief Householder QR decomposition of a matrix
   *
-  * \param MatrixType the type of the matrix of which we are computing the QR decomposition
+  * \tparam _MatrixType the type of the matrix of which we are computing the QR decomposition
   *
   * This class performs a QR decomposition of a matrix \b A into matrices \b Q and \b R
   * such that 
@@ -37,6 +37,8 @@
   * This Householder QR decomposition is faster, but less numerically stable and less feature-full than
   * FullPivHouseholderQR or ColPivHouseholderQR.
   *
+  * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism.
+  *
   * \sa MatrixBase::householderQr()
   */
 template<typename _MatrixType> class HouseholderQR
@@ -47,13 +49,13 @@
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
-      Options = MatrixType::Options,
       MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
       MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    // FIXME should be int
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime, (MatrixType::Flags&RowMajorBit) ? RowMajor : ColMajor, MaxRowsAtCompileTime, MaxRowsAtCompileTime> MatrixQType;
     typedef typename internal::plain_diag_type<MatrixType>::type HCoeffsType;
     typedef typename internal::plain_row_type<MatrixType>::type RowVectorType;
@@ -91,13 +93,32 @@
       * 
       * \sa compute()
       */
-    HouseholderQR(const MatrixType& matrix)
+    template<typename InputType>
+    explicit HouseholderQR(const EigenBase<InputType>& matrix)
       : m_qr(matrix.rows(), matrix.cols()),
         m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
         m_temp(matrix.cols()),
         m_isInitialized(false)
     {
-      compute(matrix);
+      compute(matrix.derived());
+    }
+
+
+    /** \brief Constructs a QR factorization from a given matrix
+      *
+      * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when
+      * \c MatrixType is a Eigen::Ref.
+      *
+      * \sa HouseholderQR(const EigenBase&)
+      */
+    template<typename InputType>
+    explicit HouseholderQR(EigenBase<InputType>& matrix)
+      : m_qr(matrix.derived()),
+        m_hCoeffs((std::min)(matrix.rows(),matrix.cols())),
+        m_temp(matrix.cols()),
+        m_isInitialized(false)
+    {
+      computeInPlace();
     }
 
     /** This method finds a solution x to the equation Ax=b, where A is the matrix of which
@@ -107,9 +128,6 @@
       *
       * \returns a solution.
       *
-      * \note The case where b is a matrix is not yet implemented. Also, this
-      *       code is space inefficient.
-      *
       * \note_about_checking_solutions
       *
       * \note_about_arbitrary_choice_of_solution
@@ -118,11 +136,11 @@
       * Output: \verbinclude HouseholderQR_solve.out
       */
     template<typename Rhs>
-    inline const internal::solve_retval<HouseholderQR, Rhs>
+    inline const Solve<HouseholderQR, Rhs>
     solve(const MatrixBase<Rhs>& b) const
     {
       eigen_assert(m_isInitialized && "HouseholderQR is not initialized.");
-      return internal::solve_retval<HouseholderQR, Rhs>(*this, b.derived());
+      return Solve<HouseholderQR, Rhs>(*this, b.derived());
     }
 
     /** This method returns an expression of the unitary matrix Q as a sequence of Householder transformations.
@@ -148,7 +166,12 @@
         return m_qr;
     }
 
-    HouseholderQR& compute(const MatrixType& matrix);
+    template<typename InputType>
+    HouseholderQR& compute(const EigenBase<InputType>& matrix) {
+      m_qr = matrix.derived();
+      computeInPlace();
+      return *this;
+    }
 
     /** \returns the absolute value of the determinant of the matrix of which
       * *this is the QR decomposition. It has only linear complexity
@@ -187,6 +210,12 @@
       * For advanced uses only.
       */
     const HCoeffsType& hCoeffs() const { return m_hCoeffs; }
+    
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    void _solve_impl(const RhsType &rhs, DstType &dst) const;
+    #endif
 
   protected:
     
@@ -194,6 +223,8 @@
     {
       EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
     }
+
+    void computeInPlace();
     
     MatrixType m_qr;
     HCoeffsType m_hCoeffs;
@@ -224,7 +255,6 @@
 template<typename MatrixQR, typename HCoeffs>
 void householder_qr_inplace_unblocked(MatrixQR& mat, HCoeffs& hCoeffs, typename MatrixQR::Scalar* tempData = 0)
 {
-  typedef typename MatrixQR::Index Index;
   typedef typename MatrixQR::Scalar Scalar;
   typedef typename MatrixQR::RealScalar RealScalar;
   Index rows = mat.rows();
@@ -263,11 +293,9 @@
 struct householder_qr_inplace_blocked
 {
   // This is specialized for MKL-supported Scalar types in HouseholderQR_MKL.h
-  static void run(MatrixQR& mat, HCoeffs& hCoeffs,
-      typename MatrixQR::Index maxBlockSize=32,
+  static void run(MatrixQR& mat, HCoeffs& hCoeffs, Index maxBlockSize=32,
       typename MatrixQR::Scalar* tempData = 0)
   {
-    typedef typename MatrixQR::Index Index;
     typedef typename MatrixQR::Scalar Scalar;
     typedef Block<MatrixQR,Dynamic,Dynamic> BlockType;
 
@@ -289,8 +317,8 @@
     for (k = 0; k < size; k += blockSize)
     {
       Index bs = (std::min)(size-k,blockSize);  // actual size of the block
-      Index tcols = cols - k - bs;            // trailing columns
-      Index brows = rows-k;                   // rows of the block
+      Index tcols = cols - k - bs;              // trailing columns
+      Index brows = rows-k;                     // rows of the block
 
       // partition the matrix:
       //        A00 | A01 | A02
@@ -308,44 +336,39 @@
       if(tcols)
       {
         BlockType A21_22 = mat.block(k,k+bs,brows,tcols);
-        apply_block_householder_on_the_left(A21_22,A11_21,hCoeffsSegment.adjoint());
+        apply_block_householder_on_the_left(A21_22,A11_21,hCoeffsSegment, false); // false == backward
       }
     }
   }
 };
 
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<HouseholderQR<_MatrixType>, Rhs>
-  : solve_retval_base<HouseholderQR<_MatrixType>, Rhs>
-{
-  EIGEN_MAKE_SOLVE_HELPERS(HouseholderQR<_MatrixType>,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    const Index rows = dec().rows(), cols = dec().cols();
-    const Index rank = (std::min)(rows, cols);
-    eigen_assert(rhs().rows() == rows);
-
-    typename Rhs::PlainObject c(rhs());
-
-    // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
-    c.applyOnTheLeft(householderSequence(
-      dec().matrixQR().leftCols(rank),
-      dec().hCoeffs().head(rank)).transpose()
-    );
-
-    dec().matrixQR()
-       .topLeftCorner(rank, rank)
-       .template triangularView<Upper>()
-       .solveInPlace(c.topRows(rank));
-
-    dst.topRows(rank) = c.topRows(rank);
-    dst.bottomRows(cols-rank).setZero();
-  }
-};
-
 } // end namespace internal
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename _MatrixType>
+template<typename RhsType, typename DstType>
+void HouseholderQR<_MatrixType>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  const Index rank = (std::min)(rows(), cols());
+  eigen_assert(rhs.rows() == rows());
+
+  typename RhsType::PlainObject c(rhs);
+
+  // Note that the matrix Q = H_0^* H_1^*... so its inverse is Q^* = (H_0 H_1 ...)^T
+  c.applyOnTheLeft(householderSequence(
+    m_qr.leftCols(rank),
+    m_hCoeffs.head(rank)).transpose()
+  );
+
+  m_qr.topLeftCorner(rank, rank)
+      .template triangularView<Upper>()
+      .solveInPlace(c.topRows(rank));
+
+  dst.topRows(rank) = c.topRows(rank);
+  dst.bottomRows(cols()-rank).setZero();
+}
+#endif
+
 /** Performs the QR factorization of the given matrix \a matrix. The result of
   * the factorization is stored into \c *this, and a reference to \c *this
   * is returned.
@@ -353,15 +376,14 @@
   * \sa class HouseholderQR, HouseholderQR(const MatrixType&)
   */
 template<typename MatrixType>
-HouseholderQR<MatrixType>& HouseholderQR<MatrixType>::compute(const MatrixType& matrix)
+void HouseholderQR<MatrixType>::computeInPlace()
 {
   check_template_parameters();
   
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
+  Index rows = m_qr.rows();
+  Index cols = m_qr.cols();
   Index size = (std::min)(rows,cols);
 
-  m_qr = matrix;
   m_hCoeffs.resize(size);
 
   m_temp.resize(cols);
@@ -369,7 +391,6 @@
   internal::householder_qr_inplace_blocked<MatrixType, HCoeffsType>::run(m_qr, m_hCoeffs, 48, m_temp.data());
 
   m_isInitialized = true;
-  return *this;
 }
 
 /** \return the Householder QR decomposition of \c *this.
diff --git a/Eigen/src/QR/HouseholderQR_MKL.h b/Eigen/src/QR/HouseholderQR_LAPACKE.h
similarity index 77%
rename from Eigen/src/QR/HouseholderQR_MKL.h
rename to Eigen/src/QR/HouseholderQR_LAPACKE.h
index b80f1b4..1dc7d53 100644
--- a/Eigen/src/QR/HouseholderQR_MKL.h
+++ b/Eigen/src/QR/HouseholderQR_LAPACKE.h
@@ -25,47 +25,44 @@
  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
  ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
+ *   Content : Eigen bindings to LAPACKe
  *    Householder QR decomposition of a matrix w/o pivoting based on
  *    LAPACKE_?geqrf function.
  ********************************************************************************
 */
 
-#ifndef EIGEN_QR_MKL_H
-#define EIGEN_QR_MKL_H
-
-#include "../Core/util/MKL_support.h"
+#ifndef EIGEN_QR_LAPACKE_H
+#define EIGEN_QR_LAPACKE_H
 
 namespace Eigen { 
 
-  namespace internal {
+namespace internal {
 
-    /** \internal Specialization for the data types supported by MKL */
+/** \internal Specialization for the data types supported by LAPACKe */
 
-#define EIGEN_MKL_QR_NOPIV(EIGTYPE, MKLTYPE, MKLPREFIX) \
+#define EIGEN_LAPACKE_QR_NOPIV(EIGTYPE, LAPACKE_TYPE, LAPACKE_PREFIX) \
 template<typename MatrixQR, typename HCoeffs> \
 struct householder_qr_inplace_blocked<MatrixQR, HCoeffs, EIGTYPE, true> \
 { \
-  static void run(MatrixQR& mat, HCoeffs& hCoeffs, \
-      typename MatrixQR::Index = 32, \
+  static void run(MatrixQR& mat, HCoeffs& hCoeffs, Index = 32, \
       typename MatrixQR::Scalar* = 0) \
   { \
     lapack_int m = (lapack_int) mat.rows(); \
     lapack_int n = (lapack_int) mat.cols(); \
     lapack_int lda = (lapack_int) mat.outerStride(); \
     lapack_int matrix_order = (MatrixQR::IsRowMajor) ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \
-    LAPACKE_##MKLPREFIX##geqrf( matrix_order, m, n, (MKLTYPE*)mat.data(), lda, (MKLTYPE*)hCoeffs.data()); \
+    LAPACKE_##LAPACKE_PREFIX##geqrf( matrix_order, m, n, (LAPACKE_TYPE*)mat.data(), lda, (LAPACKE_TYPE*)hCoeffs.data()); \
     hCoeffs.adjointInPlace(); \
   } \
 };
 
-EIGEN_MKL_QR_NOPIV(double, double, d)
-EIGEN_MKL_QR_NOPIV(float, float, s)
-EIGEN_MKL_QR_NOPIV(dcomplex, MKL_Complex16, z)
-EIGEN_MKL_QR_NOPIV(scomplex, MKL_Complex8, c)
+EIGEN_LAPACKE_QR_NOPIV(double, double, d)
+EIGEN_LAPACKE_QR_NOPIV(float, float, s)
+EIGEN_LAPACKE_QR_NOPIV(dcomplex, lapack_complex_double, z)
+EIGEN_LAPACKE_QR_NOPIV(scomplex, lapack_complex_float, c)
 
 } // end namespace internal
 
 } // end namespace Eigen
 
-#endif // EIGEN_QR_MKL_H
+#endif // EIGEN_QR_LAPACKE_H
diff --git a/Eigen/src/SPQRSupport/CMakeLists.txt b/Eigen/src/SPQRSupport/CMakeLists.txt
deleted file mode 100644
index 4968bea..0000000
--- a/Eigen/src/SPQRSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_SPQRSupport_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_SPQRSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/SPQRSupport/ COMPONENT Devel
-  )
diff --git a/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h b/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h
index de00877..953d57c 100644
--- a/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h
+++ b/Eigen/src/SPQRSupport/SuiteSparseQRSupport.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2012 Desire Nuentsa <desire.nuentsa_wakam@inria.fr>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -32,45 +33,54 @@
   } // End namespace internal
   
 /**
- * \ingroup SPQRSupport_Module
- * \class SPQR
- * \brief Sparse QR factorization based on SuiteSparseQR library
- * 
- * This class is used to perform a multithreaded and multifrontal rank-revealing QR decomposition 
- * of sparse matrices. The result is then used to solve linear leasts_square systems.
- * Clearly, a QR factorization is returned such that A*P = Q*R where :
- * 
- * P is the column permutation. Use colsPermutation() to get it.
- * 
- * Q is the orthogonal matrix represented as Householder reflectors. 
- * Use matrixQ() to get an expression and matrixQ().transpose() to get the transpose.
- * You can then apply it to a vector.
- * 
- * R is the sparse triangular factor. Use matrixQR() to get it as SparseMatrix.
- * NOTE : The Index type of R is always UF_long. You can get it with SPQR::Index
- * 
- * \tparam _MatrixType The type of the sparse matrix A, must be a column-major SparseMatrix<>
- * NOTE 
- * 
- */
+  * \ingroup SPQRSupport_Module
+  * \class SPQR
+  * \brief Sparse QR factorization based on SuiteSparseQR library
+  *
+  * This class is used to perform a multithreaded and multifrontal rank-revealing QR decomposition
+  * of sparse matrices. The result is then used to solve linear leasts_square systems.
+  * Clearly, a QR factorization is returned such that A*P = Q*R where :
+  *
+  * P is the column permutation. Use colsPermutation() to get it.
+  *
+  * Q is the orthogonal matrix represented as Householder reflectors.
+  * Use matrixQ() to get an expression and matrixQ().transpose() to get the transpose.
+  * You can then apply it to a vector.
+  *
+  * R is the sparse triangular factor. Use matrixQR() to get it as SparseMatrix.
+  * NOTE : The Index type of R is always SuiteSparse_long. You can get it with SPQR::Index
+  *
+  * \tparam _MatrixType The type of the sparse matrix A, must be a column-major SparseMatrix<>
+  *
+  * \implsparsesolverconcept
+  *
+  *
+  */
 template<typename _MatrixType>
-class SPQR
+class SPQR : public SparseSolverBase<SPQR<_MatrixType> >
 {
+  protected:
+    typedef SparseSolverBase<SPQR<_MatrixType> > Base;
+    using Base::m_isInitialized;
   public:
     typedef typename _MatrixType::Scalar Scalar;
     typedef typename _MatrixType::RealScalar RealScalar;
-    typedef UF_long Index ; 
-    typedef SparseMatrix<Scalar, ColMajor, Index> MatrixType;
-    typedef PermutationMatrix<Dynamic, Dynamic> PermutationType;
+    typedef SuiteSparse_long StorageIndex ;
+    typedef SparseMatrix<Scalar, ColMajor, StorageIndex> MatrixType;
+    typedef Map<PermutationMatrix<Dynamic, Dynamic, StorageIndex> > PermutationType;
+    enum {
+      ColsAtCompileTime = Dynamic,
+      MaxColsAtCompileTime = Dynamic
+    };
   public:
     SPQR() 
-      : m_isInitialized(false), m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
+      : m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
     { 
       cholmod_l_start(&m_cc);
     }
     
-    SPQR(const _MatrixType& matrix)
-    : m_isInitialized(false), m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
+    explicit SPQR(const _MatrixType& matrix)
+    : m_ordering(SPQR_ORDERING_DEFAULT), m_allow_tol(SPQR_DEFAULT_TOL), m_tolerance (NumTraits<Scalar>::epsilon()), m_useDefaultThreshold(true)
     {
       cholmod_l_start(&m_cc);
       compute(matrix);
@@ -103,23 +113,22 @@
       RealScalar pivotThreshold = m_tolerance;
       if(m_useDefaultThreshold) 
       {
-        using std::max;
         RealScalar max2Norm = 0.0;
-        for (int j = 0; j < mat.cols(); j++) max2Norm = (max)(max2Norm, mat.col(j).norm());
+        for (int j = 0; j < mat.cols(); j++) max2Norm = numext::maxi(max2Norm, mat.col(j).norm());
         if(max2Norm==RealScalar(0))
           max2Norm = RealScalar(1);
         pivotThreshold = 20 * (mat.rows() + mat.cols()) * max2Norm * NumTraits<RealScalar>::epsilon();
       }
-      
       cholmod_sparse A; 
       A = viewAsCholmod(mat);
+      m_rows = matrix.rows();
       Index col = matrix.cols();
       m_rank = SuiteSparseQR<Scalar>(m_ordering, pivotThreshold, col, &A, 
                              &m_cR, &m_E, &m_H, &m_HPinv, &m_HTau, &m_cc);
 
       if (!m_cR)
       {
-        m_info = NumericalIssue; 
+        m_info = NumericalIssue;
         m_isInitialized = false;
         return;
       }
@@ -130,28 +139,15 @@
     /** 
      * Get the number of rows of the input matrix and the Q matrix
      */
-    inline Index rows() const {return m_cR->nrow; }
+    inline Index rows() const {return m_rows; }
     
     /** 
      * Get the number of columns of the input matrix. 
      */
     inline Index cols() const { return m_cR->ncol; }
-   
-      /** \returns the solution X of \f$ A X = B \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<SPQR, Rhs> solve(const MatrixBase<Rhs>& B) const 
-    {
-      eigen_assert(m_isInitialized && " The QR factorization should be computed first, call compute()");
-      eigen_assert(this->rows()==B.rows()
-                    && "SPQR::solve(): invalid number of rows of the right hand side matrix B");
-          return internal::solve_retval<SPQR, Rhs>(*this, B.derived());
-    }
     
     template<typename Rhs, typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_isInitialized && " The QR factorization should be computed first, call compute()");
       eigen_assert(b.cols()==1 && "This method is for vectors only");
@@ -184,7 +180,7 @@
     {
       eigen_assert(m_isInitialized && " The QR factorization should be computed first, call compute()");
       if(!m_isRUpToDate) {
-        m_R = viewAsEigen<Scalar,ColMajor, typename MatrixType::Index>(*m_cR);
+        m_R = viewAsEigen<Scalar,ColMajor, typename MatrixType::StorageIndex>(*m_cR);
         m_isRUpToDate = true;
       }
       return m_R;
@@ -198,11 +194,7 @@
     PermutationType colsPermutation() const
     { 
       eigen_assert(m_isInitialized && "Decomposition is not initialized.");
-      Index n = m_cR->ncol;
-      PermutationType colsPerm(n);
-      for(Index j = 0; j <n; j++) colsPerm.indices()(j) = m_E[j];
-      return colsPerm; 
-      
+      return PermutationType(m_E, m_cR->ncol);
     }
     /**
      * Gets the rank of the matrix. 
@@ -237,7 +229,6 @@
       return m_info;
     }
   protected:
-    bool m_isInitialized;
     bool m_analysisIsOk;
     bool m_factorizationIsOk;
     mutable bool m_isRUpToDate;
@@ -247,13 +238,14 @@
     RealScalar m_tolerance; // treat columns with 2-norm below this tolerance as zero
     mutable cholmod_sparse *m_cR; // The sparse R factor in cholmod format
     mutable MatrixType m_R; // The sparse matrix R in Eigen format
-    mutable Index *m_E; // The permutation applied to columns
+    mutable StorageIndex *m_E; // The permutation applied to columns
     mutable cholmod_sparse *m_H;  //The householder vectors
-    mutable Index *m_HPinv; // The row permutation of H
+    mutable StorageIndex *m_HPinv; // The row permutation of H
     mutable cholmod_dense *m_HTau; // The Householder coefficients
     mutable Index m_rank; // The rank of the matrix
     mutable cholmod_common m_cc; // Workspace and parameters
     bool m_useDefaultThreshold;     // Use default threshold
+    Index m_rows;
     template<typename ,typename > friend struct SPQR_QProduct;
 };
 
@@ -261,7 +253,7 @@
 struct SPQR_QProduct : ReturnByValue<SPQR_QProduct<SPQRType,Derived> >
 {
   typedef typename SPQRType::Scalar Scalar;
-  typedef typename SPQRType::Index Index;
+  typedef typename SPQRType::StorageIndex StorageIndex;
   //Define the constructor to get reference to argument types
   SPQR_QProduct(const SPQRType& spqr, const Derived& other, bool transpose) : m_spqr(spqr),m_other(other),m_transpose(transpose) {}
   
@@ -317,22 +309,5 @@
   const SPQRType& m_spqr;
 };
 
-namespace internal {
-  
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<SPQR<_MatrixType>, Rhs>
-  : solve_retval_base<SPQR<_MatrixType>, Rhs>
-{
-  typedef SPQR<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-} // end namespace internal
-
 }// End namespace Eigen
 #endif
diff --git a/Eigen/src/SVD/BDCSVD.h b/Eigen/src/SVD/BDCSVD.h
new file mode 100644
index 0000000..1134d66
--- /dev/null
+++ b/Eigen/src/SVD/BDCSVD.h
@@ -0,0 +1,1246 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+// 
+// We used the "A Divide-And-Conquer Algorithm for the Bidiagonal SVD"
+// research report written by Ming Gu and Stanley C.Eisenstat
+// The code variable names correspond to the names they used in their 
+// report
+//
+// Copyright (C) 2013 Gauthier Brun <brun.gauthier@gmail.com>
+// Copyright (C) 2013 Nicolas Carre <nicolas.carre@ensimag.fr>
+// Copyright (C) 2013 Jean Ceccato <jean.ceccato@ensimag.fr>
+// Copyright (C) 2013 Pierre Zoppitelli <pierre.zoppitelli@ensimag.fr>
+// Copyright (C) 2013 Jitse Niesen <jitse@maths.leeds.ac.uk>
+// Copyright (C) 2014-2017 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_BDCSVD_H
+#define EIGEN_BDCSVD_H
+// #define EIGEN_BDCSVD_DEBUG_VERBOSE
+// #define EIGEN_BDCSVD_SANITY_CHECKS
+
+namespace Eigen {
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+IOFormat bdcsvdfmt(8, 0, ", ", "\n", "  [", "]");
+#endif
+  
+template<typename _MatrixType> class BDCSVD;
+
+namespace internal {
+
+template<typename _MatrixType> 
+struct traits<BDCSVD<_MatrixType> >
+{
+  typedef _MatrixType MatrixType;
+};  
+
+} // end namespace internal
+  
+  
+/** \ingroup SVD_Module
+ *
+ *
+ * \class BDCSVD
+ *
+ * \brief class Bidiagonal Divide and Conquer SVD
+ *
+ * \tparam _MatrixType the type of the matrix of which we are computing the SVD decomposition
+ *
+ * This class first reduces the input matrix to bi-diagonal form using class UpperBidiagonalization,
+ * and then performs a divide-and-conquer diagonalization. Small blocks are diagonalized using class JacobiSVD.
+ * You can control the switching size with the setSwitchSize() method, default is 16.
+ * For small matrice (<16), it is thus preferable to directly use JacobiSVD. For larger ones, BDCSVD is highly
+ * recommended and can several order of magnitude faster.
+ *
+ * \warning this algorithm is unlikely to provide accurate result when compiled with unsafe math optimizations.
+ * For instance, this concerns Intel's compiler (ICC), which perfroms such optimization by default unless
+ * you compile with the \c -fp-model \c precise option. Likewise, the \c -ffast-math option of GCC or clang will
+ * significantly degrade the accuracy.
+ *
+ * \sa class JacobiSVD
+ */
+template<typename _MatrixType> 
+class BDCSVD : public SVDBase<BDCSVD<_MatrixType> >
+{
+  typedef SVDBase<BDCSVD> Base;
+    
+public:
+  using Base::rows;
+  using Base::cols;
+  using Base::computeU;
+  using Base::computeV;
+  
+  typedef _MatrixType MatrixType;
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+  typedef typename NumTraits<RealScalar>::Literal Literal;
+  enum {
+    RowsAtCompileTime = MatrixType::RowsAtCompileTime, 
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime, 
+    DiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime, ColsAtCompileTime), 
+    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime, 
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime, 
+    MaxDiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(MaxRowsAtCompileTime, MaxColsAtCompileTime), 
+    MatrixOptions = MatrixType::Options
+  };
+
+  typedef typename Base::MatrixUType MatrixUType;
+  typedef typename Base::MatrixVType MatrixVType;
+  typedef typename Base::SingularValuesType SingularValuesType;
+  
+  typedef Matrix<Scalar, Dynamic, Dynamic, ColMajor> MatrixX;
+  typedef Matrix<RealScalar, Dynamic, Dynamic, ColMajor> MatrixXr;
+  typedef Matrix<RealScalar, Dynamic, 1> VectorType;
+  typedef Array<RealScalar, Dynamic, 1> ArrayXr;
+  typedef Array<Index,1,Dynamic> ArrayXi;
+  typedef Ref<ArrayXr> ArrayRef;
+  typedef Ref<ArrayXi> IndicesRef;
+
+  /** \brief Default Constructor.
+   *
+   * The default constructor is useful in cases in which the user intends to
+   * perform decompositions via BDCSVD::compute(const MatrixType&).
+   */
+  BDCSVD() : m_algoswap(16), m_numIters(0)
+  {}
+
+
+  /** \brief Default Constructor with memory preallocation
+   *
+   * Like the default constructor but with preallocation of the internal data
+   * according to the specified problem size.
+   * \sa BDCSVD()
+   */
+  BDCSVD(Index rows, Index cols, unsigned int computationOptions = 0)
+    : m_algoswap(16), m_numIters(0)
+  {
+    allocate(rows, cols, computationOptions);
+  }
+
+  /** \brief Constructor performing the decomposition of given matrix.
+   *
+   * \param matrix the matrix to decompose
+   * \param computationOptions optional parameter allowing to specify if you want full or thin U or V unitaries to be computed.
+   *                           By default, none is computed. This is a bit - field, the possible bits are #ComputeFullU, #ComputeThinU, 
+   *                           #ComputeFullV, #ComputeThinV.
+   *
+   * Thin unitaries are only available if your matrix type has a Dynamic number of columns (for example MatrixXf). They also are not
+   * available with the (non - default) FullPivHouseholderQR preconditioner.
+   */
+  BDCSVD(const MatrixType& matrix, unsigned int computationOptions = 0)
+    : m_algoswap(16), m_numIters(0)
+  {
+    compute(matrix, computationOptions);
+  }
+
+  ~BDCSVD() 
+  {
+  }
+  
+  /** \brief Method performing the decomposition of given matrix using custom options.
+   *
+   * \param matrix the matrix to decompose
+   * \param computationOptions optional parameter allowing to specify if you want full or thin U or V unitaries to be computed.
+   *                           By default, none is computed. This is a bit - field, the possible bits are #ComputeFullU, #ComputeThinU, 
+   *                           #ComputeFullV, #ComputeThinV.
+   *
+   * Thin unitaries are only available if your matrix type has a Dynamic number of columns (for example MatrixXf). They also are not
+   * available with the (non - default) FullPivHouseholderQR preconditioner.
+   */
+  BDCSVD& compute(const MatrixType& matrix, unsigned int computationOptions);
+
+  /** \brief Method performing the decomposition of given matrix using current options.
+   *
+   * \param matrix the matrix to decompose
+   *
+   * This method uses the current \a computationOptions, as already passed to the constructor or to compute(const MatrixType&, unsigned int).
+   */
+  BDCSVD& compute(const MatrixType& matrix)
+  {
+    return compute(matrix, this->m_computationOptions);
+  }
+
+  void setSwitchSize(int s) 
+  {
+    eigen_assert(s>3 && "BDCSVD the size of the algo switch has to be greater than 3");
+    m_algoswap = s;
+  }
+ 
+private:
+  void allocate(Index rows, Index cols, unsigned int computationOptions);
+  void divide(Index firstCol, Index lastCol, Index firstRowW, Index firstColW, Index shift);
+  void computeSVDofM(Index firstCol, Index n, MatrixXr& U, VectorType& singVals, MatrixXr& V);
+  void computeSingVals(const ArrayRef& col0, const ArrayRef& diag, const IndicesRef& perm, VectorType& singVals, ArrayRef shifts, ArrayRef mus);
+  void perturbCol0(const ArrayRef& col0, const ArrayRef& diag, const IndicesRef& perm, const VectorType& singVals, const ArrayRef& shifts, const ArrayRef& mus, ArrayRef zhat);
+  void computeSingVecs(const ArrayRef& zhat, const ArrayRef& diag, const IndicesRef& perm, const VectorType& singVals, const ArrayRef& shifts, const ArrayRef& mus, MatrixXr& U, MatrixXr& V);
+  void deflation43(Index firstCol, Index shift, Index i, Index size);
+  void deflation44(Index firstColu , Index firstColm, Index firstRowW, Index firstColW, Index i, Index j, Index size);
+  void deflation(Index firstCol, Index lastCol, Index k, Index firstRowW, Index firstColW, Index shift);
+  template<typename HouseholderU, typename HouseholderV, typename NaiveU, typename NaiveV>
+  void copyUV(const HouseholderU &householderU, const HouseholderV &householderV, const NaiveU &naiveU, const NaiveV &naivev);
+  void structured_update(Block<MatrixXr,Dynamic,Dynamic> A, const MatrixXr &B, Index n1);
+  static RealScalar secularEq(RealScalar x, const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm, const ArrayRef& diagShifted, RealScalar shift);
+
+protected:
+  MatrixXr m_naiveU, m_naiveV;
+  MatrixXr m_computed;
+  Index m_nRec;
+  ArrayXr m_workspace;
+  ArrayXi m_workspaceI;
+  int m_algoswap;
+  bool m_isTranspose, m_compU, m_compV;
+  
+  using Base::m_singularValues;
+  using Base::m_diagSize;
+  using Base::m_computeFullU;
+  using Base::m_computeFullV;
+  using Base::m_computeThinU;
+  using Base::m_computeThinV;
+  using Base::m_matrixU;
+  using Base::m_matrixV;
+  using Base::m_isInitialized;
+  using Base::m_nonzeroSingularValues;
+
+public:  
+  int m_numIters;
+}; //end class BDCSVD
+
+
+// Method to allocate and initialize matrix and attributes
+template<typename MatrixType>
+void BDCSVD<MatrixType>::allocate(Index rows, Index cols, unsigned int computationOptions)
+{
+  m_isTranspose = (cols > rows);
+
+  if (Base::allocate(rows, cols, computationOptions))
+    return;
+  
+  m_computed = MatrixXr::Zero(m_diagSize + 1, m_diagSize );
+  m_compU = computeV();
+  m_compV = computeU();
+  if (m_isTranspose)
+    std::swap(m_compU, m_compV);
+  
+  if (m_compU) m_naiveU = MatrixXr::Zero(m_diagSize + 1, m_diagSize + 1 );
+  else         m_naiveU = MatrixXr::Zero(2, m_diagSize + 1 );
+  
+  if (m_compV) m_naiveV = MatrixXr::Zero(m_diagSize, m_diagSize);
+  
+  m_workspace.resize((m_diagSize+1)*(m_diagSize+1)*3);
+  m_workspaceI.resize(3*m_diagSize);
+}// end allocate
+
+template<typename MatrixType>
+BDCSVD<MatrixType>& BDCSVD<MatrixType>::compute(const MatrixType& matrix, unsigned int computationOptions) 
+{
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "\n\n\n======================================================================================================================\n\n\n";
+#endif
+  allocate(matrix.rows(), matrix.cols(), computationOptions);
+  using std::abs;
+
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  
+  //**** step -1 - If the problem is too small, directly falls back to JacobiSVD and return
+  if(matrix.cols() < m_algoswap)
+  {
+    // FIXME this line involves temporaries
+    JacobiSVD<MatrixType> jsvd(matrix,computationOptions);
+    if(computeU()) m_matrixU = jsvd.matrixU();
+    if(computeV()) m_matrixV = jsvd.matrixV();
+    m_singularValues = jsvd.singularValues();
+    m_nonzeroSingularValues = jsvd.nonzeroSingularValues();
+    m_isInitialized = true;
+    return *this;
+  }
+  
+  //**** step 0 - Copy the input matrix and apply scaling to reduce over/under-flows
+  RealScalar scale = matrix.cwiseAbs().maxCoeff();
+  if(scale==Literal(0)) scale = Literal(1);
+  MatrixX copy;
+  if (m_isTranspose) copy = matrix.adjoint()/scale;
+  else               copy = matrix/scale;
+  
+  //**** step 1 - Bidiagonalization
+  // FIXME this line involves temporaries
+  internal::UpperBidiagonalization<MatrixX> bid(copy);
+
+  //**** step 2 - Divide & Conquer
+  m_naiveU.setZero();
+  m_naiveV.setZero();
+  // FIXME this line involves a temporary matrix
+  m_computed.topRows(m_diagSize) = bid.bidiagonal().toDenseMatrix().transpose();
+  m_computed.template bottomRows<1>().setZero();
+  divide(0, m_diagSize - 1, 0, 0, 0);
+
+  //**** step 3 - Copy singular values and vectors
+  for (int i=0; i<m_diagSize; i++)
+  {
+    RealScalar a = abs(m_computed.coeff(i, i));
+    m_singularValues.coeffRef(i) = a * scale;
+    if (a<considerZero)
+    {
+      m_nonzeroSingularValues = i;
+      m_singularValues.tail(m_diagSize - i - 1).setZero();
+      break;
+    }
+    else if (i == m_diagSize - 1)
+    {
+      m_nonzeroSingularValues = i + 1;
+      break;
+    }
+  }
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+//   std::cout << "m_naiveU\n" << m_naiveU << "\n\n";
+//   std::cout << "m_naiveV\n" << m_naiveV << "\n\n";
+#endif
+  if(m_isTranspose) copyUV(bid.householderV(), bid.householderU(), m_naiveV, m_naiveU);
+  else              copyUV(bid.householderU(), bid.householderV(), m_naiveU, m_naiveV);
+
+  m_isInitialized = true;
+  return *this;
+}// end compute
+
+
+template<typename MatrixType>
+template<typename HouseholderU, typename HouseholderV, typename NaiveU, typename NaiveV>
+void BDCSVD<MatrixType>::copyUV(const HouseholderU &householderU, const HouseholderV &householderV, const NaiveU &naiveU, const NaiveV &naiveV)
+{
+  // Note exchange of U and V: m_matrixU is set from m_naiveV and vice versa
+  if (computeU())
+  {
+    Index Ucols = m_computeThinU ? m_diagSize : householderU.cols();
+    m_matrixU = MatrixX::Identity(householderU.cols(), Ucols);
+    m_matrixU.topLeftCorner(m_diagSize, m_diagSize) = naiveV.template cast<Scalar>().topLeftCorner(m_diagSize, m_diagSize);
+    householderU.applyThisOnTheLeft(m_matrixU); // FIXME this line involves a temporary buffer
+  }
+  if (computeV())
+  {
+    Index Vcols = m_computeThinV ? m_diagSize : householderV.cols();
+    m_matrixV = MatrixX::Identity(householderV.cols(), Vcols);
+    m_matrixV.topLeftCorner(m_diagSize, m_diagSize) = naiveU.template cast<Scalar>().topLeftCorner(m_diagSize, m_diagSize);
+    householderV.applyThisOnTheLeft(m_matrixV); // FIXME this line involves a temporary buffer
+  }
+}
+
+/** \internal
+  * Performs A = A * B exploiting the special structure of the matrix A. Splitting A as:
+  *  A = [A1]
+  *      [A2]
+  * such that A1.rows()==n1, then we assume that at least half of the columns of A1 and A2 are zeros.
+  * We can thus pack them prior to the the matrix product. However, this is only worth the effort if the matrix is large
+  * enough.
+  */
+template<typename MatrixType>
+void BDCSVD<MatrixType>::structured_update(Block<MatrixXr,Dynamic,Dynamic> A, const MatrixXr &B, Index n1)
+{
+  Index n = A.rows();
+  if(n>100)
+  {
+    // If the matrices are large enough, let's exploit the sparse structure of A by
+    // splitting it in half (wrt n1), and packing the non-zero columns.
+    Index n2 = n - n1;
+    Map<MatrixXr> A1(m_workspace.data()      , n1, n);
+    Map<MatrixXr> A2(m_workspace.data()+ n1*n, n2, n);
+    Map<MatrixXr> B1(m_workspace.data()+  n*n, n,  n);
+    Map<MatrixXr> B2(m_workspace.data()+2*n*n, n,  n);
+    Index k1=0, k2=0;
+    for(Index j=0; j<n; ++j)
+    {
+      if( (A.col(j).head(n1).array()!=Literal(0)).any() )
+      {
+        A1.col(k1) = A.col(j).head(n1);
+        B1.row(k1) = B.row(j);
+        ++k1;
+      }
+      if( (A.col(j).tail(n2).array()!=Literal(0)).any() )
+      {
+        A2.col(k2) = A.col(j).tail(n2);
+        B2.row(k2) = B.row(j);
+        ++k2;
+      }
+    }
+  
+    A.topRows(n1).noalias()    = A1.leftCols(k1) * B1.topRows(k1);
+    A.bottomRows(n2).noalias() = A2.leftCols(k2) * B2.topRows(k2);
+  }
+  else
+  {
+    Map<MatrixXr,Aligned> tmp(m_workspace.data(),n,n);
+    tmp.noalias() = A*B;
+    A = tmp;
+  }
+}
+
+// The divide algorithm is done "in place", we are always working on subsets of the same matrix. The divide methods takes as argument the 
+// place of the submatrix we are currently working on.
+
+//@param firstCol : The Index of the first column of the submatrix of m_computed and for m_naiveU;
+//@param lastCol : The Index of the last column of the submatrix of m_computed and for m_naiveU; 
+// lastCol + 1 - firstCol is the size of the submatrix.
+//@param firstRowW : The Index of the first row of the matrix W that we are to change. (see the reference paper section 1 for more information on W)
+//@param firstRowW : Same as firstRowW with the column.
+//@param shift : Each time one takes the left submatrix, one must add 1 to the shift. Why? Because! We actually want the last column of the U submatrix 
+// to become the first column (*coeff) and to shift all the other columns to the right. There are more details on the reference paper.
+template<typename MatrixType>
+void BDCSVD<MatrixType>::divide (Index firstCol, Index lastCol, Index firstRowW, Index firstColW, Index shift)
+{
+  // requires rows = cols + 1;
+  using std::pow;
+  using std::sqrt;
+  using std::abs;
+  const Index n = lastCol - firstCol + 1;
+  const Index k = n/2;
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  RealScalar alphaK;
+  RealScalar betaK; 
+  RealScalar r0; 
+  RealScalar lambda, phi, c0, s0;
+  VectorType l, f;
+  // We use the other algorithm which is more efficient for small 
+  // matrices.
+  if (n < m_algoswap)
+  {
+    // FIXME this line involves temporaries
+    JacobiSVD<MatrixXr> b(m_computed.block(firstCol, firstCol, n + 1, n), ComputeFullU | (m_compV ? ComputeFullV : 0));
+    if (m_compU)
+      m_naiveU.block(firstCol, firstCol, n + 1, n + 1).real() = b.matrixU();
+    else 
+    {
+      m_naiveU.row(0).segment(firstCol, n + 1).real() = b.matrixU().row(0);
+      m_naiveU.row(1).segment(firstCol, n + 1).real() = b.matrixU().row(n);
+    }
+    if (m_compV) m_naiveV.block(firstRowW, firstColW, n, n).real() = b.matrixV();
+    m_computed.block(firstCol + shift, firstCol + shift, n + 1, n).setZero();
+    m_computed.diagonal().segment(firstCol + shift, n) = b.singularValues().head(n);
+    return;
+  }
+  // We use the divide and conquer algorithm
+  alphaK =  m_computed(firstCol + k, firstCol + k);
+  betaK = m_computed(firstCol + k + 1, firstCol + k);
+  // The divide must be done in that order in order to have good results. Divide change the data inside the submatrices
+  // and the divide of the right submatrice reads one column of the left submatrice. That's why we need to treat the 
+  // right submatrix before the left one. 
+  divide(k + 1 + firstCol, lastCol, k + 1 + firstRowW, k + 1 + firstColW, shift);
+  divide(firstCol, k - 1 + firstCol, firstRowW, firstColW + 1, shift + 1);
+
+  if (m_compU)
+  {
+    lambda = m_naiveU(firstCol + k, firstCol + k);
+    phi = m_naiveU(firstCol + k + 1, lastCol + 1);
+  } 
+  else 
+  {
+    lambda = m_naiveU(1, firstCol + k);
+    phi = m_naiveU(0, lastCol + 1);
+  }
+  r0 = sqrt((abs(alphaK * lambda) * abs(alphaK * lambda)) + abs(betaK * phi) * abs(betaK * phi));
+  if (m_compU)
+  {
+    l = m_naiveU.row(firstCol + k).segment(firstCol, k);
+    f = m_naiveU.row(firstCol + k + 1).segment(firstCol + k + 1, n - k - 1);
+  } 
+  else 
+  {
+    l = m_naiveU.row(1).segment(firstCol, k);
+    f = m_naiveU.row(0).segment(firstCol + k + 1, n - k - 1);
+  }
+  if (m_compV) m_naiveV(firstRowW+k, firstColW) = Literal(1);
+  if (r0<considerZero)
+  {
+    c0 = Literal(1);
+    s0 = Literal(0);
+  }
+  else
+  {
+    c0 = alphaK * lambda / r0;
+    s0 = betaK * phi / r0;
+  }
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  if (m_compU)
+  {
+    MatrixXr q1 (m_naiveU.col(firstCol + k).segment(firstCol, k + 1));     
+    // we shiftW Q1 to the right
+    for (Index i = firstCol + k - 1; i >= firstCol; i--) 
+      m_naiveU.col(i + 1).segment(firstCol, k + 1) = m_naiveU.col(i).segment(firstCol, k + 1);
+    // we shift q1 at the left with a factor c0
+    m_naiveU.col(firstCol).segment( firstCol, k + 1) = (q1 * c0);
+    // last column = q1 * - s0
+    m_naiveU.col(lastCol + 1).segment(firstCol, k + 1) = (q1 * ( - s0));
+    // first column = q2 * s0
+    m_naiveU.col(firstCol).segment(firstCol + k + 1, n - k) = m_naiveU.col(lastCol + 1).segment(firstCol + k + 1, n - k) * s0; 
+    // q2 *= c0
+    m_naiveU.col(lastCol + 1).segment(firstCol + k + 1, n - k) *= c0;
+  } 
+  else 
+  {
+    RealScalar q1 = m_naiveU(0, firstCol + k);
+    // we shift Q1 to the right
+    for (Index i = firstCol + k - 1; i >= firstCol; i--) 
+      m_naiveU(0, i + 1) = m_naiveU(0, i);
+    // we shift q1 at the left with a factor c0
+    m_naiveU(0, firstCol) = (q1 * c0);
+    // last column = q1 * - s0
+    m_naiveU(0, lastCol + 1) = (q1 * ( - s0));
+    // first column = q2 * s0
+    m_naiveU(1, firstCol) = m_naiveU(1, lastCol + 1) *s0; 
+    // q2 *= c0
+    m_naiveU(1, lastCol + 1) *= c0;
+    m_naiveU.row(1).segment(firstCol + 1, k).setZero();
+    m_naiveU.row(0).segment(firstCol + k + 1, n - k - 1).setZero();
+  }
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  m_computed(firstCol + shift, firstCol + shift) = r0;
+  m_computed.col(firstCol + shift).segment(firstCol + shift + 1, k) = alphaK * l.transpose().real();
+  m_computed.col(firstCol + shift).segment(firstCol + shift + k + 1, n - k - 1) = betaK * f.transpose().real();
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  ArrayXr tmp1 = (m_computed.block(firstCol+shift, firstCol+shift, n, n)).jacobiSvd().singularValues();
+#endif
+  // Second part: try to deflate singular values in combined matrix
+  deflation(firstCol, lastCol, k, firstRowW, firstColW, shift);
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  ArrayXr tmp2 = (m_computed.block(firstCol+shift, firstCol+shift, n, n)).jacobiSvd().singularValues();
+  std::cout << "\n\nj1 = " << tmp1.transpose().format(bdcsvdfmt) << "\n";
+  std::cout << "j2 = " << tmp2.transpose().format(bdcsvdfmt) << "\n\n";
+  std::cout << "err:      " << ((tmp1-tmp2).abs()>1e-12*tmp2.abs()).transpose() << "\n";
+  static int count = 0;
+  std::cout << "# " << ++count << "\n\n";
+  assert((tmp1-tmp2).matrix().norm() < 1e-14*tmp2.matrix().norm());
+//   assert(count<681);
+//   assert(((tmp1-tmp2).abs()<1e-13*tmp2.abs()).all());
+#endif
+  
+  // Third part: compute SVD of combined matrix
+  MatrixXr UofSVD, VofSVD;
+  VectorType singVals;
+  computeSVDofM(firstCol + shift, n, UofSVD, singVals, VofSVD);
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(UofSVD.allFinite());
+  assert(VofSVD.allFinite());
+#endif
+  
+  if (m_compU)
+    structured_update(m_naiveU.block(firstCol, firstCol, n + 1, n + 1), UofSVD, (n+2)/2);
+  else
+  {
+    Map<Matrix<RealScalar,2,Dynamic>,Aligned> tmp(m_workspace.data(),2,n+1);
+    tmp.noalias() = m_naiveU.middleCols(firstCol, n+1) * UofSVD;
+    m_naiveU.middleCols(firstCol, n + 1) = tmp;
+  }
+  
+  if (m_compV)  structured_update(m_naiveV.block(firstRowW, firstColW, n, n), VofSVD, (n+1)/2);
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  m_computed.block(firstCol + shift, firstCol + shift, n, n).setZero();
+  m_computed.block(firstCol + shift, firstCol + shift, n, n).diagonal() = singVals;
+}// end divide
+
+// Compute SVD of m_computed.block(firstCol, firstCol, n + 1, n); this block only has non-zeros in
+// the first column and on the diagonal and has undergone deflation, so diagonal is in increasing
+// order except for possibly the (0,0) entry. The computed SVD is stored U, singVals and V, except
+// that if m_compV is false, then V is not computed. Singular values are sorted in decreasing order.
+//
+// TODO Opportunities for optimization: better root finding algo, better stopping criterion, better
+// handling of round-off errors, be consistent in ordering
+// For instance, to solve the secular equation using FMM, see http://www.stat.uchicago.edu/~lekheng/courses/302/classics/greengard-rokhlin.pdf
+template <typename MatrixType>
+void BDCSVD<MatrixType>::computeSVDofM(Index firstCol, Index n, MatrixXr& U, VectorType& singVals, MatrixXr& V)
+{
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  using std::abs;
+  ArrayRef col0 = m_computed.col(firstCol).segment(firstCol, n);
+  m_workspace.head(n) =  m_computed.block(firstCol, firstCol, n, n).diagonal();
+  ArrayRef diag = m_workspace.head(n);
+  diag(0) = Literal(0);
+
+  // Allocate space for singular values and vectors
+  singVals.resize(n);
+  U.resize(n+1, n+1);
+  if (m_compV) V.resize(n, n);
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  if (col0.hasNaN() || diag.hasNaN())
+    std::cout << "\n\nHAS NAN\n\n";
+#endif
+  
+  // Many singular values might have been deflated, the zero ones have been moved to the end,
+  // but others are interleaved and we must ignore them at this stage.
+  // To this end, let's compute a permutation skipping them:
+  Index actual_n = n;
+  while(actual_n>1 && diag(actual_n-1)==Literal(0)) --actual_n;
+  Index m = 0; // size of the deflated problem
+  for(Index k=0;k<actual_n;++k)
+    if(abs(col0(k))>considerZero)
+      m_workspaceI(m++) = k;
+  Map<ArrayXi> perm(m_workspaceI.data(),m);
+  
+  Map<ArrayXr> shifts(m_workspace.data()+1*n, n);
+  Map<ArrayXr> mus(m_workspace.data()+2*n, n);
+  Map<ArrayXr> zhat(m_workspace.data()+3*n, n);
+
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "computeSVDofM using:\n";
+  std::cout << "  z: " << col0.transpose() << "\n";
+  std::cout << "  d: " << diag.transpose() << "\n";
+#endif
+  
+  // Compute singVals, shifts, and mus
+  computeSingVals(col0, diag, perm, singVals, shifts, mus);
+  
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "  j:        " << (m_computed.block(firstCol, firstCol, n, n)).jacobiSvd().singularValues().transpose().reverse() << "\n\n";
+  std::cout << "  sing-val: " << singVals.transpose() << "\n";
+  std::cout << "  mu:       " << mus.transpose() << "\n";
+  std::cout << "  shift:    " << shifts.transpose() << "\n";
+  
+  {
+    Index actual_n = n;
+    while(actual_n>1 && abs(col0(actual_n-1))<considerZero) --actual_n;
+    std::cout << "\n\n    mus:    " << mus.head(actual_n).transpose() << "\n\n";
+    std::cout << "    check1 (expect0) : " << ((singVals.array()-(shifts+mus)) / singVals.array()).head(actual_n).transpose() << "\n\n";
+    std::cout << "    check2 (>0)      : " << ((singVals.array()-diag) / singVals.array()).head(actual_n).transpose() << "\n\n";
+    std::cout << "    check3 (>0)      : " << ((diag.segment(1,actual_n-1)-singVals.head(actual_n-1).array()) / singVals.head(actual_n-1).array()).transpose() << "\n\n\n";
+    std::cout << "    check4 (>0)      : " << ((singVals.segment(1,actual_n-1)-singVals.head(actual_n-1))).transpose() << "\n\n\n";
+  }
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(singVals.allFinite());
+  assert(mus.allFinite());
+  assert(shifts.allFinite());
+#endif
+  
+  // Compute zhat
+  perturbCol0(col0, diag, perm, singVals, shifts, mus, zhat);
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "  zhat: " << zhat.transpose() << "\n";
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(zhat.allFinite());
+#endif
+  
+  computeSingVecs(zhat, diag, perm, singVals, shifts, mus, U, V);
+  
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "U^T U: " << (U.transpose() * U - MatrixXr(MatrixXr::Identity(U.cols(),U.cols()))).norm() << "\n";
+  std::cout << "V^T V: " << (V.transpose() * V - MatrixXr(MatrixXr::Identity(V.cols(),V.cols()))).norm() << "\n";
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(U.allFinite());
+  assert(V.allFinite());
+  assert((U.transpose() * U - MatrixXr(MatrixXr::Identity(U.cols(),U.cols()))).norm() < 1e-14 * n);
+  assert((V.transpose() * V - MatrixXr(MatrixXr::Identity(V.cols(),V.cols()))).norm() < 1e-14 * n);
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+  
+  // Because of deflation, the singular values might not be completely sorted.
+  // Fortunately, reordering them is a O(n) problem
+  for(Index i=0; i<actual_n-1; ++i)
+  {
+    if(singVals(i)>singVals(i+1))
+    {
+      using std::swap;
+      swap(singVals(i),singVals(i+1));
+      U.col(i).swap(U.col(i+1));
+      if(m_compV) V.col(i).swap(V.col(i+1));
+    }
+  }
+  
+  // Reverse order so that singular values in increased order
+  // Because of deflation, the zeros singular-values are already at the end
+  singVals.head(actual_n).reverseInPlace();
+  U.leftCols(actual_n).rowwise().reverseInPlace();
+  if (m_compV) V.leftCols(actual_n).rowwise().reverseInPlace();
+  
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  JacobiSVD<MatrixXr> jsvd(m_computed.block(firstCol, firstCol, n, n) );
+  std::cout << "  * j:        " << jsvd.singularValues().transpose() << "\n\n";
+  std::cout << "  * sing-val: " << singVals.transpose() << "\n";
+//   std::cout << "  * err:      " << ((jsvd.singularValues()-singVals)>1e-13*singVals.norm()).transpose() << "\n";
+#endif
+}
+
+template <typename MatrixType>
+typename BDCSVD<MatrixType>::RealScalar BDCSVD<MatrixType>::secularEq(RealScalar mu, const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm, const ArrayRef& diagShifted, RealScalar shift)
+{
+  Index m = perm.size();
+  RealScalar res = Literal(1);
+  for(Index i=0; i<m; ++i)
+  {
+    Index j = perm(i);
+    // The following expression could be rewritten to involve only a single division,
+    // but this would make the expression more sensitive to overflow.
+    res += (col0(j) / (diagShifted(j) - mu)) * (col0(j) / (diag(j) + shift + mu));
+  }
+  return res;
+
+}
+
+template <typename MatrixType>
+void BDCSVD<MatrixType>::computeSingVals(const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm,
+                                         VectorType& singVals, ArrayRef shifts, ArrayRef mus)
+{
+  using std::abs;
+  using std::swap;
+  using std::sqrt;
+
+  Index n = col0.size();
+  Index actual_n = n;
+  // Note that here actual_n is computed based on col0(i)==0 instead of diag(i)==0 as above
+  // because 1) we have diag(i)==0 => col0(i)==0 and 2) if col0(i)==0, then diag(i) is already a singular value.
+  while(actual_n>1 && col0(actual_n-1)==Literal(0)) --actual_n;
+
+  for (Index k = 0; k < n; ++k)
+  {
+    if (col0(k) == Literal(0) || actual_n==1)
+    {
+      // if col0(k) == 0, then entry is deflated, so singular value is on diagonal
+      // if actual_n==1, then the deflated problem is already diagonalized
+      singVals(k) = k==0 ? col0(0) : diag(k);
+      mus(k) = Literal(0);
+      shifts(k) = k==0 ? col0(0) : diag(k);
+      continue;
+    } 
+
+    // otherwise, use secular equation to find singular value
+    RealScalar left = diag(k);
+    RealScalar right; // was: = (k != actual_n-1) ? diag(k+1) : (diag(actual_n-1) + col0.matrix().norm());
+    if(k==actual_n-1)
+      right = (diag(actual_n-1) + col0.matrix().norm());
+    else
+    {
+      // Skip deflated singular values,
+      // recall that at this stage we assume that z[j]!=0 and all entries for which z[j]==0 have been put aside.
+      // This should be equivalent to using perm[]
+      Index l = k+1;
+      while(col0(l)==Literal(0)) { ++l; eigen_internal_assert(l<actual_n); }
+      right = diag(l);
+    }
+
+    // first decide whether it's closer to the left end or the right end
+    RealScalar mid = left + (right-left) / Literal(2);
+    RealScalar fMid = secularEq(mid, col0, diag, perm, diag, Literal(0));
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+    std::cout << right-left << "\n";
+    std::cout << "fMid = " << fMid << " " << secularEq(mid-left, col0, diag, perm, diag-left, left) << " " << secularEq(mid-right, col0, diag, perm, diag-right, right)   << "\n";
+    std::cout << "     = " << secularEq(0.1*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.2*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.3*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.4*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.49*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.5*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.51*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.6*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.7*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.8*(left+right), col0, diag, perm, diag, 0)
+              << " "       << secularEq(0.9*(left+right), col0, diag, perm, diag, 0) << "\n";
+#endif
+    RealScalar shift = (k == actual_n-1 || fMid > Literal(0)) ? left : right;
+    
+    // measure everything relative to shift
+    Map<ArrayXr> diagShifted(m_workspace.data()+4*n, n);
+    diagShifted = diag - shift;
+    
+    // initial guess
+    RealScalar muPrev, muCur;
+    if (shift == left)
+    {
+      muPrev = (right - left) * RealScalar(0.1);
+      if (k == actual_n-1) muCur = right - left;
+      else                 muCur = (right - left) * RealScalar(0.5);
+    }
+    else
+    {
+      muPrev = -(right - left) * RealScalar(0.1);
+      muCur = -(right - left) * RealScalar(0.5);
+    }
+
+    RealScalar fPrev = secularEq(muPrev, col0, diag, perm, diagShifted, shift);
+    RealScalar fCur = secularEq(muCur, col0, diag, perm, diagShifted, shift);
+    if (abs(fPrev) < abs(fCur))
+    {
+      swap(fPrev, fCur);
+      swap(muPrev, muCur);
+    }
+
+    // rational interpolation: fit a function of the form a / mu + b through the two previous
+    // iterates and use its zero to compute the next iterate
+    bool useBisection = fPrev*fCur>Literal(0);
+    while (fCur!=Literal(0) && abs(muCur - muPrev) > Literal(8) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(muCur), abs(muPrev)) && abs(fCur - fPrev)>NumTraits<RealScalar>::epsilon() && !useBisection)
+    {
+      ++m_numIters;
+
+      // Find a and b such that the function f(mu) = a / mu + b matches the current and previous samples.
+      RealScalar a = (fCur - fPrev) / (Literal(1)/muCur - Literal(1)/muPrev);
+      RealScalar b = fCur - a / muCur;
+      // And find mu such that f(mu)==0:
+      RealScalar muZero = -a/b;
+      RealScalar fZero = secularEq(muZero, col0, diag, perm, diagShifted, shift);
+      
+      muPrev = muCur;
+      fPrev = fCur;
+      muCur = muZero;
+      fCur = fZero;
+      
+      
+      if (shift == left  && (muCur < Literal(0) || muCur > right - left)) useBisection = true;
+      if (shift == right && (muCur < -(right - left) || muCur > Literal(0))) useBisection = true;
+      if (abs(fCur)>abs(fPrev)) useBisection = true;
+    }
+
+    // fall back on bisection method if rational interpolation did not work
+    if (useBisection)
+    {
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "useBisection for k = " << k << ", actual_n = " << actual_n << "\n";
+#endif
+      RealScalar leftShifted, rightShifted;
+      if (shift == left)
+      {
+        // to avoid overflow, we must have mu > max(real_min, |z(k)|/sqrt(real_max)),
+        // the factor 2 is to be more conservative
+        leftShifted = numext::maxi<RealScalar>( (std::numeric_limits<RealScalar>::min)(), Literal(2) * abs(col0(k)) / sqrt((std::numeric_limits<RealScalar>::max)()) );
+
+        // check that we did it right:
+        eigen_internal_assert( (numext::isfinite)( (col0(k)/leftShifted)*(col0(k)/(diag(k)+shift+leftShifted)) ) );
+        // I don't understand why the case k==0 would be special there:
+        // if (k == 0) rightShifted = right - left; else
+        rightShifted = (k==actual_n-1) ? right : ((right - left) * RealScalar(0.51)); // theoretically we can take 0.5, but let's be safe
+      }
+      else
+      {
+        leftShifted = -(right - left) * RealScalar(0.51);
+        if(k+1<n)
+          rightShifted = -numext::maxi<RealScalar>( (std::numeric_limits<RealScalar>::min)(), abs(col0(k+1)) / sqrt((std::numeric_limits<RealScalar>::max)()) );
+        else
+          rightShifted = -(std::numeric_limits<RealScalar>::min)();
+      }
+      
+      RealScalar fLeft = secularEq(leftShifted, col0, diag, perm, diagShifted, shift);
+
+#if defined EIGEN_INTERNAL_DEBUGGING || defined EIGEN_BDCSVD_DEBUG_VERBOSE
+      RealScalar fRight = secularEq(rightShifted, col0, diag, perm, diagShifted, shift);
+#endif
+
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      if(!(fLeft * fRight<0))
+      {
+        std::cout << "fLeft: " << leftShifted << " - " << diagShifted.head(10).transpose()  << "\n ; " << bool(left==shift) << " " << (left-shift) << "\n";
+        std::cout << k << " : " <<  fLeft << " * " << fRight << " == " << fLeft * fRight << "  ;  " << left << " - " << right << " -> " <<  leftShifted << " " << rightShifted << "   shift=" << shift << "\n";
+      }
+#endif
+      eigen_internal_assert(fLeft * fRight < Literal(0));
+      
+      while (rightShifted - leftShifted > Literal(2) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(abs(leftShifted), abs(rightShifted)))
+      {
+        RealScalar midShifted = (leftShifted + rightShifted) / Literal(2);
+        fMid = secularEq(midShifted, col0, diag, perm, diagShifted, shift);
+        if (fLeft * fMid < Literal(0))
+        {
+          rightShifted = midShifted;
+        }
+        else
+        {
+          leftShifted = midShifted;
+          fLeft = fMid;
+        }
+      }
+
+      muCur = (leftShifted + rightShifted) / Literal(2);
+    }
+      
+    singVals[k] = shift + muCur;
+    shifts[k] = shift;
+    mus[k] = muCur;
+
+    // perturb singular value slightly if it equals diagonal entry to avoid division by zero later
+    // (deflation is supposed to avoid this from happening)
+    // - this does no seem to be necessary anymore -
+//     if (singVals[k] == left) singVals[k] *= 1 + NumTraits<RealScalar>::epsilon();
+//     if (singVals[k] == right) singVals[k] *= 1 - NumTraits<RealScalar>::epsilon();
+  }
+}
+
+
+// zhat is perturbation of col0 for which singular vectors can be computed stably (see Section 3.1)
+template <typename MatrixType>
+void BDCSVD<MatrixType>::perturbCol0
+   (const ArrayRef& col0, const ArrayRef& diag, const IndicesRef &perm, const VectorType& singVals,
+    const ArrayRef& shifts, const ArrayRef& mus, ArrayRef zhat)
+{
+  using std::sqrt;
+  Index n = col0.size();
+  Index m = perm.size();
+  if(m==0)
+  {
+    zhat.setZero();
+    return;
+  }
+  Index last = perm(m-1);
+  // The offset permits to skip deflated entries while computing zhat
+  for (Index k = 0; k < n; ++k)
+  {
+    if (col0(k) == Literal(0)) // deflated
+      zhat(k) = Literal(0);
+    else
+    {
+      // see equation (3.6)
+      RealScalar dk = diag(k);
+      RealScalar prod = (singVals(last) + dk) * (mus(last) + (shifts(last) - dk));
+
+      for(Index l = 0; l<m; ++l)
+      {
+        Index i = perm(l);
+        if(i!=k)
+        {
+          Index j = i<k ? i : perm(l-1);
+          prod *= ((singVals(j)+dk) / ((diag(i)+dk))) * ((mus(j)+(shifts(j)-dk)) / ((diag(i)-dk)));
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+          if(i!=k && std::abs(((singVals(j)+dk)*(mus(j)+(shifts(j)-dk)))/((diag(i)+dk)*(diag(i)-dk)) - 1) > 0.9 )
+            std::cout << "     " << ((singVals(j)+dk)*(mus(j)+(shifts(j)-dk)))/((diag(i)+dk)*(diag(i)-dk)) << " == (" << (singVals(j)+dk) << " * " << (mus(j)+(shifts(j)-dk))
+                       << ") / (" << (diag(i)+dk) << " * " << (diag(i)-dk) << ")\n";
+#endif
+        }
+      }
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "zhat(" << k << ") =  sqrt( " << prod << ")  ;  " << (singVals(last) + dk) << " * " << mus(last) + shifts(last) << " - " << dk << "\n";
+#endif
+      RealScalar tmp = sqrt(prod);
+      zhat(k) = col0(k) > Literal(0) ? tmp : -tmp;
+    }
+  }
+}
+
+// compute singular vectors
+template <typename MatrixType>
+void BDCSVD<MatrixType>::computeSingVecs
+   (const ArrayRef& zhat, const ArrayRef& diag, const IndicesRef &perm, const VectorType& singVals,
+    const ArrayRef& shifts, const ArrayRef& mus, MatrixXr& U, MatrixXr& V)
+{
+  Index n = zhat.size();
+  Index m = perm.size();
+  
+  for (Index k = 0; k < n; ++k)
+  {
+    if (zhat(k) == Literal(0))
+    {
+      U.col(k) = VectorType::Unit(n+1, k);
+      if (m_compV) V.col(k) = VectorType::Unit(n, k);
+    }
+    else
+    {
+      U.col(k).setZero();
+      for(Index l=0;l<m;++l)
+      {
+        Index i = perm(l);
+        U(i,k) = zhat(i)/(((diag(i) - shifts(k)) - mus(k)) )/( (diag(i) + singVals[k]));
+      }
+      U(n,k) = Literal(0);
+      U.col(k).normalize();
+    
+      if (m_compV)
+      {
+        V.col(k).setZero();
+        for(Index l=1;l<m;++l)
+        {
+          Index i = perm(l);
+          V(i,k) = diag(i) * zhat(i) / (((diag(i) - shifts(k)) - mus(k)) )/( (diag(i) + singVals[k]));
+        }
+        V(0,k) = Literal(-1);
+        V.col(k).normalize();
+      }
+    }
+  }
+  U.col(n) = VectorType::Unit(n+1, n);
+}
+
+
+// page 12_13
+// i >= 1, di almost null and zi non null.
+// We use a rotation to zero out zi applied to the left of M
+template <typename MatrixType>
+void BDCSVD<MatrixType>::deflation43(Index firstCol, Index shift, Index i, Index size)
+{
+  using std::abs;
+  using std::sqrt;
+  using std::pow;
+  Index start = firstCol + shift;
+  RealScalar c = m_computed(start, start);
+  RealScalar s = m_computed(start+i, start);
+  RealScalar r = numext::hypot(c,s);
+  if (r == Literal(0))
+  {
+    m_computed(start+i, start+i) = Literal(0);
+    return;
+  }
+  m_computed(start,start) = r;  
+  m_computed(start+i, start) = Literal(0);
+  m_computed(start+i, start+i) = Literal(0);
+  
+  JacobiRotation<RealScalar> J(c/r,-s/r);
+  if (m_compU)  m_naiveU.middleRows(firstCol, size+1).applyOnTheRight(firstCol, firstCol+i, J);
+  else          m_naiveU.applyOnTheRight(firstCol, firstCol+i, J);
+}// end deflation 43
+
+
+// page 13
+// i,j >= 1, i!=j and |di - dj| < epsilon * norm2(M)
+// We apply two rotations to have zj = 0;
+// TODO deflation44 is still broken and not properly tested
+template <typename MatrixType>
+void BDCSVD<MatrixType>::deflation44(Index firstColu , Index firstColm, Index firstRowW, Index firstColW, Index i, Index j, Index size)
+{
+  using std::abs;
+  using std::sqrt;
+  using std::conj;
+  using std::pow;
+  RealScalar c = m_computed(firstColm+i, firstColm);
+  RealScalar s = m_computed(firstColm+j, firstColm);
+  RealScalar r = sqrt(numext::abs2(c) + numext::abs2(s));
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "deflation 4.4: " << i << "," << j << " -> " << c << " " << s << " " << r << " ; "
+    << m_computed(firstColm + i-1, firstColm)  << " "
+    << m_computed(firstColm + i, firstColm)  << " "
+    << m_computed(firstColm + i+1, firstColm) << " "
+    << m_computed(firstColm + i+2, firstColm) << "\n";
+  std::cout << m_computed(firstColm + i-1, firstColm + i-1)  << " "
+    << m_computed(firstColm + i, firstColm+i)  << " "
+    << m_computed(firstColm + i+1, firstColm+i+1) << " "
+    << m_computed(firstColm + i+2, firstColm+i+2) << "\n";
+#endif
+  if (r==Literal(0))
+  {
+    m_computed(firstColm + i, firstColm + i) = m_computed(firstColm + j, firstColm + j);
+    return;
+  }
+  c/=r;
+  s/=r;
+  m_computed(firstColm + i, firstColm) = r;  
+  m_computed(firstColm + j, firstColm + j) = m_computed(firstColm + i, firstColm + i);
+  m_computed(firstColm + j, firstColm) = Literal(0);
+
+  JacobiRotation<RealScalar> J(c,-s);
+  if (m_compU)  m_naiveU.middleRows(firstColu, size+1).applyOnTheRight(firstColu + i, firstColu + j, J);
+  else          m_naiveU.applyOnTheRight(firstColu+i, firstColu+j, J);
+  if (m_compV)  m_naiveV.middleRows(firstRowW, size).applyOnTheRight(firstColW + i, firstColW + j, J);
+}// end deflation 44
+
+
+// acts on block from (firstCol+shift, firstCol+shift) to (lastCol+shift, lastCol+shift) [inclusive]
+template <typename MatrixType>
+void BDCSVD<MatrixType>::deflation(Index firstCol, Index lastCol, Index k, Index firstRowW, Index firstColW, Index shift)
+{
+  using std::sqrt;
+  using std::abs;
+  const Index length = lastCol + 1 - firstCol;
+  
+  Block<MatrixXr,Dynamic,1> col0(m_computed, firstCol+shift, firstCol+shift, length, 1);
+  Diagonal<MatrixXr> fulldiag(m_computed);
+  VectorBlock<Diagonal<MatrixXr>,Dynamic> diag(fulldiag, firstCol+shift, length);
+  
+  const RealScalar considerZero = (std::numeric_limits<RealScalar>::min)();
+  RealScalar maxDiag = diag.tail((std::max)(Index(1),length-1)).cwiseAbs().maxCoeff();
+  RealScalar epsilon_strict = numext::maxi<RealScalar>(considerZero,NumTraits<RealScalar>::epsilon() * maxDiag);
+  RealScalar epsilon_coarse = Literal(8) * NumTraits<RealScalar>::epsilon() * numext::maxi<RealScalar>(col0.cwiseAbs().maxCoeff(), maxDiag);
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE  
+  std::cout << "\ndeflate:" << diag.head(k+1).transpose() << "  |  " << diag.segment(k+1,length-k-1).transpose() << "\n";
+#endif
+  
+  //condition 4.1
+  if (diag(0) < epsilon_coarse)
+  { 
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+    std::cout << "deflation 4.1, because " << diag(0) << " < " << epsilon_coarse << "\n";
+#endif
+    diag(0) = epsilon_coarse;
+  }
+
+  //condition 4.2
+  for (Index i=1;i<length;++i)
+    if (abs(col0(i)) < epsilon_strict)
+    {
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "deflation 4.2, set z(" << i << ") to zero because " << abs(col0(i)) << " < " << epsilon_strict << "  (diag(" << i << ")=" << diag(i) << ")\n";
+#endif
+      col0(i) = Literal(0);
+    }
+
+  //condition 4.3
+  for (Index i=1;i<length; i++)
+    if (diag(i) < epsilon_coarse)
+    {
+#ifdef  EIGEN_BDCSVD_DEBUG_VERBOSE
+      std::cout << "deflation 4.3, cancel z(" << i << ")=" << col0(i) << " because diag(" << i << ")=" << diag(i) << " < " << epsilon_coarse << "\n";
+#endif
+      deflation43(firstCol, shift, i, length);
+    }
+
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "to be sorted: " << diag.transpose() << "\n\n";
+#endif
+  {
+    // Check for total deflation
+    // If we have a total deflation, then we have to consider col0(0)==diag(0) as a singular value during sorting
+    bool total_deflation = (col0.tail(length-1).array()<considerZero).all();
+    
+    // Sort the diagonal entries, since diag(1:k-1) and diag(k:length) are already sorted, let's do a sorted merge.
+    // First, compute the respective permutation.
+    Index *permutation = m_workspaceI.data();
+    {
+      permutation[0] = 0;
+      Index p = 1;
+      
+      // Move deflated diagonal entries at the end.
+      for(Index i=1; i<length; ++i)
+        if(abs(diag(i))<considerZero)
+          permutation[p++] = i;
+        
+      Index i=1, j=k+1;
+      for( ; p < length; ++p)
+      {
+             if (i > k)             permutation[p] = j++;
+        else if (j >= length)       permutation[p] = i++;
+        else if (diag(i) < diag(j)) permutation[p] = j++;
+        else                        permutation[p] = i++;
+      }
+    }
+    
+    // If we have a total deflation, then we have to insert diag(0) at the right place
+    if(total_deflation)
+    {
+      for(Index i=1; i<length; ++i)
+      {
+        Index pi = permutation[i];
+        if(abs(diag(pi))<considerZero || diag(0)<diag(pi))
+          permutation[i-1] = permutation[i];
+        else
+        {
+          permutation[i-1] = 0;
+          break;
+        }
+      }
+    }
+    
+    // Current index of each col, and current column of each index
+    Index *realInd = m_workspaceI.data()+length;
+    Index *realCol = m_workspaceI.data()+2*length;
+    
+    for(int pos = 0; pos< length; pos++)
+    {
+      realCol[pos] = pos;
+      realInd[pos] = pos;
+    }
+    
+    for(Index i = total_deflation?0:1; i < length; i++)
+    {
+      const Index pi = permutation[length - (total_deflation ? i+1 : i)];
+      const Index J = realCol[pi];
+      
+      using std::swap;
+      // swap diagonal and first column entries:
+      swap(diag(i), diag(J));
+      if(i!=0 && J!=0) swap(col0(i), col0(J));
+
+      // change columns
+      if (m_compU) m_naiveU.col(firstCol+i).segment(firstCol, length + 1).swap(m_naiveU.col(firstCol+J).segment(firstCol, length + 1));
+      else         m_naiveU.col(firstCol+i).segment(0, 2)                .swap(m_naiveU.col(firstCol+J).segment(0, 2));
+      if (m_compV) m_naiveV.col(firstColW + i).segment(firstRowW, length).swap(m_naiveV.col(firstColW + J).segment(firstRowW, length));
+
+      //update real pos
+      const Index realI = realInd[i];
+      realCol[realI] = J;
+      realCol[pi] = i;
+      realInd[J] = realI;
+      realInd[i] = pi;
+    }
+  }
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+  std::cout << "sorted: " << diag.transpose().format(bdcsvdfmt) << "\n";
+  std::cout << "      : " << col0.transpose() << "\n\n";
+#endif
+    
+  //condition 4.4
+  {
+    Index i = length-1;
+    while(i>0 && (abs(diag(i))<considerZero || abs(col0(i))<considerZero)) --i;
+    for(; i>1;--i)
+       if( (diag(i) - diag(i-1)) < NumTraits<RealScalar>::epsilon()*maxDiag )
+      {
+#ifdef EIGEN_BDCSVD_DEBUG_VERBOSE
+        std::cout << "deflation 4.4 with i = " << i << " because " << (diag(i) - diag(i-1)) << " < " << NumTraits<RealScalar>::epsilon()*diag(i) << "\n";
+#endif
+        eigen_internal_assert(abs(diag(i) - diag(i-1))<epsilon_coarse && " diagonal entries are not properly sorted");
+        deflation44(firstCol, firstCol + shift, firstRowW, firstColW, i-1, i, length);
+      }
+  }
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  for(Index j=2;j<length;++j)
+    assert(diag(j-1)<=diag(j) || abs(diag(j))<considerZero);
+#endif
+  
+#ifdef EIGEN_BDCSVD_SANITY_CHECKS
+  assert(m_naiveU.allFinite());
+  assert(m_naiveV.allFinite());
+  assert(m_computed.allFinite());
+#endif
+}//end deflation
+
+#ifndef __CUDACC__
+/** \svd_module
+  *
+  * \return the singular value decomposition of \c *this computed by Divide & Conquer algorithm
+  *
+  * \sa class BDCSVD
+  */
+template<typename Derived>
+BDCSVD<typename MatrixBase<Derived>::PlainObject>
+MatrixBase<Derived>::bdcSvd(unsigned int computationOptions) const
+{
+  return BDCSVD<PlainObject>(*this, computationOptions);
+}
+#endif
+
+} // end namespace Eigen
+
+#endif
diff --git a/Eigen/src/SVD/CMakeLists.txt b/Eigen/src/SVD/CMakeLists.txt
deleted file mode 100644
index 55efc44..0000000
--- a/Eigen/src/SVD/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_SVD_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_SVD_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/SVD COMPONENT Devel
-  )
diff --git a/Eigen/src/SVD/JacobiSVD.h b/Eigen/src/SVD/JacobiSVD.h
index 1b29774..43488b1 100644
--- a/Eigen/src/SVD/JacobiSVD.h
+++ b/Eigen/src/SVD/JacobiSVD.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2009-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2013-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -51,7 +52,6 @@
 class qr_preconditioner_impl<MatrixType, QRPreconditioner, Case, false>
 {
 public:
-  typedef typename MatrixType::Index Index;
   void allocate(const JacobiSVD<MatrixType, QRPreconditioner>&) {}
   bool run(JacobiSVD<MatrixType, QRPreconditioner>&, const MatrixType&)
   {
@@ -65,7 +65,6 @@
 class qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -106,7 +105,6 @@
 class qr_preconditioner_impl<MatrixType, FullPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -114,9 +112,11 @@
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-    Options = MatrixType::Options
+    TrOptions = RowsAtCompileTime==1 ? (MatrixType::Options & ~(RowMajor))
+              : ColsAtCompileTime==1 ? (MatrixType::Options |   RowMajor)
+              : MatrixType::Options
   };
-  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
+  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, TrOptions, MaxColsAtCompileTime, MaxRowsAtCompileTime>
           TransposeTypeWithSameStorageOrder;
 
   void allocate(const JacobiSVD<MatrixType, FullPivHouseholderQRPreconditioner>& svd)
@@ -156,8 +156,6 @@
 class qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
-
   void allocate(const JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd)
   {
     if (svd.rows() != m_qr.rows() || svd.cols() != m_qr.cols())
@@ -197,7 +195,6 @@
 class qr_preconditioner_impl<MatrixType, ColPivHouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -205,10 +202,12 @@
     ColsAtCompileTime = MatrixType::ColsAtCompileTime,
     MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
     MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
-    Options = MatrixType::Options
+    TrOptions = RowsAtCompileTime==1 ? (MatrixType::Options & ~(RowMajor))
+              : ColsAtCompileTime==1 ? (MatrixType::Options |   RowMajor)
+              : MatrixType::Options
   };
 
-  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, Options, MaxColsAtCompileTime, MaxRowsAtCompileTime>
+  typedef Matrix<Scalar, ColsAtCompileTime, RowsAtCompileTime, TrOptions, MaxColsAtCompileTime, MaxRowsAtCompileTime>
           TransposeTypeWithSameStorageOrder;
 
   void allocate(const JacobiSVD<MatrixType, ColPivHouseholderQRPreconditioner>& svd)
@@ -256,8 +255,6 @@
 class qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreRowsThanCols, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
-
   void allocate(const JacobiSVD<MatrixType, HouseholderQRPreconditioner>& svd)
   {
     if (svd.rows() != m_qr.rows() || svd.cols() != m_qr.cols())
@@ -296,7 +293,6 @@
 class qr_preconditioner_impl<MatrixType, HouseholderQRPreconditioner, PreconditionIfMoreColsThanRows, true>
 {
 public:
-  typedef typename MatrixType::Index Index;
   typedef typename MatrixType::Scalar Scalar;
   enum
   {
@@ -358,8 +354,8 @@
 struct svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner, false>
 {
   typedef JacobiSVD<MatrixType, QRPreconditioner> SVD;
-  typedef typename SVD::Index Index;
-  static void run(typename SVD::WorkMatrixType&, SVD&, Index, Index) {}
+  typedef typename MatrixType::RealScalar RealScalar;
+  static bool run(typename SVD::WorkMatrixType&, SVD&, Index, Index, RealScalar&) { return true; }
 };
 
 template<typename MatrixType, int QRPreconditioner>
@@ -368,20 +364,30 @@
   typedef JacobiSVD<MatrixType, QRPreconditioner> SVD;
   typedef typename MatrixType::Scalar Scalar;
   typedef typename MatrixType::RealScalar RealScalar;
-  typedef typename SVD::Index Index;
-  static void run(typename SVD::WorkMatrixType& work_matrix, SVD& svd, Index p, Index q)
+  static bool run(typename SVD::WorkMatrixType& work_matrix, SVD& svd, Index p, Index q, RealScalar& maxDiagEntry)
   {
     using std::sqrt;
+    using std::abs;
     Scalar z;
     JacobiRotation<Scalar> rot;
     RealScalar n = sqrt(numext::abs2(work_matrix.coeff(p,p)) + numext::abs2(work_matrix.coeff(q,p)));
-    
+
+    const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
+    const RealScalar precision = NumTraits<Scalar>::epsilon();
+
     if(n==0)
     {
-      z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
-      work_matrix.row(p) *= z;
-      if(svd.computeU()) svd.m_matrixU.col(p) *= conj(z);
-      if(work_matrix.coeff(q,q)!=Scalar(0))
+      // make sure first column is zero
+      work_matrix.coeffRef(p,p) = work_matrix.coeffRef(q,p) = Scalar(0);
+
+      if(abs(numext::imag(work_matrix.coeff(p,q)))>considerAsZero)
+      {
+        // work_matrix.coeff(p,q) can be zero if work_matrix.coeff(q,p) is not zero but small enough to underflow when computing n
+        z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
+        work_matrix.row(p) *= z;
+        if(svd.computeU()) svd.m_matrixU.col(p) *= conj(z);
+      }
+      if(abs(numext::imag(work_matrix.coeff(q,q)))>considerAsZero)
       {
         z = abs(work_matrix.coeff(q,q)) / work_matrix.coeff(q,q);
         work_matrix.row(q) *= z;
@@ -395,52 +401,33 @@
       rot.s() = work_matrix.coeff(q,p) / n;
       work_matrix.applyOnTheLeft(p,q,rot);
       if(svd.computeU()) svd.m_matrixU.applyOnTheRight(p,q,rot.adjoint());
-      if(work_matrix.coeff(p,q) != Scalar(0))
+      if(abs(numext::imag(work_matrix.coeff(p,q)))>considerAsZero)
       {
-        Scalar z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
+        z = abs(work_matrix.coeff(p,q)) / work_matrix.coeff(p,q);
         work_matrix.col(q) *= z;
         if(svd.computeV()) svd.m_matrixV.col(q) *= z;
       }
-      if(work_matrix.coeff(q,q) != Scalar(0))
+      if(abs(numext::imag(work_matrix.coeff(q,q)))>considerAsZero)
       {
         z = abs(work_matrix.coeff(q,q)) / work_matrix.coeff(q,q);
         work_matrix.row(q) *= z;
         if(svd.computeU()) svd.m_matrixU.col(q) *= conj(z);
       }
     }
+
+    // update largest diagonal entry
+    maxDiagEntry = numext::maxi<RealScalar>(maxDiagEntry,numext::maxi<RealScalar>(abs(work_matrix.coeff(p,p)), abs(work_matrix.coeff(q,q))));
+    // and check whether the 2x2 block is already diagonal
+    RealScalar threshold = numext::maxi<RealScalar>(considerAsZero, precision * maxDiagEntry);
+    return abs(work_matrix.coeff(p,q))>threshold || abs(work_matrix.coeff(q,p)) > threshold;
   }
 };
 
-template<typename MatrixType, typename RealScalar, typename Index>
-void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
-                            JacobiRotation<RealScalar> *j_left,
-                            JacobiRotation<RealScalar> *j_right)
+template<typename _MatrixType, int QRPreconditioner> 
+struct traits<JacobiSVD<_MatrixType,QRPreconditioner> >
 {
-  using std::sqrt;
-  using std::abs;
-  Matrix<RealScalar,2,2> m;
-  m << numext::real(matrix.coeff(p,p)), numext::real(matrix.coeff(p,q)),
-       numext::real(matrix.coeff(q,p)), numext::real(matrix.coeff(q,q));
-  JacobiRotation<RealScalar> rot1;
-  RealScalar t = m.coeff(0,0) + m.coeff(1,1);
-  RealScalar d = m.coeff(1,0) - m.coeff(0,1);
-  if(t == RealScalar(0))
-  {
-    rot1.c() = RealScalar(0);
-    rot1.s() = d > RealScalar(0) ? RealScalar(1) : RealScalar(-1);
-  }
-  else
-  {
-    RealScalar t2d2 = numext::hypot(t,d);
-    rot1.c() = abs(t)/t2d2;
-    rot1.s() = d/t2d2;
-    if(t<RealScalar(0))
-      rot1.s() = -rot1.s();
-  }
-  m.applyOnTheLeft(0,1,rot1);
-  j_right->makeJacobi(m,0,1);
-  *j_left  = rot1 * j_right->transpose();
-}
+  typedef _MatrixType MatrixType;
+};
 
 } // end namespace internal
 
@@ -451,8 +438,8 @@
   *
   * \brief Two-sided Jacobi SVD decomposition of a rectangular matrix
   *
-  * \param MatrixType the type of the matrix of which we are computing the SVD decomposition
-  * \param QRPreconditioner this optional parameter allows to specify the type of QR decomposition that will be used internally
+  * \tparam _MatrixType the type of the matrix of which we are computing the SVD decomposition
+  * \tparam QRPreconditioner this optional parameter allows to specify the type of QR decomposition that will be used internally
   *                        for the R-SVD step for non-square matrices. See discussion of possible values below.
   *
   * SVD decomposition consists in decomposing any n-by-p matrix \a A as a product
@@ -498,13 +485,14 @@
   * \sa MatrixBase::jacobiSvd()
   */
 template<typename _MatrixType, int QRPreconditioner> class JacobiSVD
+ : public SVDBase<JacobiSVD<_MatrixType,QRPreconditioner> >
 {
+    typedef SVDBase<JacobiSVD> Base;
   public:
 
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
-    typedef typename MatrixType::Index Index;
     enum {
       RowsAtCompileTime = MatrixType::RowsAtCompileTime,
       ColsAtCompileTime = MatrixType::ColsAtCompileTime,
@@ -515,13 +503,10 @@
       MatrixOptions = MatrixType::Options
     };
 
-    typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime,
-                   MatrixOptions, MaxRowsAtCompileTime, MaxRowsAtCompileTime>
-            MatrixUType;
-    typedef Matrix<Scalar, ColsAtCompileTime, ColsAtCompileTime,
-                   MatrixOptions, MaxColsAtCompileTime, MaxColsAtCompileTime>
-            MatrixVType;
-    typedef typename internal::plain_diag_type<MatrixType, RealScalar>::type SingularValuesType;
+    typedef typename Base::MatrixUType MatrixUType;
+    typedef typename Base::MatrixVType MatrixVType;
+    typedef typename Base::SingularValuesType SingularValuesType;
+    
     typedef typename internal::plain_row_type<MatrixType>::type RowType;
     typedef typename internal::plain_col_type<MatrixType>::type ColType;
     typedef Matrix<Scalar, DiagSizeAtCompileTime, DiagSizeAtCompileTime,
@@ -534,11 +519,6 @@
       * perform decompositions via JacobiSVD::compute(const MatrixType&).
       */
     JacobiSVD()
-      : m_isInitialized(false),
-        m_isAllocated(false),
-        m_usePrescribedThreshold(false),
-        m_computationOptions(0),
-        m_rows(-1), m_cols(-1), m_diagSize(0)
     {}
 
 
@@ -549,11 +529,6 @@
       * \sa JacobiSVD()
       */
     JacobiSVD(Index rows, Index cols, unsigned int computationOptions = 0)
-      : m_isInitialized(false),
-        m_isAllocated(false),
-        m_usePrescribedThreshold(false),
-        m_computationOptions(0),
-        m_rows(-1), m_cols(-1)
     {
       allocate(rows, cols, computationOptions);
     }
@@ -568,12 +543,7 @@
      * Thin unitaries are only available if your matrix type has a Dynamic number of columns (for example MatrixXf). They also are not
      * available with the (non-default) FullPivHouseholderQR preconditioner.
      */
-    JacobiSVD(const MatrixType& matrix, unsigned int computationOptions = 0)
-      : m_isInitialized(false),
-        m_isAllocated(false),
-        m_usePrescribedThreshold(false),
-        m_computationOptions(0),
-        m_rows(-1), m_cols(-1)
+    explicit JacobiSVD(const MatrixType& matrix, unsigned int computationOptions = 0)
     {
       compute(matrix, computationOptions);
     }
@@ -601,164 +571,33 @@
       return compute(matrix, m_computationOptions);
     }
 
-    /** \returns the \a U matrix.
-     *
-     * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
-     * the U matrix is n-by-n if you asked for #ComputeFullU, and is n-by-m if you asked for #ComputeThinU.
-     *
-     * The \a m first columns of \a U are the left singular vectors of the matrix being decomposed.
-     *
-     * This method asserts that you asked for \a U to be computed.
-     */
-    const MatrixUType& matrixU() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      eigen_assert(computeU() && "This JacobiSVD decomposition didn't compute U. Did you ask for it?");
-      return m_matrixU;
-    }
-
-    /** \returns the \a V matrix.
-     *
-     * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
-     * the V matrix is p-by-p if you asked for #ComputeFullV, and is p-by-m if you asked for ComputeThinV.
-     *
-     * The \a m first columns of \a V are the right singular vectors of the matrix being decomposed.
-     *
-     * This method asserts that you asked for \a V to be computed.
-     */
-    const MatrixVType& matrixV() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      eigen_assert(computeV() && "This JacobiSVD decomposition didn't compute V. Did you ask for it?");
-      return m_matrixV;
-    }
-
-    /** \returns the vector of singular values.
-     *
-     * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p, the
-     * returned vector has size \a m.  Singular values are always sorted in decreasing order.
-     */
-    const SingularValuesType& singularValues() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      return m_singularValues;
-    }
-
-    /** \returns true if \a U (full or thin) is asked for in this SVD decomposition */
-    inline bool computeU() const { return m_computeFullU || m_computeThinU; }
-    /** \returns true if \a V (full or thin) is asked for in this SVD decomposition */
-    inline bool computeV() const { return m_computeFullV || m_computeThinV; }
-
-    /** \returns a (least squares) solution of \f$ A x = b \f$ using the current SVD decomposition of A.
-      *
-      * \param b the right-hand-side of the equation to solve.
-      *
-      * \note Solving requires both U and V to be computed. Thin U and V are enough, there is no need for full U or V.
-      *
-      * \note SVD solving is implicitly least-squares. Thus, this method serves both purposes of exact solving and least-squares solving.
-      * In other words, the returned solution is guaranteed to minimize the Euclidean norm \f$ \Vert A x - b \Vert \f$.
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<JacobiSVD, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      eigen_assert(computeU() && computeV() && "JacobiSVD::solve() requires both unitaries U and V to be computed (thin unitaries suffice).");
-      return internal::solve_retval<JacobiSVD, Rhs>(*this, b.derived());
-    }
-
-    /** \returns the number of singular values that are not exactly 0 */
-    Index nonzeroSingularValues() const
-    {
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      return m_nonzeroSingularValues;
-    }
-    
-    /** \returns the rank of the matrix of which \c *this is the SVD.
-      *
-      * \note This method has to determine which singular values should be considered nonzero.
-      *       For that, it uses the threshold value that you can control by calling
-      *       setThreshold(const RealScalar&).
-      */
-    inline Index rank() const
-    {
-      using std::abs;
-      eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
-      if(m_singularValues.size()==0) return 0;
-      RealScalar premultiplied_threshold = m_singularValues.coeff(0) * threshold();
-      Index i = m_nonzeroSingularValues-1;
-      while(i>=0 && m_singularValues.coeff(i) < premultiplied_threshold) --i;
-      return i+1;
-    }
-    
-    /** Allows to prescribe a threshold to be used by certain methods, such as rank() and solve(),
-      * which need to determine when singular values are to be considered nonzero.
-      * This is not used for the SVD decomposition itself.
-      *
-      * When it needs to get the threshold value, Eigen calls threshold().
-      * The default is \c NumTraits<Scalar>::epsilon()
-      *
-      * \param threshold The new value to use as the threshold.
-      *
-      * A singular value will be considered nonzero if its value is strictly greater than
-      *  \f$ \vert singular value \vert \leqslant threshold \times \vert max singular value \vert \f$.
-      *
-      * If you want to come back to the default behavior, call setThreshold(Default_t)
-      */
-    JacobiSVD& setThreshold(const RealScalar& threshold)
-    {
-      m_usePrescribedThreshold = true;
-      m_prescribedThreshold = threshold;
-      return *this;
-    }
-
-    /** Allows to come back to the default behavior, letting Eigen use its default formula for
-      * determining the threshold.
-      *
-      * You should pass the special object Eigen::Default as parameter here.
-      * \code svd.setThreshold(Eigen::Default); \endcode
-      *
-      * See the documentation of setThreshold(const RealScalar&).
-      */
-    JacobiSVD& setThreshold(Default_t)
-    {
-      m_usePrescribedThreshold = false;
-      return *this;
-    }
-
-    /** Returns the threshold that will be used by certain methods such as rank().
-      *
-      * See the documentation of setThreshold(const RealScalar&).
-      */
-    RealScalar threshold() const
-    {
-      eigen_assert(m_isInitialized || m_usePrescribedThreshold);
-      return m_usePrescribedThreshold ? m_prescribedThreshold
-                                      : (std::max<Index>)(1,m_diagSize)*NumTraits<Scalar>::epsilon();
-    }
-
-    inline Index rows() const { return m_rows; }
-    inline Index cols() const { return m_cols; }
+    using Base::computeU;
+    using Base::computeV;
+    using Base::rows;
+    using Base::cols;
+    using Base::rank;
 
   private:
     void allocate(Index rows, Index cols, unsigned int computationOptions);
-    
-    static void check_template_parameters()
-    {
-      EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
-    }
 
   protected:
-    MatrixUType m_matrixU;
-    MatrixVType m_matrixV;
-    SingularValuesType m_singularValues;
+    using Base::m_matrixU;
+    using Base::m_matrixV;
+    using Base::m_singularValues;
+    using Base::m_isInitialized;
+    using Base::m_isAllocated;
+    using Base::m_usePrescribedThreshold;
+    using Base::m_computeFullU;
+    using Base::m_computeThinU;
+    using Base::m_computeFullV;
+    using Base::m_computeThinV;
+    using Base::m_computationOptions;
+    using Base::m_nonzeroSingularValues;
+    using Base::m_rows;
+    using Base::m_cols;
+    using Base::m_diagSize;
+    using Base::m_prescribedThreshold;
     WorkMatrixType m_workMatrix;
-    bool m_isInitialized, m_isAllocated, m_usePrescribedThreshold;
-    bool m_computeFullU, m_computeThinU;
-    bool m_computeFullV, m_computeThinV;
-    unsigned int m_computationOptions;
-    Index m_nonzeroSingularValues, m_rows, m_cols, m_diagSize;
-    RealScalar m_prescribedThreshold;
 
     template<typename __MatrixType, int _QRPreconditioner, bool _IsComplex>
     friend struct internal::svd_precondition_2x2_block_to_be_real;
@@ -816,15 +655,13 @@
   
   if(m_cols>m_rows)   m_qr_precond_morecols.allocate(*this);
   if(m_rows>m_cols)   m_qr_precond_morerows.allocate(*this);
-  if(m_cols!=m_cols)  m_scaledMatrix.resize(rows,cols);
+  if(m_rows!=m_cols)  m_scaledMatrix.resize(rows,cols);
 }
 
 template<typename MatrixType, int QRPreconditioner>
 JacobiSVD<MatrixType, QRPreconditioner>&
 JacobiSVD<MatrixType, QRPreconditioner>::compute(const MatrixType& matrix, unsigned int computationOptions)
 {
-  check_template_parameters();
-  
   using std::abs;
   allocate(matrix.rows(), matrix.cols(), computationOptions);
 
@@ -832,8 +669,8 @@
   // only worsening the precision of U and V as we accumulate more rotations
   const RealScalar precision = RealScalar(2) * NumTraits<Scalar>::epsilon();
 
-  // limit for very small denormal numbers to be considered zero in order to avoid infinite loops (see bug 286)
-  const RealScalar considerAsZero = RealScalar(2) * std::numeric_limits<RealScalar>::denorm_min();
+  // limit for denormal numbers to be considered zero in order to avoid infinite loops (see bug 286)
+  const RealScalar considerAsZero = (std::numeric_limits<RealScalar>::min)();
 
   // Scaling factor to reduce over/under-flows
   RealScalar scale = matrix.cwiseAbs().maxCoeff();
@@ -857,6 +694,7 @@
   }
 
   /*** step 2. The main Jacobi SVD iteration. ***/
+  RealScalar maxDiagEntry = m_workMatrix.cwiseAbs().diagonal().maxCoeff();
 
   bool finished = false;
   while(!finished)
@@ -872,25 +710,27 @@
         // if this 2x2 sub-matrix is not diagonal already...
         // notice that this comparison will evaluate to false if any NaN is involved, ensuring that NaN's don't
         // keep us iterating forever. Similarly, small denormal numbers are considered zero.
-        using std::max;
-        RealScalar threshold = (max)(considerAsZero, precision * (max)(abs(m_workMatrix.coeff(p,p)),
-                                                                       abs(m_workMatrix.coeff(q,q))));
-        // We compare both values to threshold instead of calling max to be robust to NaN (See bug 791)
+        RealScalar threshold = numext::maxi<RealScalar>(considerAsZero, precision * maxDiagEntry);
         if(abs(m_workMatrix.coeff(p,q))>threshold || abs(m_workMatrix.coeff(q,p)) > threshold)
         {
           finished = false;
-
           // perform SVD decomposition of 2x2 sub-matrix corresponding to indices p,q to make it diagonal
-          internal::svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner>::run(m_workMatrix, *this, p, q);
-          JacobiRotation<RealScalar> j_left, j_right;
-          internal::real_2x2_jacobi_svd(m_workMatrix, p, q, &j_left, &j_right);
+          // the complex to real operation returns true if the updated 2x2 block is not already diagonal
+          if(internal::svd_precondition_2x2_block_to_be_real<MatrixType, QRPreconditioner>::run(m_workMatrix, *this, p, q, maxDiagEntry))
+          {
+            JacobiRotation<RealScalar> j_left, j_right;
+            internal::real_2x2_jacobi_svd(m_workMatrix, p, q, &j_left, &j_right);
 
-          // accumulate resulting Jacobi rotations
-          m_workMatrix.applyOnTheLeft(p,q,j_left);
-          if(computeU()) m_matrixU.applyOnTheRight(p,q,j_left.transpose());
+            // accumulate resulting Jacobi rotations
+            m_workMatrix.applyOnTheLeft(p,q,j_left);
+            if(computeU()) m_matrixU.applyOnTheRight(p,q,j_left.transpose());
 
-          m_workMatrix.applyOnTheRight(p,q,j_right);
-          if(computeV()) m_matrixV.applyOnTheRight(p,q,j_right);
+            m_workMatrix.applyOnTheRight(p,q,j_right);
+            if(computeV()) m_matrixV.applyOnTheRight(p,q,j_right);
+
+            // keep track of the largest diagonal coefficient
+            maxDiagEntry = numext::maxi<RealScalar>(maxDiagEntry,numext::maxi<RealScalar>(abs(m_workMatrix.coeff(p,p)), abs(m_workMatrix.coeff(q,q))));
+          }
         }
       }
     }
@@ -900,10 +740,25 @@
 
   for(Index i = 0; i < m_diagSize; ++i)
   {
-    RealScalar a = abs(m_workMatrix.coeff(i,i));
-    m_singularValues.coeffRef(i) = a;
-    if(computeU() && (a!=RealScalar(0))) m_matrixU.col(i) *= m_workMatrix.coeff(i,i)/a;
+    // For a complex matrix, some diagonal coefficients might note have been
+    // treated by svd_precondition_2x2_block_to_be_real, and the imaginary part
+    // of some diagonal entry might not be null.
+    if(NumTraits<Scalar>::IsComplex && abs(numext::imag(m_workMatrix.coeff(i,i)))>considerAsZero)
+    {
+      RealScalar a = abs(m_workMatrix.coeff(i,i));
+      m_singularValues.coeffRef(i) = abs(a);
+      if(computeU()) m_matrixU.col(i) *= m_workMatrix.coeff(i,i)/a;
+    }
+    else
+    {
+      // m_workMatrix.coeff(i,i) is already real, no difficulty:
+      RealScalar a = numext::real(m_workMatrix.coeff(i,i));
+      m_singularValues.coeffRef(i) = abs(a);
+      if(computeU() && (a<RealScalar(0))) m_matrixU.col(i) = -m_matrixU.col(i);
+    }
   }
+  
+  m_singularValues *= scale;
 
   /*** step 4. Sort singular values in descending order and compute the number of nonzero singular values ***/
 
@@ -925,38 +780,11 @@
       if(computeV()) m_matrixV.col(pos).swap(m_matrixV.col(i));
     }
   }
-  
-  m_singularValues *= scale;
 
   m_isInitialized = true;
   return *this;
 }
 
-namespace internal {
-template<typename _MatrixType, int QRPreconditioner, typename Rhs>
-struct solve_retval<JacobiSVD<_MatrixType, QRPreconditioner>, Rhs>
-  : solve_retval_base<JacobiSVD<_MatrixType, QRPreconditioner>, Rhs>
-{
-  typedef JacobiSVD<_MatrixType, QRPreconditioner> JacobiSVDType;
-  EIGEN_MAKE_SOLVE_HELPERS(JacobiSVDType,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    eigen_assert(rhs().rows() == dec().rows());
-
-    // A = U S V^*
-    // So A^{-1} = V S^{-1} U^*
-
-    Matrix<Scalar, Dynamic, Rhs::ColsAtCompileTime, 0, _MatrixType::MaxRowsAtCompileTime, Rhs::MaxColsAtCompileTime> tmp;
-    Index rank = dec().rank();
-    
-    tmp.noalias() = dec().matrixU().leftCols(rank).adjoint() * rhs();
-    tmp = dec().singularValues().head(rank).asDiagonal().inverse() * tmp;
-    dst = dec().matrixV().leftCols(rank) * tmp;
-  }
-};
-} // end namespace internal
-
 /** \svd_module
   *
   * \return the singular value decomposition of \c *this computed by two-sided
diff --git a/Eigen/src/SVD/JacobiSVD_LAPACKE.h b/Eigen/src/SVD/JacobiSVD_LAPACKE.h
new file mode 100644
index 0000000..ff0516f
--- /dev/null
+++ b/Eigen/src/SVD/JacobiSVD_LAPACKE.h
@@ -0,0 +1,91 @@
+/*
+ Copyright (c) 2011, Intel Corporation. All rights reserved.
+
+ Redistribution and use in source and binary forms, with or without modification,
+ are permitted provided that the following conditions are met:
+
+ * Redistributions of source code must retain the above copyright notice, this
+   list of conditions and the following disclaimer.
+ * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+ * Neither the name of Intel Corporation nor the names of its contributors may
+   be used to endorse or promote products derived from this software without
+   specific prior written permission.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+ ********************************************************************************
+ *   Content : Eigen bindings to LAPACKe
+ *    Singular Value Decomposition - SVD.
+ ********************************************************************************
+*/
+
+#ifndef EIGEN_JACOBISVD_LAPACKE_H
+#define EIGEN_JACOBISVD_LAPACKE_H
+
+namespace Eigen { 
+
+/** \internal Specialization for the data types supported by LAPACKe */
+
+#define EIGEN_LAPACKE_SVD(EIGTYPE, LAPACKE_TYPE, LAPACKE_RTYPE, LAPACKE_PREFIX, EIGCOLROW, LAPACKE_COLROW) \
+template<> inline \
+JacobiSVD<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>, ColPivHouseholderQRPreconditioner>& \
+JacobiSVD<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>, ColPivHouseholderQRPreconditioner>::compute(const Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>& matrix, unsigned int computationOptions) \
+{ \
+  typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> MatrixType; \
+  /*typedef MatrixType::Scalar Scalar;*/ \
+  /*typedef MatrixType::RealScalar RealScalar;*/ \
+  allocate(matrix.rows(), matrix.cols(), computationOptions); \
+\
+  /*const RealScalar precision = RealScalar(2) * NumTraits<Scalar>::epsilon();*/ \
+  m_nonzeroSingularValues = m_diagSize; \
+\
+  lapack_int lda = internal::convert_index<lapack_int>(matrix.outerStride()), ldu, ldvt; \
+  lapack_int matrix_order = LAPACKE_COLROW; \
+  char jobu, jobvt; \
+  LAPACKE_TYPE *u, *vt, dummy; \
+  jobu  = (m_computeFullU) ? 'A' : (m_computeThinU) ? 'S' : 'N'; \
+  jobvt = (m_computeFullV) ? 'A' : (m_computeThinV) ? 'S' : 'N'; \
+  if (computeU()) { \
+    ldu  = internal::convert_index<lapack_int>(m_matrixU.outerStride()); \
+    u    = (LAPACKE_TYPE*)m_matrixU.data(); \
+  } else { ldu=1; u=&dummy; }\
+  MatrixType localV; \
+  lapack_int vt_rows = (m_computeFullV) ? internal::convert_index<lapack_int>(m_cols) : (m_computeThinV) ? internal::convert_index<lapack_int>(m_diagSize) : 1; \
+  if (computeV()) { \
+    localV.resize(vt_rows, m_cols); \
+    ldvt  = internal::convert_index<lapack_int>(localV.outerStride()); \
+    vt   = (LAPACKE_TYPE*)localV.data(); \
+  } else { ldvt=1; vt=&dummy; }\
+  Matrix<LAPACKE_RTYPE, Dynamic, Dynamic> superb; superb.resize(m_diagSize, 1); \
+  MatrixType m_temp; m_temp = matrix; \
+  LAPACKE_##LAPACKE_PREFIX##gesvd( matrix_order, jobu, jobvt, internal::convert_index<lapack_int>(m_rows), internal::convert_index<lapack_int>(m_cols), (LAPACKE_TYPE*)m_temp.data(), lda, (LAPACKE_RTYPE*)m_singularValues.data(), u, ldu, vt, ldvt, superb.data()); \
+  if (computeV()) m_matrixV = localV.adjoint(); \
+ /* for(int i=0;i<m_diagSize;i++) if (m_singularValues.coeffRef(i) < precision) { m_nonzeroSingularValues--; m_singularValues.coeffRef(i)=RealScalar(0);}*/ \
+  m_isInitialized = true; \
+  return *this; \
+}
+
+EIGEN_LAPACKE_SVD(double,   double,                double, d, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SVD(float,    float,                 float , s, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SVD(dcomplex, lapack_complex_double, double, z, ColMajor, LAPACK_COL_MAJOR)
+EIGEN_LAPACKE_SVD(scomplex, lapack_complex_float,  float , c, ColMajor, LAPACK_COL_MAJOR)
+
+EIGEN_LAPACKE_SVD(double,   double,                double, d, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SVD(float,    float,                 float , s, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SVD(dcomplex, lapack_complex_double, double, z, RowMajor, LAPACK_ROW_MAJOR)
+EIGEN_LAPACKE_SVD(scomplex, lapack_complex_float,  float , c, RowMajor, LAPACK_ROW_MAJOR)
+
+} // end namespace Eigen
+
+#endif // EIGEN_JACOBISVD_LAPACKE_H
diff --git a/Eigen/src/SVD/JacobiSVD_MKL.h b/Eigen/src/SVD/JacobiSVD_MKL.h
deleted file mode 100644
index decda75..0000000
--- a/Eigen/src/SVD/JacobiSVD_MKL.h
+++ /dev/null
@@ -1,92 +0,0 @@
-/*
- Copyright (c) 2011, Intel Corporation. All rights reserved.
-
- Redistribution and use in source and binary forms, with or without modification,
- are permitted provided that the following conditions are met:
-
- * Redistributions of source code must retain the above copyright notice, this
-   list of conditions and the following disclaimer.
- * Redistributions in binary form must reproduce the above copyright notice,
-   this list of conditions and the following disclaimer in the documentation
-   and/or other materials provided with the distribution.
- * Neither the name of Intel Corporation nor the names of its contributors may
-   be used to endorse or promote products derived from this software without
-   specific prior written permission.
-
- THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
- ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
- WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
- DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
- ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
- (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
- LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
- ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
- SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
- ********************************************************************************
- *   Content : Eigen bindings to Intel(R) MKL
- *    Singular Value Decomposition - SVD.
- ********************************************************************************
-*/
-
-#ifndef EIGEN_JACOBISVD_MKL_H
-#define EIGEN_JACOBISVD_MKL_H
-
-#include "Eigen/src/Core/util/MKL_support.h"
-
-namespace Eigen { 
-
-/** \internal Specialization for the data types supported by MKL */
-
-#define EIGEN_MKL_SVD(EIGTYPE, MKLTYPE, MKLRTYPE, MKLPREFIX, EIGCOLROW, MKLCOLROW) \
-template<> inline \
-JacobiSVD<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>, ColPivHouseholderQRPreconditioner>& \
-JacobiSVD<Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>, ColPivHouseholderQRPreconditioner>::compute(const Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic>& matrix, unsigned int computationOptions) \
-{ \
-  typedef Matrix<EIGTYPE, Dynamic, Dynamic, EIGCOLROW, Dynamic, Dynamic> MatrixType; \
-  typedef MatrixType::Scalar Scalar; \
-  typedef MatrixType::RealScalar RealScalar; \
-  allocate(matrix.rows(), matrix.cols(), computationOptions); \
-\
-  /*const RealScalar precision = RealScalar(2) * NumTraits<Scalar>::epsilon();*/ \
-  m_nonzeroSingularValues = m_diagSize; \
-\
-  lapack_int lda = matrix.outerStride(), ldu, ldvt; \
-  lapack_int matrix_order = MKLCOLROW; \
-  char jobu, jobvt; \
-  MKLTYPE *u, *vt, dummy; \
-  jobu  = (m_computeFullU) ? 'A' : (m_computeThinU) ? 'S' : 'N'; \
-  jobvt = (m_computeFullV) ? 'A' : (m_computeThinV) ? 'S' : 'N'; \
-  if (computeU()) { \
-    ldu  = m_matrixU.outerStride(); \
-    u    = (MKLTYPE*)m_matrixU.data(); \
-  } else { ldu=1; u=&dummy; }\
-  MatrixType localV; \
-  ldvt = (m_computeFullV) ? m_cols : (m_computeThinV) ? m_diagSize : 1; \
-  if (computeV()) { \
-    localV.resize(ldvt, m_cols); \
-    vt   = (MKLTYPE*)localV.data(); \
-  } else { ldvt=1; vt=&dummy; }\
-  Matrix<MKLRTYPE, Dynamic, Dynamic> superb; superb.resize(m_diagSize, 1); \
-  MatrixType m_temp; m_temp = matrix; \
-  LAPACKE_##MKLPREFIX##gesvd( matrix_order, jobu, jobvt, m_rows, m_cols, (MKLTYPE*)m_temp.data(), lda, (MKLRTYPE*)m_singularValues.data(), u, ldu, vt, ldvt, superb.data()); \
-  if (computeV()) m_matrixV = localV.adjoint(); \
- /* for(int i=0;i<m_diagSize;i++) if (m_singularValues.coeffRef(i) < precision) { m_nonzeroSingularValues--; m_singularValues.coeffRef(i)=RealScalar(0);}*/ \
-  m_isInitialized = true; \
-  return *this; \
-}
-
-EIGEN_MKL_SVD(double,   double,        double, d, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_SVD(float,    float,         float , s, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_SVD(dcomplex, MKL_Complex16, double, z, ColMajor, LAPACK_COL_MAJOR)
-EIGEN_MKL_SVD(scomplex, MKL_Complex8,  float , c, ColMajor, LAPACK_COL_MAJOR)
-
-EIGEN_MKL_SVD(double,   double,        double, d, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_SVD(float,    float,         float , s, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_SVD(dcomplex, MKL_Complex16, double, z, RowMajor, LAPACK_ROW_MAJOR)
-EIGEN_MKL_SVD(scomplex, MKL_Complex8,  float , c, RowMajor, LAPACK_ROW_MAJOR)
-
-} // end namespace Eigen
-
-#endif // EIGEN_JACOBISVD_MKL_H
diff --git a/Eigen/src/SVD/SVDBase.h b/Eigen/src/SVD/SVDBase.h
new file mode 100644
index 0000000..3d1ef37
--- /dev/null
+++ b/Eigen/src/SVD/SVDBase.h
@@ -0,0 +1,315 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// Copyright (C) 2013 Gauthier Brun <brun.gauthier@gmail.com>
+// Copyright (C) 2013 Nicolas Carre <nicolas.carre@ensimag.fr>
+// Copyright (C) 2013 Jean Ceccato <jean.ceccato@ensimag.fr>
+// Copyright (C) 2013 Pierre Zoppitelli <pierre.zoppitelli@ensimag.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SVDBASE_H
+#define EIGEN_SVDBASE_H
+
+namespace Eigen {
+/** \ingroup SVD_Module
+ *
+ *
+ * \class SVDBase
+ *
+ * \brief Base class of SVD algorithms
+ *
+ * \tparam Derived the type of the actual SVD decomposition
+ *
+ * SVD decomposition consists in decomposing any n-by-p matrix \a A as a product
+ *   \f[ A = U S V^* \f]
+ * where \a U is a n-by-n unitary, \a V is a p-by-p unitary, and \a S is a n-by-p real positive matrix which is zero outside of its main diagonal;
+ * the diagonal entries of S are known as the \em singular \em values of \a A and the columns of \a U and \a V are known as the left
+ * and right \em singular \em vectors of \a A respectively.
+ *
+ * Singular values are always sorted in decreasing order.
+ *
+ * 
+ * You can ask for only \em thin \a U or \a V to be computed, meaning the following. In case of a rectangular n-by-p matrix, letting \a m be the
+ * smaller value among \a n and \a p, there are only \a m singular vectors; the remaining columns of \a U and \a V do not correspond to actual
+ * singular vectors. Asking for \em thin \a U or \a V means asking for only their \a m first columns to be formed. So \a U is then a n-by-m matrix,
+ * and \a V is then a p-by-m matrix. Notice that thin \a U and \a V are all you need for (least squares) solving.
+ *  
+ * If the input matrix has inf or nan coefficients, the result of the computation is undefined, but the computation is guaranteed to
+ * terminate in finite (and reasonable) time.
+ * \sa class BDCSVD, class JacobiSVD
+ */
+template<typename Derived>
+class SVDBase
+{
+
+public:
+  typedef typename internal::traits<Derived>::MatrixType MatrixType;
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename NumTraits<typename MatrixType::Scalar>::Real RealScalar;
+  typedef typename MatrixType::StorageIndex StorageIndex;
+  typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
+  enum {
+    RowsAtCompileTime = MatrixType::RowsAtCompileTime,
+    ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+    DiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_DYNAMIC(RowsAtCompileTime,ColsAtCompileTime),
+    MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime,
+    MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime,
+    MaxDiagSizeAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(MaxRowsAtCompileTime,MaxColsAtCompileTime),
+    MatrixOptions = MatrixType::Options
+  };
+
+  typedef Matrix<Scalar, RowsAtCompileTime, RowsAtCompileTime, MatrixOptions, MaxRowsAtCompileTime, MaxRowsAtCompileTime> MatrixUType;
+  typedef Matrix<Scalar, ColsAtCompileTime, ColsAtCompileTime, MatrixOptions, MaxColsAtCompileTime, MaxColsAtCompileTime> MatrixVType;
+  typedef typename internal::plain_diag_type<MatrixType, RealScalar>::type SingularValuesType;
+  
+  Derived& derived() { return *static_cast<Derived*>(this); }
+  const Derived& derived() const { return *static_cast<const Derived*>(this); }
+
+  /** \returns the \a U matrix.
+   *
+   * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
+   * the U matrix is n-by-n if you asked for \link Eigen::ComputeFullU ComputeFullU \endlink, and is n-by-m if you asked for \link Eigen::ComputeThinU ComputeThinU \endlink.
+   *
+   * The \a m first columns of \a U are the left singular vectors of the matrix being decomposed.
+   *
+   * This method asserts that you asked for \a U to be computed.
+   */
+  const MatrixUType& matrixU() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    eigen_assert(computeU() && "This SVD decomposition didn't compute U. Did you ask for it?");
+    return m_matrixU;
+  }
+
+  /** \returns the \a V matrix.
+   *
+   * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p,
+   * the V matrix is p-by-p if you asked for \link Eigen::ComputeFullV ComputeFullV \endlink, and is p-by-m if you asked for \link Eigen::ComputeThinV ComputeThinV \endlink.
+   *
+   * The \a m first columns of \a V are the right singular vectors of the matrix being decomposed.
+   *
+   * This method asserts that you asked for \a V to be computed.
+   */
+  const MatrixVType& matrixV() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    eigen_assert(computeV() && "This SVD decomposition didn't compute V. Did you ask for it?");
+    return m_matrixV;
+  }
+
+  /** \returns the vector of singular values.
+   *
+   * For the SVD decomposition of a n-by-p matrix, letting \a m be the minimum of \a n and \a p, the
+   * returned vector has size \a m.  Singular values are always sorted in decreasing order.
+   */
+  const SingularValuesType& singularValues() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    return m_singularValues;
+  }
+
+  /** \returns the number of singular values that are not exactly 0 */
+  Index nonzeroSingularValues() const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    return m_nonzeroSingularValues;
+  }
+  
+  /** \returns the rank of the matrix of which \c *this is the SVD.
+    *
+    * \note This method has to determine which singular values should be considered nonzero.
+    *       For that, it uses the threshold value that you can control by calling
+    *       setThreshold(const RealScalar&).
+    */
+  inline Index rank() const
+  {
+    using std::abs;
+    eigen_assert(m_isInitialized && "JacobiSVD is not initialized.");
+    if(m_singularValues.size()==0) return 0;
+    RealScalar premultiplied_threshold = numext::maxi<RealScalar>(m_singularValues.coeff(0) * threshold(), (std::numeric_limits<RealScalar>::min)());
+    Index i = m_nonzeroSingularValues-1;
+    while(i>=0 && m_singularValues.coeff(i) < premultiplied_threshold) --i;
+    return i+1;
+  }
+  
+  /** Allows to prescribe a threshold to be used by certain methods, such as rank() and solve(),
+    * which need to determine when singular values are to be considered nonzero.
+    * This is not used for the SVD decomposition itself.
+    *
+    * When it needs to get the threshold value, Eigen calls threshold().
+    * The default is \c NumTraits<Scalar>::epsilon()
+    *
+    * \param threshold The new value to use as the threshold.
+    *
+    * A singular value will be considered nonzero if its value is strictly greater than
+    *  \f$ \vert singular value \vert \leqslant threshold \times \vert max singular value \vert \f$.
+    *
+    * If you want to come back to the default behavior, call setThreshold(Default_t)
+    */
+  Derived& setThreshold(const RealScalar& threshold)
+  {
+    m_usePrescribedThreshold = true;
+    m_prescribedThreshold = threshold;
+    return derived();
+  }
+
+  /** Allows to come back to the default behavior, letting Eigen use its default formula for
+    * determining the threshold.
+    *
+    * You should pass the special object Eigen::Default as parameter here.
+    * \code svd.setThreshold(Eigen::Default); \endcode
+    *
+    * See the documentation of setThreshold(const RealScalar&).
+    */
+  Derived& setThreshold(Default_t)
+  {
+    m_usePrescribedThreshold = false;
+    return derived();
+  }
+
+  /** Returns the threshold that will be used by certain methods such as rank().
+    *
+    * See the documentation of setThreshold(const RealScalar&).
+    */
+  RealScalar threshold() const
+  {
+    eigen_assert(m_isInitialized || m_usePrescribedThreshold);
+    // this temporary is needed to workaround a MSVC issue
+    Index diagSize = (std::max<Index>)(1,m_diagSize);
+    return m_usePrescribedThreshold ? m_prescribedThreshold
+                                    : diagSize*NumTraits<Scalar>::epsilon();
+  }
+
+  /** \returns true if \a U (full or thin) is asked for in this SVD decomposition */
+  inline bool computeU() const { return m_computeFullU || m_computeThinU; }
+  /** \returns true if \a V (full or thin) is asked for in this SVD decomposition */
+  inline bool computeV() const { return m_computeFullV || m_computeThinV; }
+
+  inline Index rows() const { return m_rows; }
+  inline Index cols() const { return m_cols; }
+  
+  /** \returns a (least squares) solution of \f$ A x = b \f$ using the current SVD decomposition of A.
+    *
+    * \param b the right-hand-side of the equation to solve.
+    *
+    * \note Solving requires both U and V to be computed. Thin U and V are enough, there is no need for full U or V.
+    *
+    * \note SVD solving is implicitly least-squares. Thus, this method serves both purposes of exact solving and least-squares solving.
+    * In other words, the returned solution is guaranteed to minimize the Euclidean norm \f$ \Vert A x - b \Vert \f$.
+    */
+  template<typename Rhs>
+  inline const Solve<Derived, Rhs>
+  solve(const MatrixBase<Rhs>& b) const
+  {
+    eigen_assert(m_isInitialized && "SVD is not initialized.");
+    eigen_assert(computeU() && computeV() && "SVD::solve() requires both unitaries U and V to be computed (thin unitaries suffice).");
+    return Solve<Derived, Rhs>(derived(), b.derived());
+  }
+  
+  #ifndef EIGEN_PARSED_BY_DOXYGEN
+  template<typename RhsType, typename DstType>
+  EIGEN_DEVICE_FUNC
+  void _solve_impl(const RhsType &rhs, DstType &dst) const;
+  #endif
+
+protected:
+  
+  static void check_template_parameters()
+  {
+    EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar);
+  }
+  
+  // return true if already allocated
+  bool allocate(Index rows, Index cols, unsigned int computationOptions) ;
+
+  MatrixUType m_matrixU;
+  MatrixVType m_matrixV;
+  SingularValuesType m_singularValues;
+  bool m_isInitialized, m_isAllocated, m_usePrescribedThreshold;
+  bool m_computeFullU, m_computeThinU;
+  bool m_computeFullV, m_computeThinV;
+  unsigned int m_computationOptions;
+  Index m_nonzeroSingularValues, m_rows, m_cols, m_diagSize;
+  RealScalar m_prescribedThreshold;
+
+  /** \brief Default Constructor.
+   *
+   * Default constructor of SVDBase
+   */
+  SVDBase()
+    : m_isInitialized(false),
+      m_isAllocated(false),
+      m_usePrescribedThreshold(false),
+      m_computationOptions(0),
+      m_rows(-1), m_cols(-1), m_diagSize(0)
+  {
+    check_template_parameters();
+  }
+
+
+};
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename Derived>
+template<typename RhsType, typename DstType>
+void SVDBase<Derived>::_solve_impl(const RhsType &rhs, DstType &dst) const
+{
+  eigen_assert(rhs.rows() == rows());
+
+  // A = U S V^*
+  // So A^{-1} = V S^{-1} U^*
+
+  Matrix<Scalar, Dynamic, RhsType::ColsAtCompileTime, 0, MatrixType::MaxRowsAtCompileTime, RhsType::MaxColsAtCompileTime> tmp;
+  Index l_rank = rank();
+  tmp.noalias() =  m_matrixU.leftCols(l_rank).adjoint() * rhs;
+  tmp = m_singularValues.head(l_rank).asDiagonal().inverse() * tmp;
+  dst = m_matrixV.leftCols(l_rank) * tmp;
+}
+#endif
+
+template<typename MatrixType>
+bool SVDBase<MatrixType>::allocate(Index rows, Index cols, unsigned int computationOptions)
+{
+  eigen_assert(rows >= 0 && cols >= 0);
+
+  if (m_isAllocated &&
+      rows == m_rows &&
+      cols == m_cols &&
+      computationOptions == m_computationOptions)
+  {
+    return true;
+  }
+
+  m_rows = rows;
+  m_cols = cols;
+  m_isInitialized = false;
+  m_isAllocated = true;
+  m_computationOptions = computationOptions;
+  m_computeFullU = (computationOptions & ComputeFullU) != 0;
+  m_computeThinU = (computationOptions & ComputeThinU) != 0;
+  m_computeFullV = (computationOptions & ComputeFullV) != 0;
+  m_computeThinV = (computationOptions & ComputeThinV) != 0;
+  eigen_assert(!(m_computeFullU && m_computeThinU) && "SVDBase: you can't ask for both full and thin U");
+  eigen_assert(!(m_computeFullV && m_computeThinV) && "SVDBase: you can't ask for both full and thin V");
+  eigen_assert(EIGEN_IMPLIES(m_computeThinU || m_computeThinV, MatrixType::ColsAtCompileTime==Dynamic) &&
+	       "SVDBase: thin U and V are only available when your matrix has a dynamic number of columns.");
+
+  m_diagSize = (std::min)(m_rows, m_cols);
+  m_singularValues.resize(m_diagSize);
+  if(RowsAtCompileTime==Dynamic)
+    m_matrixU.resize(m_rows, m_computeFullU ? m_rows : m_computeThinU ? m_diagSize : 0);
+  if(ColsAtCompileTime==Dynamic)
+    m_matrixV.resize(m_cols, m_computeFullV ? m_cols : m_computeThinV ? m_diagSize : 0);
+
+  return false;
+}
+
+}// end namespace
+
+#endif // EIGEN_SVDBASE_H
diff --git a/Eigen/src/SVD/UpperBidiagonalization.h b/Eigen/src/SVD/UpperBidiagonalization.h
index 587de37..11ac847 100644
--- a/Eigen/src/SVD/UpperBidiagonalization.h
+++ b/Eigen/src/SVD/UpperBidiagonalization.h
@@ -2,6 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2013-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -28,15 +29,15 @@
     };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef Eigen::Index Index; ///< \deprecated since Eigen 3.3
     typedef Matrix<Scalar, 1, ColsAtCompileTime> RowVectorType;
     typedef Matrix<Scalar, RowsAtCompileTime, 1> ColVectorType;
-    typedef BandMatrix<RealScalar, ColsAtCompileTime, ColsAtCompileTime, 1, 0> BidiagonalType;
+    typedef BandMatrix<RealScalar, ColsAtCompileTime, ColsAtCompileTime, 1, 0, RowMajor> BidiagonalType;
     typedef Matrix<Scalar, ColsAtCompileTime, 1> DiagVectorType;
     typedef Matrix<Scalar, ColsAtCompileTimeMinusOne, 1> SuperDiagVectorType;
     typedef HouseholderSequence<
               const MatrixType,
-              CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, const Diagonal<const MatrixType,0> >
+              const typename internal::remove_all<typename Diagonal<const MatrixType,0>::ConjugateReturnType>::type
             > HouseholderUSequenceType;
     typedef HouseholderSequence<
               const typename internal::remove_all<typename MatrixType::ConjugateReturnType>::type,
@@ -52,7 +53,7 @@
     */
     UpperBidiagonalization() : m_householder(), m_bidiagonal(), m_isInitialized(false) {}
 
-    UpperBidiagonalization(const MatrixType& matrix)
+    explicit UpperBidiagonalization(const MatrixType& matrix)
       : m_householder(matrix.rows(), matrix.cols()),
         m_bidiagonal(matrix.cols(), matrix.cols()),
         m_isInitialized(false)
@@ -61,6 +62,7 @@
     }
     
     UpperBidiagonalization& compute(const MatrixType& matrix);
+    UpperBidiagonalization& computeUnblocked(const MatrixType& matrix);
     
     const MatrixType& householder() const { return m_householder; }
     const BidiagonalType& bidiagonal() const { return m_bidiagonal; }
@@ -85,45 +87,309 @@
     bool m_isInitialized;
 };
 
-template<typename _MatrixType>
-UpperBidiagonalization<_MatrixType>& UpperBidiagonalization<_MatrixType>::compute(const _MatrixType& matrix)
+// Standard upper bidiagonalization without fancy optimizations
+// This version should be faster for small matrix size
+template<typename MatrixType>
+void upperbidiagonalization_inplace_unblocked(MatrixType& mat,
+                                              typename MatrixType::RealScalar *diagonal,
+                                              typename MatrixType::RealScalar *upper_diagonal,
+                                              typename MatrixType::Scalar* tempData = 0)
 {
-  Index rows = matrix.rows();
-  Index cols = matrix.cols();
-  
-  eigen_assert(rows >= cols && "UpperBidiagonalization is only for matrices satisfying rows>=cols.");
-  
-  m_householder = matrix;
+  typedef typename MatrixType::Scalar Scalar;
 
-  ColVectorType temp(rows);
+  Index rows = mat.rows();
+  Index cols = mat.cols();
+
+  typedef Matrix<Scalar,Dynamic,1,ColMajor,MatrixType::MaxRowsAtCompileTime,1> TempType;
+  TempType tempVector;
+  if(tempData==0)
+  {
+    tempVector.resize(rows);
+    tempData = tempVector.data();
+  }
 
   for (Index k = 0; /* breaks at k==cols-1 below */ ; ++k)
   {
     Index remainingRows = rows - k;
     Index remainingCols = cols - k - 1;
 
-    // construct left householder transform in-place in m_householder
-    m_householder.col(k).tail(remainingRows)
-                 .makeHouseholderInPlace(m_householder.coeffRef(k,k),
-                                         m_bidiagonal.template diagonal<0>().coeffRef(k));
-    // apply householder transform to remaining part of m_householder on the left
-    m_householder.bottomRightCorner(remainingRows, remainingCols)
-                 .applyHouseholderOnTheLeft(m_householder.col(k).tail(remainingRows-1),
-                                            m_householder.coeff(k,k),
-                                            temp.data());
+    // construct left householder transform in-place in A
+    mat.col(k).tail(remainingRows)
+       .makeHouseholderInPlace(mat.coeffRef(k,k), diagonal[k]);
+    // apply householder transform to remaining part of A on the left
+    mat.bottomRightCorner(remainingRows, remainingCols)
+       .applyHouseholderOnTheLeft(mat.col(k).tail(remainingRows-1), mat.coeff(k,k), tempData);
 
     if(k == cols-1) break;
-    
-    // construct right householder transform in-place in m_householder
-    m_householder.row(k).tail(remainingCols)
-                 .makeHouseholderInPlace(m_householder.coeffRef(k,k+1),
-                                         m_bidiagonal.template diagonal<1>().coeffRef(k));
-    // apply householder transform to remaining part of m_householder on the left
-    m_householder.bottomRightCorner(remainingRows-1, remainingCols)
-                 .applyHouseholderOnTheRight(m_householder.row(k).tail(remainingCols-1).transpose(),
-                                             m_householder.coeff(k,k+1),
-                                             temp.data());
+
+    // construct right householder transform in-place in mat
+    mat.row(k).tail(remainingCols)
+       .makeHouseholderInPlace(mat.coeffRef(k,k+1), upper_diagonal[k]);
+    // apply householder transform to remaining part of mat on the left
+    mat.bottomRightCorner(remainingRows-1, remainingCols)
+       .applyHouseholderOnTheRight(mat.row(k).tail(remainingCols-1).transpose(), mat.coeff(k,k+1), tempData);
   }
+}
+
+/** \internal
+  * Helper routine for the block reduction to upper bidiagonal form.
+  *
+  * Let's partition the matrix A:
+  * 
+  *      | A00 A01 |
+  *  A = |         |
+  *      | A10 A11 |
+  *
+  * This function reduces to bidiagonal form the left \c rows x \a blockSize vertical panel [A00/A10]
+  * and the \a blockSize x \c cols horizontal panel [A00 A01] of the matrix \a A. The bottom-right block A11
+  * is updated using matrix-matrix products:
+  *   A22 -= V * Y^T - X * U^T
+  * where V and U contains the left and right Householder vectors. U and V are stored in A10, and A01
+  * respectively, and the update matrices X and Y are computed during the reduction.
+  * 
+  */
+template<typename MatrixType>
+void upperbidiagonalization_blocked_helper(MatrixType& A,
+                                           typename MatrixType::RealScalar *diagonal,
+                                           typename MatrixType::RealScalar *upper_diagonal,
+                                           Index bs,
+                                           Ref<Matrix<typename MatrixType::Scalar, Dynamic, Dynamic,
+                                                      traits<MatrixType>::Flags & RowMajorBit> > X,
+                                           Ref<Matrix<typename MatrixType::Scalar, Dynamic, Dynamic,
+                                                      traits<MatrixType>::Flags & RowMajorBit> > Y)
+{
+  typedef typename MatrixType::Scalar Scalar;
+  typedef typename MatrixType::RealScalar RealScalar;
+  typedef typename NumTraits<RealScalar>::Literal Literal;
+  enum { StorageOrder = traits<MatrixType>::Flags & RowMajorBit };
+  typedef InnerStride<int(StorageOrder) == int(ColMajor) ? 1 : Dynamic> ColInnerStride;
+  typedef InnerStride<int(StorageOrder) == int(ColMajor) ? Dynamic : 1> RowInnerStride;
+  typedef Ref<Matrix<Scalar, Dynamic, 1>, 0, ColInnerStride>    SubColumnType;
+  typedef Ref<Matrix<Scalar, 1, Dynamic>, 0, RowInnerStride>    SubRowType;
+  typedef Ref<Matrix<Scalar, Dynamic, Dynamic, StorageOrder > > SubMatType;
+  
+  Index brows = A.rows();
+  Index bcols = A.cols();
+
+  Scalar tau_u, tau_u_prev(0), tau_v;
+
+  for(Index k = 0; k < bs; ++k)
+  {
+    Index remainingRows = brows - k;
+    Index remainingCols = bcols - k - 1;
+
+    SubMatType X_k1( X.block(k,0, remainingRows,k) );
+    SubMatType V_k1( A.block(k,0, remainingRows,k) );
+
+    // 1 - update the k-th column of A
+    SubColumnType v_k = A.col(k).tail(remainingRows);
+          v_k -= V_k1 * Y.row(k).head(k).adjoint();
+    if(k) v_k -= X_k1 * A.col(k).head(k);
+    
+    // 2 - construct left Householder transform in-place
+    v_k.makeHouseholderInPlace(tau_v, diagonal[k]);
+       
+    if(k+1<bcols)
+    {
+      SubMatType Y_k  ( Y.block(k+1,0, remainingCols, k+1) );
+      SubMatType U_k1 ( A.block(0,k+1, k,remainingCols) );
+      
+      // this eases the application of Householder transforAions
+      // A(k,k) will store tau_v later
+      A(k,k) = Scalar(1);
+
+      // 3 - Compute y_k^T = tau_v * ( A^T*v_k - Y_k-1*V_k-1^T*v_k - U_k-1*X_k-1^T*v_k )
+      {
+        SubColumnType y_k( Y.col(k).tail(remainingCols) );
+        
+        // let's use the begining of column k of Y as a temporary vector
+        SubColumnType tmp( Y.col(k).head(k) );
+        y_k.noalias()  = A.block(k,k+1, remainingRows,remainingCols).adjoint() * v_k; // bottleneck
+        tmp.noalias()  = V_k1.adjoint()  * v_k;
+        y_k.noalias() -= Y_k.leftCols(k) * tmp;
+        tmp.noalias()  = X_k1.adjoint()  * v_k;
+        y_k.noalias() -= U_k1.adjoint()  * tmp;
+        y_k *= numext::conj(tau_v);
+      }
+
+      // 4 - update k-th row of A (it will become u_k)
+      SubRowType u_k( A.row(k).tail(remainingCols) );
+      u_k = u_k.conjugate();
+      {
+        u_k -= Y_k * A.row(k).head(k+1).adjoint();
+        if(k) u_k -= U_k1.adjoint() * X.row(k).head(k).adjoint();
+      }
+
+      // 5 - construct right Householder transform in-place
+      u_k.makeHouseholderInPlace(tau_u, upper_diagonal[k]);
+
+      // this eases the application of Householder transformations
+      // A(k,k+1) will store tau_u later
+      A(k,k+1) = Scalar(1);
+
+      // 6 - Compute x_k = tau_u * ( A*u_k - X_k-1*U_k-1^T*u_k - V_k*Y_k^T*u_k )
+      {
+        SubColumnType x_k ( X.col(k).tail(remainingRows-1) );
+        
+        // let's use the begining of column k of X as a temporary vectors
+        // note that tmp0 and tmp1 overlaps
+        SubColumnType tmp0 ( X.col(k).head(k) ),
+                      tmp1 ( X.col(k).head(k+1) );
+                    
+        x_k.noalias()   = A.block(k+1,k+1, remainingRows-1,remainingCols) * u_k.transpose(); // bottleneck
+        tmp0.noalias()  = U_k1 * u_k.transpose();
+        x_k.noalias()  -= X_k1.bottomRows(remainingRows-1) * tmp0;
+        tmp1.noalias()  = Y_k.adjoint() * u_k.transpose();
+        x_k.noalias()  -= A.block(k+1,0, remainingRows-1,k+1) * tmp1;
+        x_k *= numext::conj(tau_u);
+        tau_u = numext::conj(tau_u);
+        u_k = u_k.conjugate();
+      }
+
+      if(k>0) A.coeffRef(k-1,k) = tau_u_prev;
+      tau_u_prev = tau_u;
+    }
+    else
+      A.coeffRef(k-1,k) = tau_u_prev;
+
+    A.coeffRef(k,k) = tau_v;
+  }
+  
+  if(bs<bcols)
+    A.coeffRef(bs-1,bs) = tau_u_prev;
+
+  // update A22
+  if(bcols>bs && brows>bs)
+  {
+    SubMatType A11( A.bottomRightCorner(brows-bs,bcols-bs) );
+    SubMatType A10( A.block(bs,0, brows-bs,bs) );
+    SubMatType A01( A.block(0,bs, bs,bcols-bs) );
+    Scalar tmp = A01(bs-1,0);
+    A01(bs-1,0) = Literal(1);
+    A11.noalias() -= A10 * Y.topLeftCorner(bcols,bs).bottomRows(bcols-bs).adjoint();
+    A11.noalias() -= X.topLeftCorner(brows,bs).bottomRows(brows-bs) * A01;
+    A01(bs-1,0) = tmp;
+  }
+}
+
+/** \internal
+  *
+  * Implementation of a block-bidiagonal reduction.
+  * It is based on the following paper:
+  *   The Design of a Parallel Dense Linear Algebra Software Library: Reduction to Hessenberg, Tridiagonal, and Bidiagonal Form.
+  *   by Jaeyoung Choi, Jack J. Dongarra, David W. Walker. (1995)
+  *   section 3.3
+  */
+template<typename MatrixType, typename BidiagType>
+void upperbidiagonalization_inplace_blocked(MatrixType& A, BidiagType& bidiagonal,
+                                            Index maxBlockSize=32,
+                                            typename MatrixType::Scalar* /*tempData*/ = 0)
+{
+  typedef typename MatrixType::Scalar Scalar;
+  typedef Block<MatrixType,Dynamic,Dynamic> BlockType;
+
+  Index rows = A.rows();
+  Index cols = A.cols();
+  Index size = (std::min)(rows, cols);
+
+  // X and Y are work space
+  enum { StorageOrder = traits<MatrixType>::Flags & RowMajorBit };
+  Matrix<Scalar,
+         MatrixType::RowsAtCompileTime,
+         Dynamic,
+         StorageOrder,
+         MatrixType::MaxRowsAtCompileTime> X(rows,maxBlockSize);
+  Matrix<Scalar,
+         MatrixType::ColsAtCompileTime,
+         Dynamic,
+         StorageOrder,
+         MatrixType::MaxColsAtCompileTime> Y(cols,maxBlockSize);
+  Index blockSize = (std::min)(maxBlockSize,size);
+
+  Index k = 0;
+  for(k = 0; k < size; k += blockSize)
+  {
+    Index bs = (std::min)(size-k,blockSize);  // actual size of the block
+    Index brows = rows - k;                   // rows of the block
+    Index bcols = cols - k;                   // columns of the block
+
+    // partition the matrix A:
+    // 
+    //      | A00 A01 A02 |
+    //      |             |
+    // A  = | A10 A11 A12 |
+    //      |             |
+    //      | A20 A21 A22 |
+    //
+    // where A11 is a bs x bs diagonal block,
+    // and let:
+    //      | A11 A12 |
+    //  B = |         |
+    //      | A21 A22 |
+
+    BlockType B = A.block(k,k,brows,bcols);
+    
+    // This stage performs the bidiagonalization of A11, A21, A12, and updating of A22.
+    // Finally, the algorithm continue on the updated A22.
+    //
+    // However, if B is too small, or A22 empty, then let's use an unblocked strategy
+    if(k+bs==cols || bcols<48) // somewhat arbitrary threshold
+    {
+      upperbidiagonalization_inplace_unblocked(B,
+                                               &(bidiagonal.template diagonal<0>().coeffRef(k)),
+                                               &(bidiagonal.template diagonal<1>().coeffRef(k)),
+                                               X.data()
+                                              );
+      break; // We're done
+    }
+    else
+    {
+      upperbidiagonalization_blocked_helper<BlockType>( B,
+                                                        &(bidiagonal.template diagonal<0>().coeffRef(k)),
+                                                        &(bidiagonal.template diagonal<1>().coeffRef(k)),
+                                                        bs,
+                                                        X.topLeftCorner(brows,bs),
+                                                        Y.topLeftCorner(bcols,bs)
+                                                      );
+    }
+  }
+}
+
+template<typename _MatrixType>
+UpperBidiagonalization<_MatrixType>& UpperBidiagonalization<_MatrixType>::computeUnblocked(const _MatrixType& matrix)
+{
+  Index rows = matrix.rows();
+  Index cols = matrix.cols();
+  EIGEN_ONLY_USED_FOR_DEBUG(cols);
+
+  eigen_assert(rows >= cols && "UpperBidiagonalization is only for Arices satisfying rows>=cols.");
+
+  m_householder = matrix;
+
+  ColVectorType temp(rows);
+
+  upperbidiagonalization_inplace_unblocked(m_householder,
+                                           &(m_bidiagonal.template diagonal<0>().coeffRef(0)),
+                                           &(m_bidiagonal.template diagonal<1>().coeffRef(0)),
+                                           temp.data());
+
+  m_isInitialized = true;
+  return *this;
+}
+
+template<typename _MatrixType>
+UpperBidiagonalization<_MatrixType>& UpperBidiagonalization<_MatrixType>::compute(const _MatrixType& matrix)
+{
+  Index rows = matrix.rows();
+  Index cols = matrix.cols();
+  EIGEN_ONLY_USED_FOR_DEBUG(rows);
+  EIGEN_ONLY_USED_FOR_DEBUG(cols);
+
+  eigen_assert(rows >= cols && "UpperBidiagonalization is only for Arices satisfying rows>=cols.");
+
+  m_householder = matrix;
+  upperbidiagonalization_inplace_blocked(m_householder, m_bidiagonal);
+            
   m_isInitialized = true;
   return *this;
 }
diff --git a/Eigen/src/SparseCholesky/CMakeLists.txt b/Eigen/src/SparseCholesky/CMakeLists.txt
deleted file mode 100644
index 375a59d..0000000
--- a/Eigen/src/SparseCholesky/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_SparseCholesky_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_SparseCholesky_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/SparseCholesky COMPONENT Devel
-  )
diff --git a/Eigen/src/SparseCholesky/SimplicialCholesky.h b/Eigen/src/SparseCholesky/SimplicialCholesky.h
index e1f96ba..2907f65 100644
--- a/Eigen/src/SparseCholesky/SimplicialCholesky.h
+++ b/Eigen/src/SparseCholesky/SimplicialCholesky.h
@@ -17,43 +17,74 @@
   SimplicialCholeskyLDLT
 };
 
+namespace internal {
+  template<typename CholMatrixType, typename InputMatrixType>
+  struct simplicial_cholesky_grab_input {
+    typedef CholMatrixType const * ConstCholMatrixPtr;
+    static void run(const InputMatrixType& input, ConstCholMatrixPtr &pmat, CholMatrixType &tmp)
+    {
+      tmp = input;
+      pmat = &tmp;
+    }
+  };
+  
+  template<typename MatrixType>
+  struct simplicial_cholesky_grab_input<MatrixType,MatrixType> {
+    typedef MatrixType const * ConstMatrixPtr;
+    static void run(const MatrixType& input, ConstMatrixPtr &pmat, MatrixType &/*tmp*/)
+    {
+      pmat = &input;
+    }
+  };
+} // end namespace internal
+
 /** \ingroup SparseCholesky_Module
-  * \brief A direct sparse Cholesky factorizations
+  * \brief A base class for direct sparse Cholesky factorizations
   *
-  * These classes provide LL^T and LDL^T Cholesky factorizations of sparse matrices that are
-  * selfadjoint and positive definite. The factorization allows for solving A.X = B where
+  * This is a base class for LL^T and LDL^T Cholesky factorizations of sparse matrices that are
+  * selfadjoint and positive definite. These factorizations allow for solving A.X = B where
   * X and B can be either dense or sparse.
   * 
   * In order to reduce the fill-in, a symmetric permutation P is applied prior to the factorization
   * such that the factorized matrix is P A P^-1.
   *
-  * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
-  * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower
-  *               or Upper. Default is Lower.
+  * \tparam Derived the type of the derived class, that is the actual factorization type.
   *
   */
 template<typename Derived>
-class SimplicialCholeskyBase : internal::noncopyable
+class SimplicialCholeskyBase : public SparseSolverBase<Derived>
 {
+    typedef SparseSolverBase<Derived> Base;
+    using Base::m_isInitialized;
+    
   public:
     typedef typename internal::traits<Derived>::MatrixType MatrixType;
     typedef typename internal::traits<Derived>::OrderingType OrderingType;
     enum { UpLo = internal::traits<Derived>::UpLo };
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> CholMatrixType;
+    typedef CholMatrixType const * ConstCholMatrixPtr;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
+    
+    using Base::derived;
 
     /** Default constructor */
     SimplicialCholeskyBase()
-      : m_info(Success), m_isInitialized(false), m_shiftOffset(0), m_shiftScale(1)
+      : m_info(Success), m_shiftOffset(0), m_shiftScale(1)
     {}
 
-    SimplicialCholeskyBase(const MatrixType& matrix)
-      : m_info(Success), m_isInitialized(false), m_shiftOffset(0), m_shiftScale(1)
+    explicit SimplicialCholeskyBase(const MatrixType& matrix)
+      : m_info(Success), m_shiftOffset(0), m_shiftScale(1)
     {
       derived().compute(matrix);
     }
@@ -79,42 +110,14 @@
       return m_info;
     }
     
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<SimplicialCholeskyBase, Rhs>
-    solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Simplicial LLT or LDLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SimplicialCholeskyBase::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<SimplicialCholeskyBase, Rhs>(*this, b.derived());
-    }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<SimplicialCholeskyBase, Rhs>
-    solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "Simplicial LLT or LDLT is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SimplicialCholesky::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<SimplicialCholeskyBase, Rhs>(*this, b.derived());
-    }
-    
     /** \returns the permutation P
       * \sa permutationPinv() */
-    const PermutationMatrix<Dynamic,Dynamic,Index>& permutationP() const
+    const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& permutationP() const
     { return m_P; }
     
     /** \returns the inverse P^-1 of the permutation P
       * \sa permutationP() */
-    const PermutationMatrix<Dynamic,Dynamic,Index>& permutationPinv() const
+    const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& permutationPinv() const
     { return m_Pinv; }
 
     /** Sets the shift parameters that will be used to adjust the diagonal coefficients during the numerical factorization.
@@ -150,7 +153,7 @@
 
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       eigen_assert(m_matrix.rows()==b.rows());
@@ -175,6 +178,12 @@
       if(m_P.size()>0)
         dest = m_Pinv * dest;
     }
+    
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const SparseMatrixBase<Rhs> &b, SparseMatrixBase<Dest> &dest) const
+    {
+      internal::solve_sparse_through_dense_panels(derived(), b, dest);
+    }
 
 #endif // EIGEN_PARSED_BY_DOXYGEN
 
@@ -186,20 +195,33 @@
     {
       eigen_assert(matrix.rows()==matrix.cols());
       Index size = matrix.cols();
-      CholMatrixType ap(size,size);
-      ordering(matrix, ap);
-      analyzePattern_preordered(ap, DoLDLT);
-      factorize_preordered<DoLDLT>(ap);
+      CholMatrixType tmp(size,size);
+      ConstCholMatrixPtr pmat;
+      ordering(matrix, pmat, tmp);
+      analyzePattern_preordered(*pmat, DoLDLT);
+      factorize_preordered<DoLDLT>(*pmat);
     }
     
     template<bool DoLDLT>
     void factorize(const MatrixType& a)
     {
       eigen_assert(a.rows()==a.cols());
-      int size = a.cols();
-      CholMatrixType ap(size,size);
-      ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
-      factorize_preordered<DoLDLT>(ap);
+      Index size = a.cols();
+      CholMatrixType tmp(size,size);
+      ConstCholMatrixPtr pmat;
+      
+      if(m_P.size()==0 && (UpLo&Upper)==Upper)
+      {
+        // If there is no ordering, try to directly use the input matrix without any copy
+        internal::simplicial_cholesky_grab_input<CholMatrixType,MatrixType>::run(a, pmat, tmp);
+      }
+      else
+      {
+        tmp.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
+        pmat = &tmp;
+      }
+      
+      factorize_preordered<DoLDLT>(*pmat);
     }
 
     template<bool DoLDLT>
@@ -208,14 +230,15 @@
     void analyzePattern(const MatrixType& a, bool doLDLT)
     {
       eigen_assert(a.rows()==a.cols());
-      int size = a.cols();
-      CholMatrixType ap(size,size);
-      ordering(a, ap);
-      analyzePattern_preordered(ap,doLDLT);
+      Index size = a.cols();
+      CholMatrixType tmp(size,size);
+      ConstCholMatrixPtr pmat;
+      ordering(a, pmat, tmp);
+      analyzePattern_preordered(*pmat,doLDLT);
     }
     void analyzePattern_preordered(const CholMatrixType& a, bool doLDLT);
     
-    void ordering(const MatrixType& a, CholMatrixType& ap);
+    void ordering(const MatrixType& a, ConstCholMatrixPtr &pmat, CholMatrixType& ap);
 
     /** keeps off-diagonal entries; drops diagonal entries */
     struct keep_diag {
@@ -226,24 +249,23 @@
     };
 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     bool m_factorizationIsOk;
     bool m_analysisIsOk;
     
     CholMatrixType m_matrix;
     VectorType m_diag;                                // the diagonal coefficients (LDLT mode)
-    VectorXi m_parent;                                // elimination tree
-    VectorXi m_nonZerosPerCol;
-    PermutationMatrix<Dynamic,Dynamic,Index> m_P;     // the permutation
-    PermutationMatrix<Dynamic,Dynamic,Index> m_Pinv;  // the inverse permutation
+    VectorI m_parent;                                 // elimination tree
+    VectorI m_nonZerosPerCol;
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_P;     // the permutation
+    PermutationMatrix<Dynamic,Dynamic,StorageIndex> m_Pinv;  // the inverse permutation
 
     RealScalar m_shiftOffset;
     RealScalar m_shiftScale;
 };
 
-template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::Index> > class SimplicialLLT;
-template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::Index> > class SimplicialLDLT;
-template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::Index> > class SimplicialCholesky;
+template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::StorageIndex> > class SimplicialLLT;
+template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::StorageIndex> > class SimplicialLDLT;
+template<typename _MatrixType, int _UpLo = Lower, typename _Ordering = AMDOrdering<typename _MatrixType::StorageIndex> > class SimplicialCholesky;
 
 namespace internal {
 
@@ -253,12 +275,12 @@
   typedef _Ordering OrderingType;
   enum { UpLo = _UpLo };
   typedef typename MatrixType::Scalar                         Scalar;
-  typedef typename MatrixType::Index                          Index;
-  typedef SparseMatrix<Scalar, ColMajor, Index>               CholMatrixType;
-  typedef SparseTriangularView<CholMatrixType, Eigen::Lower>  MatrixL;
-  typedef SparseTriangularView<typename CholMatrixType::AdjointReturnType, Eigen::Upper>   MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  typedef typename MatrixType::StorageIndex                   StorageIndex;
+  typedef SparseMatrix<Scalar, ColMajor, StorageIndex>        CholMatrixType;
+  typedef TriangularView<const CholMatrixType, Eigen::Lower>  MatrixL;
+  typedef TriangularView<const typename CholMatrixType::AdjointReturnType, Eigen::Upper>   MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
 };
 
 template<typename _MatrixType,int _UpLo, typename _Ordering> struct traits<SimplicialLDLT<_MatrixType,_UpLo,_Ordering> >
@@ -267,12 +289,12 @@
   typedef _Ordering OrderingType;
   enum { UpLo = _UpLo };
   typedef typename MatrixType::Scalar                             Scalar;
-  typedef typename MatrixType::Index                              Index;
-  typedef SparseMatrix<Scalar, ColMajor, Index>                   CholMatrixType;
-  typedef SparseTriangularView<CholMatrixType, Eigen::UnitLower>  MatrixL;
-  typedef SparseTriangularView<typename CholMatrixType::AdjointReturnType, Eigen::UnitUpper> MatrixU;
-  static inline MatrixL getL(const MatrixType& m) { return m; }
-  static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); }
+  typedef typename MatrixType::StorageIndex                       StorageIndex;
+  typedef SparseMatrix<Scalar, ColMajor, StorageIndex>            CholMatrixType;
+  typedef TriangularView<const CholMatrixType, Eigen::UnitLower>  MatrixL;
+  typedef TriangularView<const typename CholMatrixType::AdjointReturnType, Eigen::UnitUpper> MatrixU;
+  static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); }
+  static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); }
 };
 
 template<typename _MatrixType, int _UpLo, typename _Ordering> struct traits<SimplicialCholesky<_MatrixType,_UpLo,_Ordering> >
@@ -300,6 +322,8 @@
   *               or Upper. Default is Lower.
   * \tparam _Ordering The ordering method to use, either AMDOrdering<> or NaturalOrdering<>. Default is AMDOrdering<>
   *
+  * \implsparsesolverconcept
+  *
   * \sa class SimplicialLDLT, class AMDOrdering, class NaturalOrdering
   */
 template<typename _MatrixType, int _UpLo, typename _Ordering>
@@ -311,7 +335,7 @@
     typedef SimplicialCholeskyBase<SimplicialLLT> Base;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
     typedef internal::traits<SimplicialLLT> Traits;
@@ -321,7 +345,7 @@
     /** Default constructor */
     SimplicialLLT() : Base() {}
     /** Constructs and performs the LLT factorization of \a matrix */
-    SimplicialLLT(const MatrixType& matrix)
+    explicit SimplicialLLT(const MatrixType& matrix)
         : Base(matrix) {}
 
     /** \returns an expression of the factor L */
@@ -389,6 +413,8 @@
   *               or Upper. Default is Lower.
   * \tparam _Ordering The ordering method to use, either AMDOrdering<> or NaturalOrdering<>. Default is AMDOrdering<>
   *
+  * \implsparsesolverconcept
+  *
   * \sa class SimplicialLLT, class AMDOrdering, class NaturalOrdering
   */
 template<typename _MatrixType, int _UpLo, typename _Ordering>
@@ -400,8 +426,8 @@
     typedef SimplicialCholeskyBase<SimplicialLDLT> Base;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> CholMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
     typedef internal::traits<SimplicialLDLT> Traits;
     typedef typename Traits::MatrixL  MatrixL;
@@ -411,7 +437,7 @@
     SimplicialLDLT() : Base() {}
 
     /** Constructs and performs the LLT factorization of \a matrix */
-    SimplicialLDLT(const MatrixType& matrix)
+    explicit SimplicialLDLT(const MatrixType& matrix)
         : Base(matrix) {}
 
     /** \returns a vector expression of the diagonal D */
@@ -482,8 +508,8 @@
     typedef SimplicialCholeskyBase<SimplicialCholesky> Base;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> CholMatrixType;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> CholMatrixType;
     typedef Matrix<Scalar,Dynamic,1> VectorType;
     typedef internal::traits<SimplicialCholesky> Traits;
     typedef internal::traits<SimplicialLDLT<MatrixType,UpLo> > LDLTTraits;
@@ -491,7 +517,7 @@
   public:
     SimplicialCholesky() : Base(), m_LDLT(true) {}
 
-    SimplicialCholesky(const MatrixType& matrix)
+    explicit SimplicialCholesky(const MatrixType& matrix)
       : Base(), m_LDLT(true)
     {
       compute(matrix);
@@ -560,7 +586,7 @@
 
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const
     {
       eigen_assert(Base::m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()");
       eigen_assert(Base::m_matrix.rows()==b.rows());
@@ -596,6 +622,13 @@
         dest = Base::m_Pinv * dest;
     }
     
+    /** \internal */
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const SparseMatrixBase<Rhs> &b, SparseMatrixBase<Dest> &dest) const
+    {
+      internal::solve_sparse_through_dense_panels(*this, b, dest);
+    }
+    
     Scalar determinant() const
     {
       if(m_LDLT)
@@ -614,58 +647,43 @@
 };
 
 template<typename Derived>
-void SimplicialCholeskyBase<Derived>::ordering(const MatrixType& a, CholMatrixType& ap)
+void SimplicialCholeskyBase<Derived>::ordering(const MatrixType& a, ConstCholMatrixPtr &pmat, CholMatrixType& ap)
 {
   eigen_assert(a.rows()==a.cols());
   const Index size = a.rows();
-  // Note that amd compute the inverse permutation
+  pmat = &ap;
+  // Note that ordering methods compute the inverse permutation
+  if(!internal::is_same<OrderingType,NaturalOrdering<Index> >::value)
   {
-    CholMatrixType C;
-    C = a.template selfadjointView<UpLo>();
+    {
+      CholMatrixType C;
+      C = a.template selfadjointView<UpLo>();
+      
+      OrderingType ordering;
+      ordering(C,m_Pinv);
+    }
+
+    if(m_Pinv.size()>0) m_P = m_Pinv.inverse();
+    else                m_P.resize(0);
     
-    OrderingType ordering;
-    ordering(C,m_Pinv);
+    ap.resize(size,size);
+    ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
   }
-
-  if(m_Pinv.size()>0)
-    m_P = m_Pinv.inverse();
   else
+  {
+    m_Pinv.resize(0);
     m_P.resize(0);
-
-  ap.resize(size,size);
-  ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>().twistedBy(m_P);
+    if(int(UpLo)==int(Lower) || MatrixType::IsRowMajor)
+    {
+      // we have to transpose the lower part to to the upper one
+      ap.resize(size,size);
+      ap.template selfadjointView<Upper>() = a.template selfadjointView<UpLo>();
+    }
+    else
+      internal::simplicial_cholesky_grab_input<CholMatrixType,MatrixType>::run(a, pmat, ap);
+  }  
 }
 
-namespace internal {
-  
-template<typename Derived, typename Rhs>
-struct solve_retval<SimplicialCholeskyBase<Derived>, Rhs>
-  : solve_retval_base<SimplicialCholeskyBase<Derived>, Rhs>
-{
-  typedef SimplicialCholeskyBase<Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec().derived()._solve(rhs(),dst);
-  }
-};
-
-template<typename Derived, typename Rhs>
-struct sparse_solve_retval<SimplicialCholeskyBase<Derived>, Rhs>
-  : sparse_solve_retval_base<SimplicialCholeskyBase<Derived>, Rhs>
-{
-  typedef SimplicialCholeskyBase<Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_SIMPLICIAL_CHOLESKY_H
diff --git a/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h b/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h
index 7aaf702..31e0699 100644
--- a/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h
+++ b/Eigen/src/SparseCholesky/SimplicialCholesky_impl.h
@@ -50,14 +50,14 @@
 template<typename Derived>
 void SimplicialCholeskyBase<Derived>::analyzePattern_preordered(const CholMatrixType& ap, bool doLDLT)
 {
-  const Index size = ap.rows();
+  const StorageIndex size = StorageIndex(ap.rows());
   m_matrix.resize(size, size);
   m_parent.resize(size);
   m_nonZerosPerCol.resize(size);
 
-  ei_declare_aligned_stack_constructed_variable(Index, tags, size, 0);
+  ei_declare_aligned_stack_constructed_variable(StorageIndex, tags, size, 0);
 
-  for(Index k = 0; k < size; ++k)
+  for(StorageIndex k = 0; k < size; ++k)
   {
     /* L(k,:) pattern: all nodes reachable in etree from nz in A(0:k-1,k) */
     m_parent[k] = -1;             /* parent of k is not yet known */
@@ -65,7 +65,7 @@
     m_nonZerosPerCol[k] = 0;      /* count of nonzeros in column k of L */
     for(typename CholMatrixType::InnerIterator it(ap,k); it; ++it)
     {
-      Index i = it.index();
+      StorageIndex i = it.index();
       if(i < k)
       {
         /* follow path from i to root of etree, stop at flagged node */
@@ -82,9 +82,9 @@
   }
 
   /* construct Lp index array from m_nonZerosPerCol column counts */
-  Index* Lp = m_matrix.outerIndexPtr();
+  StorageIndex* Lp = m_matrix.outerIndexPtr();
   Lp[0] = 0;
-  for(Index k = 0; k < size; ++k)
+  for(StorageIndex k = 0; k < size; ++k)
     Lp[k+1] = Lp[k] + m_nonZerosPerCol[k] + (doLDLT ? 0 : 1);
 
   m_matrix.resizeNonZeros(Lp[size]);
@@ -104,31 +104,31 @@
 
   eigen_assert(m_analysisIsOk && "You must first call analyzePattern()");
   eigen_assert(ap.rows()==ap.cols());
-  const Index size = ap.rows();
-  eigen_assert(m_parent.size()==size);
-  eigen_assert(m_nonZerosPerCol.size()==size);
+  eigen_assert(m_parent.size()==ap.rows());
+  eigen_assert(m_nonZerosPerCol.size()==ap.rows());
 
-  const Index* Lp = m_matrix.outerIndexPtr();
-  Index* Li = m_matrix.innerIndexPtr();
+  const StorageIndex size = StorageIndex(ap.rows());
+  const StorageIndex* Lp = m_matrix.outerIndexPtr();
+  StorageIndex* Li = m_matrix.innerIndexPtr();
   Scalar* Lx = m_matrix.valuePtr();
 
   ei_declare_aligned_stack_constructed_variable(Scalar, y, size, 0);
-  ei_declare_aligned_stack_constructed_variable(Index,  pattern, size, 0);
-  ei_declare_aligned_stack_constructed_variable(Index,  tags, size, 0);
+  ei_declare_aligned_stack_constructed_variable(StorageIndex,  pattern, size, 0);
+  ei_declare_aligned_stack_constructed_variable(StorageIndex,  tags, size, 0);
 
   bool ok = true;
   m_diag.resize(DoLDLT ? size : 0);
 
-  for(Index k = 0; k < size; ++k)
+  for(StorageIndex k = 0; k < size; ++k)
   {
     // compute nonzero pattern of kth row of L, in topological order
     y[k] = 0.0;                     // Y(0:k) is now all zero
-    Index top = size;               // stack for pattern is empty
+    StorageIndex top = size;               // stack for pattern is empty
     tags[k] = k;                    // mark node k as visited
     m_nonZerosPerCol[k] = 0;        // count of nonzeros in column k of L
-    for(typename MatrixType::InnerIterator it(ap,k); it; ++it)
+    for(typename CholMatrixType::InnerIterator it(ap,k); it; ++it)
     {
-      Index i = it.index();
+      StorageIndex i = it.index();
       if(i <= k)
       {
         y[i] += numext::conj(it.value());            /* scatter A(i,k) into Y (sum duplicates) */
diff --git a/Eigen/src/SparseCore/AmbiVector.h b/Eigen/src/SparseCore/AmbiVector.h
index 220c645..e0295f2 100644
--- a/Eigen/src/SparseCore/AmbiVector.h
+++ b/Eigen/src/SparseCore/AmbiVector.h
@@ -19,15 +19,15 @@
   *
   * See BasicSparseLLT and SparseProduct for usage examples.
   */
-template<typename _Scalar, typename _Index>
+template<typename _Scalar, typename _StorageIndex>
 class AmbiVector
 {
   public:
     typedef _Scalar Scalar;
-    typedef _Index Index;
+    typedef _StorageIndex StorageIndex;
     typedef typename NumTraits<Scalar>::Real RealScalar;
 
-    AmbiVector(Index size)
+    explicit AmbiVector(Index size)
       : m_buffer(0), m_zero(0), m_size(0), m_allocatedSize(0), m_allocatedElements(0), m_mode(-1)
     {
       resize(size);
@@ -39,7 +39,7 @@
     Index nonZeros() const;
 
     /** Specifies a sub-vector to work on */
-    void setBounds(Index start, Index end) { m_start = start; m_end = end; }
+    void setBounds(Index start, Index end) { m_start = convert_index(start); m_end = convert_index(end); }
 
     void setZero();
 
@@ -55,12 +55,16 @@
     {
       if (m_allocatedSize < size)
         reallocate(size);
-      m_size = size;
+      m_size = convert_index(size);
     }
 
-    Index size() const { return m_size; }
+    StorageIndex size() const { return m_size; }
 
   protected:
+    StorageIndex convert_index(Index idx)
+    {
+      return internal::convert_index<StorageIndex>(idx);
+    }
 
     void reallocate(Index size)
     {
@@ -70,15 +74,15 @@
       if (size<1000)
       {
         Index allocSize = (size * sizeof(ListEl) + sizeof(Scalar) - 1)/sizeof(Scalar);
-        m_allocatedElements = (allocSize*sizeof(Scalar))/sizeof(ListEl);
+        m_allocatedElements = convert_index((allocSize*sizeof(Scalar))/sizeof(ListEl));
         m_buffer = new Scalar[allocSize];
       }
       else
       {
-        m_allocatedElements = (size*sizeof(Scalar))/sizeof(ListEl);
+        m_allocatedElements = convert_index((size*sizeof(Scalar))/sizeof(ListEl));
         m_buffer = new Scalar[size];
       }
-      m_size = size;
+      m_size = convert_index(size);
       m_start = 0;
       m_end = m_size;
     }
@@ -86,11 +90,11 @@
     void reallocateSparse()
     {
       Index copyElements = m_allocatedElements;
-      m_allocatedElements = (std::min)(Index(m_allocatedElements*1.5),m_size);
+      m_allocatedElements = (std::min)(StorageIndex(m_allocatedElements*1.5),m_size);
       Index allocSize = m_allocatedElements * sizeof(ListEl);
       allocSize = (allocSize + sizeof(Scalar) - 1)/sizeof(Scalar);
       Scalar* newBuffer = new Scalar[allocSize];
-      memcpy(newBuffer,  m_buffer,  copyElements * sizeof(ListEl));
+      std::memcpy(newBuffer,  m_buffer,  copyElements * sizeof(ListEl));
       delete[] m_buffer;
       m_buffer = newBuffer;
     }
@@ -99,30 +103,30 @@
     // element type of the linked list
     struct ListEl
     {
-      Index next;
-      Index index;
+      StorageIndex next;
+      StorageIndex index;
       Scalar value;
     };
 
     // used to store data in both mode
     Scalar* m_buffer;
     Scalar m_zero;
-    Index m_size;
-    Index m_start;
-    Index m_end;
-    Index m_allocatedSize;
-    Index m_allocatedElements;
-    Index m_mode;
+    StorageIndex m_size;
+    StorageIndex m_start;
+    StorageIndex m_end;
+    StorageIndex m_allocatedSize;
+    StorageIndex m_allocatedElements;
+    StorageIndex m_mode;
 
     // linked list mode
-    Index m_llStart;
-    Index m_llCurrent;
-    Index m_llSize;
+    StorageIndex m_llStart;
+    StorageIndex m_llCurrent;
+    StorageIndex m_llSize;
 };
 
 /** \returns the number of non zeros in the current sub vector */
-template<typename _Scalar,typename _Index>
-_Index AmbiVector<_Scalar,_Index>::nonZeros() const
+template<typename _Scalar,typename _StorageIndex>
+Index AmbiVector<_Scalar,_StorageIndex>::nonZeros() const
 {
   if (m_mode==IsSparse)
     return m_llSize;
@@ -130,8 +134,8 @@
     return m_end - m_start;
 }
 
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::init(double estimatedDensity)
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::init(double estimatedDensity)
 {
   if (estimatedDensity>0.1)
     init(IsDense);
@@ -139,8 +143,8 @@
     init(IsSparse);
 }
 
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::init(int mode)
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::init(int mode)
 {
   m_mode = mode;
   if (m_mode==IsSparse)
@@ -155,15 +159,15 @@
   *
   * Don't worry, this function is extremely cheap.
   */
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::restart()
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::restart()
 {
   m_llCurrent = m_llStart;
 }
 
 /** Set all coefficients of current subvector to zero */
-template<typename _Scalar,typename _Index>
-void AmbiVector<_Scalar,_Index>::setZero()
+template<typename _Scalar,typename _StorageIndex>
+void AmbiVector<_Scalar,_StorageIndex>::setZero()
 {
   if (m_mode==IsDense)
   {
@@ -178,8 +182,8 @@
   }
 }
 
-template<typename _Scalar,typename _Index>
-_Scalar& AmbiVector<_Scalar,_Index>::coeffRef(_Index i)
+template<typename _Scalar,typename _StorageIndex>
+_Scalar& AmbiVector<_Scalar,_StorageIndex>::coeffRef(Index i)
 {
   if (m_mode==IsDense)
     return m_buffer[i];
@@ -195,7 +199,7 @@
       m_llCurrent = 0;
       ++m_llSize;
       llElements[0].value = Scalar(0);
-      llElements[0].index = i;
+      llElements[0].index = convert_index(i);
       llElements[0].next = -1;
       return llElements[0].value;
     }
@@ -204,7 +208,7 @@
       // this is going to be the new first element of the list
       ListEl& el = llElements[m_llSize];
       el.value = Scalar(0);
-      el.index = i;
+      el.index = convert_index(i);
       el.next = m_llStart;
       m_llStart = m_llSize;
       ++m_llSize;
@@ -213,7 +217,7 @@
     }
     else
     {
-      Index nextel = llElements[m_llCurrent].next;
+      StorageIndex nextel = llElements[m_llCurrent].next;
       eigen_assert(i>=llElements[m_llCurrent].index && "you must call restart() before inserting an element with lower or equal index");
       while (nextel >= 0 && llElements[nextel].index<=i)
       {
@@ -237,7 +241,7 @@
         // let's insert a new coefficient
         ListEl& el = llElements[m_llSize];
         el.value = Scalar(0);
-        el.index = i;
+        el.index = convert_index(i);
         el.next = llElements[m_llCurrent].next;
         llElements[m_llCurrent].next = m_llSize;
         ++m_llSize;
@@ -247,8 +251,8 @@
   }
 }
 
-template<typename _Scalar,typename _Index>
-_Scalar& AmbiVector<_Scalar,_Index>::coeff(_Index i)
+template<typename _Scalar,typename _StorageIndex>
+_Scalar& AmbiVector<_Scalar,_StorageIndex>::coeff(Index i)
 {
   if (m_mode==IsDense)
     return m_buffer[i];
@@ -275,8 +279,8 @@
 }
 
 /** Iterator over the nonzero coefficients */
-template<typename _Scalar,typename _Index>
-class AmbiVector<_Scalar,_Index>::Iterator
+template<typename _Scalar,typename _StorageIndex>
+class AmbiVector<_Scalar,_StorageIndex>::Iterator
 {
   public:
     typedef _Scalar Scalar;
@@ -288,7 +292,7 @@
       * In practice, all coefficients having a magnitude smaller than \a epsilon
       * are skipped.
       */
-    Iterator(const AmbiVector& vec, const RealScalar& epsilon = 0)
+    explicit Iterator(const AmbiVector& vec, const RealScalar& epsilon = 0)
       : m_vector(vec)
     {
       using std::abs;
@@ -320,7 +324,7 @@
       }
     }
 
-    Index index() const { return m_cachedIndex; }
+    StorageIndex index() const { return m_cachedIndex; }
     Scalar value() const { return m_cachedValue; }
 
     operator bool() const { return m_cachedIndex>=0; }
@@ -332,7 +336,7 @@
       {
         do {
           ++m_cachedIndex;
-        } while (m_cachedIndex<m_vector.m_end && abs(m_vector.m_buffer[m_cachedIndex])<m_epsilon);
+        } while (m_cachedIndex<m_vector.m_end && abs(m_vector.m_buffer[m_cachedIndex])<=m_epsilon);
         if (m_cachedIndex<m_vector.m_end)
           m_cachedValue = m_vector.m_buffer[m_cachedIndex];
         else
@@ -343,7 +347,7 @@
         ListEl* EIGEN_RESTRICT llElements = reinterpret_cast<ListEl*>(m_vector.m_buffer);
         do {
           m_currentEl = llElements[m_currentEl].next;
-        } while (m_currentEl>=0 && abs(llElements[m_currentEl].value)<m_epsilon);
+        } while (m_currentEl>=0 && abs(llElements[m_currentEl].value)<=m_epsilon);
         if (m_currentEl<0)
         {
           m_cachedIndex = -1;
@@ -359,9 +363,9 @@
 
   protected:
     const AmbiVector& m_vector; // the target vector
-    Index m_currentEl;            // the current element in sparse/linked-list mode
+    StorageIndex m_currentEl;   // the current element in sparse/linked-list mode
     RealScalar m_epsilon;       // epsilon used to prune zero coefficients
-    Index m_cachedIndex;          // current coordinate
+    StorageIndex m_cachedIndex; // current coordinate
     Scalar m_cachedValue;       // current value
     bool m_isDense;             // mode of the vector
 };
diff --git a/Eigen/src/SparseCore/CMakeLists.txt b/Eigen/src/SparseCore/CMakeLists.txt
deleted file mode 100644
index d860452..0000000
--- a/Eigen/src/SparseCore/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_SparseCore_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_SparseCore_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/SparseCore COMPONENT Devel
-  )
diff --git a/Eigen/src/SparseCore/CompressedStorage.h b/Eigen/src/SparseCore/CompressedStorage.h
index a667cb5..d89fa0d 100644
--- a/Eigen/src/SparseCore/CompressedStorage.h
+++ b/Eigen/src/SparseCore/CompressedStorage.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,13 +18,13 @@
   * Stores a sparse set of values as a list of values and a list of indices.
   *
   */
-template<typename _Scalar,typename _Index>
+template<typename _Scalar,typename _StorageIndex>
 class CompressedStorage
 {
   public:
 
     typedef _Scalar Scalar;
-    typedef _Index Index;
+    typedef _StorageIndex StorageIndex;
 
   protected:
 
@@ -36,7 +36,7 @@
       : m_values(0), m_indices(0), m_size(0), m_allocatedSize(0)
     {}
 
-    CompressedStorage(size_t size)
+    explicit CompressedStorage(Index size)
       : m_values(0), m_indices(0), m_size(0), m_allocatedSize(0)
     {
       resize(size);
@@ -51,8 +51,11 @@
     CompressedStorage& operator=(const CompressedStorage& other)
     {
       resize(other.size());
-      internal::smart_copy(other.m_values,  other.m_values  + m_size, m_values);
-      internal::smart_copy(other.m_indices, other.m_indices + m_size, m_indices);
+      if(other.size()>0)
+      {
+        internal::smart_copy(other.m_values,  other.m_values  + m_size, m_values);
+        internal::smart_copy(other.m_indices, other.m_indices + m_size, m_indices);
+      }
       return *this;
     }
 
@@ -70,9 +73,9 @@
       delete[] m_indices;
     }
 
-    void reserve(size_t size)
+    void reserve(Index size)
     {
-      size_t newAllocatedSize = m_size + size;
+      Index newAllocatedSize = m_size + size;
       if (newAllocatedSize > m_allocatedSize)
         reallocate(newAllocatedSize);
     }
@@ -83,39 +86,40 @@
         reallocate(m_size);
     }
 
-    void resize(size_t size, double reserveSizeFactor = 0)
+    void resize(Index size, double reserveSizeFactor = 0)
     {
       if (m_allocatedSize<size)
-        reallocate(size + size_t(reserveSizeFactor*double(size)));
+      {
+        Index realloc_size = (std::min<Index>)(NumTraits<StorageIndex>::highest(),  size + Index(reserveSizeFactor*double(size)));
+        if(realloc_size<size)
+          internal::throw_std_bad_alloc();
+        reallocate(realloc_size);
+      }
       m_size = size;
     }
 
     void append(const Scalar& v, Index i)
     {
-      Index id = static_cast<Index>(m_size);
+      Index id = m_size;
       resize(m_size+1, 1);
       m_values[id] = v;
-      m_indices[id] = i;
+      m_indices[id] = internal::convert_index<StorageIndex>(i);
     }
 
-    inline size_t size() const { return m_size; }
-    inline size_t allocatedSize() const { return m_allocatedSize; }
+    inline Index size() const { return m_size; }
+    inline Index allocatedSize() const { return m_allocatedSize; }
     inline void clear() { m_size = 0; }
 
-    inline Scalar& value(size_t i) { return m_values[i]; }
-    inline const Scalar& value(size_t i) const { return m_values[i]; }
+    const Scalar* valuePtr() const { return m_values; }
+    Scalar* valuePtr() { return m_values; }
+    const StorageIndex* indexPtr() const { return m_indices; }
+    StorageIndex* indexPtr() { return m_indices; }
 
-    inline Index& index(size_t i) { return m_indices[i]; }
-    inline const Index& index(size_t i) const { return m_indices[i]; }
+    inline Scalar& value(Index i) { eigen_internal_assert(m_values!=0); return m_values[i]; }
+    inline const Scalar& value(Index i) const { eigen_internal_assert(m_values!=0); return m_values[i]; }
 
-    static CompressedStorage Map(Index* indices, Scalar* values, size_t size)
-    {
-      CompressedStorage res;
-      res.m_indices = indices;
-      res.m_values = values;
-      res.m_allocatedSize = res.m_size = size;
-      return res;
-    }
+    inline StorageIndex& index(Index i) { eigen_internal_assert(m_indices!=0); return m_indices[i]; }
+    inline const StorageIndex& index(Index i) const { eigen_internal_assert(m_indices!=0); return m_indices[i]; }
 
     /** \returns the largest \c k such that for all \c j in [0,k) index[\c j]\<\a key */
     inline Index searchLowerIndex(Index key) const
@@ -124,17 +128,17 @@
     }
 
     /** \returns the largest \c k in [start,end) such that for all \c j in [start,k) index[\c j]\<\a key */
-    inline Index searchLowerIndex(size_t start, size_t end, Index key) const
+    inline Index searchLowerIndex(Index start, Index end, Index key) const
     {
       while(end>start)
       {
-        size_t mid = (end+start)>>1;
+        Index mid = (end+start)>>1;
         if (m_indices[mid]<key)
           start = mid+1;
         else
           end = mid;
       }
-      return static_cast<Index>(start);
+      return start;
     }
 
     /** \returns the stored value at index \a key
@@ -147,20 +151,20 @@
         return m_values[m_size-1];
       // ^^  optimization: let's first check if it is the last coefficient
       // (very common in high level algorithms)
-      const size_t id = searchLowerIndex(0,m_size-1,key);
+      const Index id = searchLowerIndex(0,m_size-1,key);
       return ((id<m_size) && (m_indices[id]==key)) ? m_values[id] : defaultValue;
     }
 
     /** Like at(), but the search is performed in the range [start,end) */
-    inline Scalar atInRange(size_t start, size_t end, Index key, const Scalar& defaultValue = Scalar(0)) const
+    inline Scalar atInRange(Index start, Index end, Index key, const Scalar &defaultValue = Scalar(0)) const
     {
       if (start>=end)
-        return Scalar(0);
+        return defaultValue;
       else if (end>start && key==m_indices[end-1])
         return m_values[end-1];
       // ^^  optimization: let's first check if it is the last coefficient
       // (very common in high level algorithms)
-      const size_t id = searchLowerIndex(start,end-1,key);
+      const Index id = searchLowerIndex(start,end-1,key);
       return ((id<end) && (m_indices[id]==key)) ? m_values[id] : defaultValue;
     }
 
@@ -169,16 +173,35 @@
       * such that the keys are sorted. */
     inline Scalar& atWithInsertion(Index key, const Scalar& defaultValue = Scalar(0))
     {
-      size_t id = searchLowerIndex(0,m_size,key);
+      Index id = searchLowerIndex(0,m_size,key);
       if (id>=m_size || m_indices[id]!=key)
       {
-        resize(m_size+1,1);
-        for (size_t j=m_size-1; j>id; --j)
+        if (m_allocatedSize<m_size+1)
         {
-          m_indices[j] = m_indices[j-1];
-          m_values[j] = m_values[j-1];
+          m_allocatedSize = 2*(m_size+1);
+          internal::scoped_array<Scalar> newValues(m_allocatedSize);
+          internal::scoped_array<StorageIndex> newIndices(m_allocatedSize);
+
+          // copy first chunk
+          internal::smart_copy(m_values,  m_values +id, newValues.ptr());
+          internal::smart_copy(m_indices, m_indices+id, newIndices.ptr());
+
+          // copy the rest
+          if(m_size>id)
+          {
+            internal::smart_copy(m_values +id,  m_values +m_size, newValues.ptr() +id+1);
+            internal::smart_copy(m_indices+id,  m_indices+m_size, newIndices.ptr()+id+1);
+          }
+          std::swap(m_values,newValues.ptr());
+          std::swap(m_indices,newIndices.ptr());
         }
-        m_indices[id] = key;
+        else if(m_size>id)
+        {
+          internal::smart_memmove(m_values +id, m_values +m_size, m_values +id+1);
+          internal::smart_memmove(m_indices+id, m_indices+m_size, m_indices+id+1);
+        }
+        m_size++;
+        m_indices[id] = internal::convert_index<StorageIndex>(key);
         m_values[id] = defaultValue;
       }
       return m_values[id];
@@ -186,9 +209,9 @@
 
     void prune(const Scalar& reference, const RealScalar& epsilon = NumTraits<RealScalar>::dummy_precision())
     {
-      size_t k = 0;
-      size_t n = size();
-      for (size_t i=0; i<n; ++i)
+      Index k = 0;
+      Index n = size();
+      for (Index i=0; i<n; ++i)
       {
         if (!internal::isMuchSmallerThan(value(i), reference, epsilon))
         {
@@ -202,27 +225,29 @@
 
   protected:
 
-    inline void reallocate(size_t size)
+    inline void reallocate(Index size)
     {
-      Scalar* newValues  = new Scalar[size];
-      Index* newIndices = new Index[size];
-      size_t copySize = (std::min)(size, m_size);
-      // copy
-      internal::smart_copy(m_values, m_values+copySize, newValues);
-      internal::smart_copy(m_indices, m_indices+copySize, newIndices);
-      // delete old stuff
-      delete[] m_values;
-      delete[] m_indices;
-      m_values = newValues;
-      m_indices = newIndices;
+      #ifdef EIGEN_SPARSE_COMPRESSED_STORAGE_REALLOCATE_PLUGIN
+        EIGEN_SPARSE_COMPRESSED_STORAGE_REALLOCATE_PLUGIN
+      #endif
+      eigen_internal_assert(size!=m_allocatedSize);
+      internal::scoped_array<Scalar> newValues(size);
+      internal::scoped_array<StorageIndex> newIndices(size);
+      Index copySize = (std::min)(size, m_size);
+      if (copySize>0) {
+        internal::smart_copy(m_values, m_values+copySize, newValues.ptr());
+        internal::smart_copy(m_indices, m_indices+copySize, newIndices.ptr());
+      }
+      std::swap(m_values,newValues.ptr());
+      std::swap(m_indices,newIndices.ptr());
       m_allocatedSize = size;
     }
 
   protected:
     Scalar* m_values;
-    Index* m_indices;
-    size_t m_size;
-    size_t m_allocatedSize;
+    StorageIndex* m_indices;
+    Index m_size;
+    Index m_allocatedSize;
 
 };
 
diff --git a/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h b/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h
index 5c320e2..9db119b 100644
--- a/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h
+++ b/Eigen/src/SparseCore/ConservativeSparseSparseProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -15,27 +15,33 @@
 namespace internal {
 
 template<typename Lhs, typename Rhs, typename ResultType>
-static void conservative_sparse_sparse_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+static void conservative_sparse_sparse_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res, bool sortedInsertion = false)
 {
-  typedef typename remove_all<Lhs>::type::Scalar Scalar;
-  typedef typename remove_all<Lhs>::type::Index Index;
+  typedef typename remove_all<Lhs>::type::Scalar LhsScalar;
+  typedef typename remove_all<Rhs>::type::Scalar RhsScalar;
+  typedef typename remove_all<ResultType>::type::Scalar ResScalar;
 
   // make sure to call innerSize/outerSize since we fake the storage order.
   Index rows = lhs.innerSize();
   Index cols = rhs.outerSize();
   eigen_assert(lhs.outerSize() == rhs.innerSize());
+  
+  ei_declare_aligned_stack_constructed_variable(bool,   mask,     rows, 0);
+  ei_declare_aligned_stack_constructed_variable(ResScalar, values,   rows, 0);
+  ei_declare_aligned_stack_constructed_variable(Index,  indices,  rows, 0);
+  
+  std::memset(mask,0,sizeof(bool)*rows);
 
-  std::vector<bool> mask(rows,false);
-  Matrix<Scalar,Dynamic,1> values(rows);
-  Matrix<Index,Dynamic,1>  indices(rows);
-
+  evaluator<Lhs> lhsEval(lhs);
+  evaluator<Rhs> rhsEval(rhs);
+  
   // estimate the number of non zero entries
   // given a rhs column containing Y non zeros, we assume that the respective Y columns
   // of the lhs differs in average of one non zeros, thus the number of non zeros for
   // the product of a rhs column with the lhs is X+Y where X is the average number of non zero
   // per column of the lhs.
   // Therefore, we have nnz(lhs*rhs) = nnz(lhs) + nnz(rhs)
-  Index estimated_nnz_prod = lhs.nonZeros() + rhs.nonZeros();
+  Index estimated_nnz_prod = lhsEval.nonZerosEstimate() + rhsEval.nonZerosEstimate();
 
   res.setZero();
   res.reserve(Index(estimated_nnz_prod));
@@ -45,14 +51,14 @@
 
     res.startVec(j);
     Index nnz = 0;
-    for (typename Rhs::InnerIterator rhsIt(rhs, j); rhsIt; ++rhsIt)
+    for (typename evaluator<Rhs>::InnerIterator rhsIt(rhsEval, j); rhsIt; ++rhsIt)
     {
-      Scalar y = rhsIt.value();
+      RhsScalar y = rhsIt.value();
       Index k = rhsIt.index();
-      for (typename Lhs::InnerIterator lhsIt(lhs, k); lhsIt; ++lhsIt)
+      for (typename evaluator<Lhs>::InnerIterator lhsIt(lhsEval, k); lhsIt; ++lhsIt)
       {
         Index i = lhsIt.index();
-        Scalar x = lhsIt.value();
+        LhsScalar x = lhsIt.value();
         if(!mask[i])
         {
           mask[i] = true;
@@ -64,53 +70,51 @@
           values[i] += x * y;
       }
     }
-
-    // unordered insertion
-    for(Index k=0; k<nnz; ++k)
+    if(!sortedInsertion)
     {
-      Index i = indices[k];
-      res.insertBackByOuterInnerUnordered(j,i) = values[i];
-      mask[i] = false;
-    }
-
-#if 0
-    // alternative ordered insertion code:
-
-    Index t200 = rows/(log2(200)*1.39);
-    Index t = (rows*100)/139;
-
-    // FIXME reserve nnz non zeros
-    // FIXME implement fast sort algorithms for very small nnz
-    // if the result is sparse enough => use a quick sort
-    // otherwise => loop through the entire vector
-    // In order to avoid to perform an expensive log2 when the
-    // result is clearly very sparse we use a linear bound up to 200.
-    //if((nnz<200 && nnz<t200) || nnz * log2(nnz) < t)
-    //res.startVec(j);
-    if(true)
-    {
-      if(nnz>1) std::sort(indices.data(),indices.data()+nnz);
+      // unordered insertion
       for(Index k=0; k<nnz; ++k)
       {
         Index i = indices[k];
-        res.insertBackByOuterInner(j,i) = values[i];
+        res.insertBackByOuterInnerUnordered(j,i) = values[i];
         mask[i] = false;
       }
     }
     else
     {
-      // dense path
-      for(Index i=0; i<rows; ++i)
+      // alternative ordered insertion code:
+      const Index t200 = rows/11; // 11 == (log2(200)*1.39)
+      const Index t = (rows*100)/139;
+
+      // FIXME reserve nnz non zeros
+      // FIXME implement faster sorting algorithms for very small nnz
+      // if the result is sparse enough => use a quick sort
+      // otherwise => loop through the entire vector
+      // In order to avoid to perform an expensive log2 when the
+      // result is clearly very sparse we use a linear bound up to 200.
+      if((nnz<200 && nnz<t200) || nnz * numext::log2(int(nnz)) < t)
       {
-        if(mask[i])
+        if(nnz>1) std::sort(indices,indices+nnz);
+        for(Index k=0; k<nnz; ++k)
         {
-          mask[i] = false;
+          Index i = indices[k];
           res.insertBackByOuterInner(j,i) = values[i];
+          mask[i] = false;
+        }
+      }
+      else
+      {
+        // dense path
+        for(Index i=0; i<rows; ++i)
+        {
+          if(mask[i])
+          {
+            mask[i] = false;
+            res.insertBackByOuterInner(j,i) = values[i];
+          }
         }
       }
     }
-#endif
-
   }
   res.finalize();
 }
@@ -134,13 +138,28 @@
 
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
-    ColMajorMatrix resCol(lhs.rows(),rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol);
-    // sort the non zeros:
-    RowMajorMatrix resRow(resCol);
-    res = resRow;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorMatrixAux;
+    typedef typename sparse_eval<ColMajorMatrixAux,ResultType::RowsAtCompileTime,ResultType::ColsAtCompileTime,ColMajorMatrixAux::Flags>::type ColMajorMatrix;
+    
+    // If the result is tall and thin (in the extreme case a column vector)
+    // then it is faster to sort the coefficients inplace instead of transposing twice.
+    // FIXME, the following heuristic is probably not very good.
+    if(lhs.rows()>rhs.cols())
+    {
+      ColMajorMatrix resCol(lhs.rows(),rhs.cols());
+      // perform sorted insertion
+      internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol, true);
+      res = resCol.markAsRValue();
+    }
+    else
+    {
+      ColMajorMatrixAux resCol(lhs.rows(),rhs.cols());
+      // ressort to transpose to sort the entries
+      internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrixAux>(lhs, rhs, resCol, false);
+      RowMajorMatrix resRow(resCol);
+      res = resRow.markAsRValue();
+    }
   }
 };
 
@@ -149,11 +168,12 @@
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-     typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-     RowMajorMatrix rhsRow = rhs;
-     RowMajorMatrix resRow(lhs.rows(), rhs.cols());
-     internal::conservative_sparse_sparse_product_impl<RowMajorMatrix,Lhs,RowMajorMatrix>(rhsRow, lhs, resRow);
-     res = resRow;
+    typedef SparseMatrix<typename Rhs::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorRhs;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorRes;
+    RowMajorRhs rhsRow = rhs;
+    RowMajorRes resRow(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<RowMajorRhs,Lhs,RowMajorRes>(rhsRow, lhs, resRow);
+    res = resRow;
   }
 };
 
@@ -162,10 +182,11 @@
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-    RowMajorMatrix lhsRow = lhs;
-    RowMajorMatrix resRow(lhs.rows(), rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<Rhs,RowMajorMatrix,RowMajorMatrix>(rhs, lhsRow, resRow);
+    typedef SparseMatrix<typename Lhs::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorLhs;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorRes;
+    RowMajorLhs lhsRow = lhs;
+    RowMajorRes resRow(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<Rhs,RowMajorLhs,RowMajorRes>(rhs, lhsRow, resRow);
     res = resRow;
   }
 };
@@ -175,7 +196,7 @@
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorMatrix;
     RowMajorMatrix resRow(lhs.rows(), rhs.cols());
     internal::conservative_sparse_sparse_product_impl<Rhs,Lhs,RowMajorMatrix>(rhs, lhs, resRow);
     res = resRow;
@@ -190,7 +211,7 @@
 
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorMatrix;
     ColMajorMatrix resCol(lhs.rows(), rhs.cols());
     internal::conservative_sparse_sparse_product_impl<Lhs,Rhs,ColMajorMatrix>(lhs, rhs, resCol);
     res = resCol;
@@ -202,10 +223,11 @@
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
-    ColMajorMatrix lhsCol = lhs;
-    ColMajorMatrix resCol(lhs.rows(), rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<ColMajorMatrix,Rhs,ColMajorMatrix>(lhsCol, rhs, resCol);
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorLhs;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRes;
+    ColMajorLhs lhsCol = lhs;
+    ColMajorRes resCol(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<ColMajorLhs,Rhs,ColMajorRes>(lhsCol, rhs, resCol);
     res = resCol;
   }
 };
@@ -215,10 +237,11 @@
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
-    ColMajorMatrix rhsCol = rhs;
-    ColMajorMatrix resCol(lhs.rows(), rhs.cols());
-    internal::conservative_sparse_sparse_product_impl<Lhs,ColMajorMatrix,ColMajorMatrix>(lhs, rhsCol, resCol);
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRhs;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRes;
+    ColMajorRhs rhsCol = rhs;
+    ColMajorRes resCol(lhs.rows(), rhs.cols());
+    internal::conservative_sparse_sparse_product_impl<Lhs,ColMajorRhs,ColMajorRes>(lhs, rhsCol, resCol);
     res = resCol;
   }
 };
@@ -228,8 +251,8 @@
 {
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::Index> RowMajorMatrix;
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> ColMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,RowMajor,typename ResultType::StorageIndex> RowMajorMatrix;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorMatrix;
     RowMajorMatrix resRow(lhs.rows(),rhs.cols());
     internal::conservative_sparse_sparse_product_impl<Rhs,Lhs,RowMajorMatrix>(rhs, lhs, resRow);
     // sort the non zeros:
@@ -240,6 +263,90 @@
 
 } // end namespace internal
 
+
+namespace internal {
+
+template<typename Lhs, typename Rhs, typename ResultType>
+static void sparse_sparse_to_dense_product_impl(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+{
+  typedef typename remove_all<Lhs>::type::Scalar LhsScalar;
+  typedef typename remove_all<Rhs>::type::Scalar RhsScalar;
+  Index cols = rhs.outerSize();
+  eigen_assert(lhs.outerSize() == rhs.innerSize());
+
+  evaluator<Lhs> lhsEval(lhs);
+  evaluator<Rhs> rhsEval(rhs);
+
+  for (Index j=0; j<cols; ++j)
+  {
+    for (typename evaluator<Rhs>::InnerIterator rhsIt(rhsEval, j); rhsIt; ++rhsIt)
+    {
+      RhsScalar y = rhsIt.value();
+      Index k = rhsIt.index();
+      for (typename evaluator<Lhs>::InnerIterator lhsIt(lhsEval, k); lhsIt; ++lhsIt)
+      {
+        Index i = lhsIt.index();
+        LhsScalar x = lhsIt.value();
+        res.coeffRef(i,j) += x * y;
+      }
+    }
+  }
+}
+
+
+} // end namespace internal
+
+namespace internal {
+
+template<typename Lhs, typename Rhs, typename ResultType,
+  int LhsStorageOrder = (traits<Lhs>::Flags&RowMajorBit) ? RowMajor : ColMajor,
+  int RhsStorageOrder = (traits<Rhs>::Flags&RowMajorBit) ? RowMajor : ColMajor>
+struct sparse_sparse_to_dense_product_selector;
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    internal::sparse_sparse_to_dense_product_impl<Lhs,Rhs,ResultType>(lhs, rhs, res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorLhs;
+    ColMajorLhs lhsCol(lhs);
+    internal::sparse_sparse_to_dense_product_impl<ColMajorLhs,Rhs,ResultType>(lhsCol, rhs, res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename ResultType::StorageIndex> ColMajorRhs;
+    ColMajorRhs rhsCol(rhs);
+    internal::sparse_sparse_to_dense_product_impl<Lhs,ColMajorRhs,ResultType>(lhs, rhsCol, res);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_to_dense_product_selector<Lhs,Rhs,ResultType,RowMajor,RowMajor>
+{
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res)
+  {
+    Transpose<ResultType> trRes(res);
+    internal::sparse_sparse_to_dense_product_impl<Rhs,Lhs,Transpose<ResultType> >(rhs, lhs, trRes);
+  }
+};
+
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_CONSERVATIVESPARSESPARSEPRODUCT_H
diff --git a/Eigen/src/SparseCore/MappedSparseMatrix.h b/Eigen/src/SparseCore/MappedSparseMatrix.h
index ab1a266..67718c8 100644
--- a/Eigen/src/SparseCore/MappedSparseMatrix.h
+++ b/Eigen/src/SparseCore/MappedSparseMatrix.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,9 +10,10 @@
 #ifndef EIGEN_MAPPED_SPARSEMATRIX_H
 #define EIGEN_MAPPED_SPARSEMATRIX_H
 
-namespace Eigen { 
+namespace Eigen {
 
-/** \class MappedSparseMatrix
+/** \deprecated Use Map<SparseMatrix<> >
+  * \class MappedSparseMatrix
   *
   * \brief Sparse matrix
   *
@@ -22,159 +23,44 @@
   *
   */
 namespace internal {
-template<typename _Scalar, int _Flags, typename _Index>
-struct traits<MappedSparseMatrix<_Scalar, _Flags, _Index> > : traits<SparseMatrix<_Scalar, _Flags, _Index> >
+template<typename _Scalar, int _Flags, typename _StorageIndex>
+struct traits<MappedSparseMatrix<_Scalar, _Flags, _StorageIndex> > : traits<SparseMatrix<_Scalar, _Flags, _StorageIndex> >
 {};
-}
+} // end namespace internal
 
-template<typename _Scalar, int _Flags, typename _Index>
+template<typename _Scalar, int _Flags, typename _StorageIndex>
 class MappedSparseMatrix
-  : public SparseMatrixBase<MappedSparseMatrix<_Scalar, _Flags, _Index> >
+  : public Map<SparseMatrix<_Scalar, _Flags, _StorageIndex> >
 {
-  public:
-    EIGEN_SPARSE_PUBLIC_INTERFACE(MappedSparseMatrix)
-    enum { IsRowMajor = Base::IsRowMajor };
-
-  protected:
-
-    Index   m_outerSize;
-    Index   m_innerSize;
-    Index   m_nnz;
-    Index*  m_outerIndex;
-    Index*  m_innerIndices;
-    Scalar* m_values;
+    typedef Map<SparseMatrix<_Scalar, _Flags, _StorageIndex> > Base;
 
   public:
-
-    inline Index rows() const { return IsRowMajor ? m_outerSize : m_innerSize; }
-    inline Index cols() const { return IsRowMajor ? m_innerSize : m_outerSize; }
-    inline Index innerSize() const { return m_innerSize; }
-    inline Index outerSize() const { return m_outerSize; }
     
-    bool isCompressed() const { return true; }
+    typedef typename Base::StorageIndex StorageIndex;
+    typedef typename Base::Scalar Scalar;
 
-    //----------------------------------------
-    // direct access interface
-    inline const Scalar* valuePtr() const { return m_values; }
-    inline Scalar* valuePtr() { return m_values; }
-
-    inline const Index* innerIndexPtr() const { return m_innerIndices; }
-    inline Index* innerIndexPtr() { return m_innerIndices; }
-
-    inline const Index* outerIndexPtr() const { return m_outerIndex; }
-    inline Index* outerIndexPtr() { return m_outerIndex; }
-    //----------------------------------------
-
-    inline Scalar coeff(Index row, Index col) const
-    {
-      const Index outer = IsRowMajor ? row : col;
-      const Index inner = IsRowMajor ? col : row;
-
-      Index start = m_outerIndex[outer];
-      Index end = m_outerIndex[outer+1];
-      if (start==end)
-        return Scalar(0);
-      else if (end>0 && inner==m_innerIndices[end-1])
-        return m_values[end-1];
-      // ^^  optimization: let's first check if it is the last coefficient
-      // (very common in high level algorithms)
-
-      const Index* r = std::lower_bound(&m_innerIndices[start],&m_innerIndices[end-1],inner);
-      const Index id = r-&m_innerIndices[0];
-      return ((*r==inner) && (id<end)) ? m_values[id] : Scalar(0);
-    }
-
-    inline Scalar& coeffRef(Index row, Index col)
-    {
-      const Index outer = IsRowMajor ? row : col;
-      const Index inner = IsRowMajor ? col : row;
-
-      Index start = m_outerIndex[outer];
-      Index end = m_outerIndex[outer+1];
-      eigen_assert(end>=start && "you probably called coeffRef on a non finalized matrix");
-      eigen_assert(end>start && "coeffRef cannot be called on a zero coefficient");
-      Index* r = std::lower_bound(&m_innerIndices[start],&m_innerIndices[end],inner);
-      const Index id = r-&m_innerIndices[0];
-      eigen_assert((*r==inner) && (id<end) && "coeffRef cannot be called on a zero coefficient");
-      return m_values[id];
-    }
-
-    class InnerIterator;
-    class ReverseInnerIterator;
-
-    /** \returns the number of non zero coefficients */
-    inline Index nonZeros() const  { return m_nnz; }
-
-    inline MappedSparseMatrix(Index rows, Index cols, Index nnz, Index* outerIndexPtr, Index* innerIndexPtr, Scalar* valuePtr)
-      : m_outerSize(IsRowMajor?rows:cols), m_innerSize(IsRowMajor?cols:rows), m_nnz(nnz), m_outerIndex(outerIndexPtr),
-        m_innerIndices(innerIndexPtr), m_values(valuePtr)
+    inline MappedSparseMatrix(Index rows, Index cols, Index nnz, StorageIndex* outerIndexPtr, StorageIndex* innerIndexPtr, Scalar* valuePtr, StorageIndex* innerNonZeroPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZeroPtr)
     {}
 
     /** Empty destructor */
     inline ~MappedSparseMatrix() {}
 };
 
-template<typename Scalar, int _Flags, typename _Index>
-class MappedSparseMatrix<Scalar,_Flags,_Index>::InnerIterator
+namespace internal {
+
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct evaluator<MappedSparseMatrix<_Scalar,_Options,_StorageIndex> >
+  : evaluator<SparseCompressedBase<MappedSparseMatrix<_Scalar,_Options,_StorageIndex> > >
 {
-  public:
-    InnerIterator(const MappedSparseMatrix& mat, Index outer)
-      : m_matrix(mat),
-        m_outer(outer),
-        m_id(mat.outerIndexPtr()[outer]),
-        m_start(m_id),
-        m_end(mat.outerIndexPtr()[outer+1])
-    {}
-
-    inline InnerIterator& operator++() { m_id++; return *this; }
-
-    inline Scalar value() const { return m_matrix.valuePtr()[m_id]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix.valuePtr()[m_id]); }
-
-    inline Index index() const { return m_matrix.innerIndexPtr()[m_id]; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id < m_end) && (m_id>=m_start); }
-
-  protected:
-    const MappedSparseMatrix& m_matrix;
-    const Index m_outer;
-    Index m_id;
-    const Index m_start;
-    const Index m_end;
+  typedef MappedSparseMatrix<_Scalar,_Options,_StorageIndex> XprType;
+  typedef evaluator<SparseCompressedBase<XprType> > Base;
+  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
 };
 
-template<typename Scalar, int _Flags, typename _Index>
-class MappedSparseMatrix<Scalar,_Flags,_Index>::ReverseInnerIterator
-{
-  public:
-    ReverseInnerIterator(const MappedSparseMatrix& mat, Index outer)
-      : m_matrix(mat),
-        m_outer(outer),
-        m_id(mat.outerIndexPtr()[outer+1]),
-        m_start(mat.outerIndexPtr()[outer]),
-        m_end(m_id)
-    {}
-
-    inline ReverseInnerIterator& operator--() { m_id--; return *this; }
-
-    inline Scalar value() const { return m_matrix.valuePtr()[m_id-1]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_matrix.valuePtr()[m_id-1]); }
-
-    inline Index index() const { return m_matrix.innerIndexPtr()[m_id-1]; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id <= m_end) && (m_id>m_start); }
-
-  protected:
-    const MappedSparseMatrix& m_matrix;
-    const Index m_outer;
-    Index m_id;
-    const Index m_start;
-    const Index m_end;
-};
+}
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/SparseCore/SparseAssign.h b/Eigen/src/SparseCore/SparseAssign.h
new file mode 100644
index 0000000..18352a8
--- /dev/null
+++ b/Eigen/src/SparseCore/SparseAssign.h
@@ -0,0 +1,216 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSEASSIGN_H
+#define EIGEN_SPARSEASSIGN_H
+
+namespace Eigen { 
+
+template<typename Derived>    
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator=(const EigenBase<OtherDerived> &other)
+{
+  internal::call_assignment_no_alias(derived(), other.derived());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator=(const ReturnByValue<OtherDerived>& other)
+{
+  // TODO use the evaluator mechanism
+  other.evalTo(derived());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+inline Derived& SparseMatrixBase<Derived>::operator=(const SparseMatrixBase<OtherDerived>& other)
+{
+  // by default sparse evaluation do not alias, so we can safely bypass the generic call_assignment routine
+  internal::Assignment<Derived,OtherDerived,internal::assign_op<Scalar,typename OtherDerived::Scalar> >
+          ::run(derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+inline Derived& SparseMatrixBase<Derived>::operator=(const Derived& other)
+{
+  internal::call_assignment_no_alias(derived(), other.derived());
+  return derived();
+}
+
+namespace internal {
+
+template<>
+struct storage_kind_to_evaluator_kind<Sparse> {
+  typedef IteratorBased Kind;
+};
+
+template<>
+struct storage_kind_to_shape<Sparse> {
+  typedef SparseShape Shape;
+};
+
+struct Sparse2Sparse {};
+struct Sparse2Dense  {};
+
+template<> struct AssignmentKind<SparseShape, SparseShape>           { typedef Sparse2Sparse Kind; };
+template<> struct AssignmentKind<SparseShape, SparseTriangularShape> { typedef Sparse2Sparse Kind; };
+template<> struct AssignmentKind<DenseShape,  SparseShape>           { typedef Sparse2Dense  Kind; };
+template<> struct AssignmentKind<DenseShape,  SparseTriangularShape> { typedef Sparse2Dense  Kind; };
+
+
+template<typename DstXprType, typename SrcXprType>
+void assign_sparse_to_sparse(DstXprType &dst, const SrcXprType &src)
+{
+  typedef typename DstXprType::Scalar Scalar;
+  typedef internal::evaluator<DstXprType> DstEvaluatorType;
+  typedef internal::evaluator<SrcXprType> SrcEvaluatorType;
+
+  SrcEvaluatorType srcEvaluator(src);
+
+  const bool transpose = (DstEvaluatorType::Flags & RowMajorBit) != (SrcEvaluatorType::Flags & RowMajorBit);
+  const Index outerEvaluationSize = (SrcEvaluatorType::Flags&RowMajorBit) ? src.rows() : src.cols();
+  if ((!transpose) && src.isRValue())
+  {
+    // eval without temporary
+    dst.resize(src.rows(), src.cols());
+    dst.setZero();
+    dst.reserve((std::max)(src.rows(),src.cols())*2);
+    for (Index j=0; j<outerEvaluationSize; ++j)
+    {
+      dst.startVec(j);
+      for (typename SrcEvaluatorType::InnerIterator it(srcEvaluator, j); it; ++it)
+      {
+        Scalar v = it.value();
+        dst.insertBackByOuterInner(j,it.index()) = v;
+      }
+    }
+    dst.finalize();
+  }
+  else
+  {
+    // eval through a temporary
+    eigen_assert(( ((internal::traits<DstXprType>::SupportedAccessPatterns & OuterRandomAccessPattern)==OuterRandomAccessPattern) ||
+              (!((DstEvaluatorType::Flags & RowMajorBit) != (SrcEvaluatorType::Flags & RowMajorBit)))) &&
+              "the transpose operation is supposed to be handled in SparseMatrix::operator=");
+
+    enum { Flip = (DstEvaluatorType::Flags & RowMajorBit) != (SrcEvaluatorType::Flags & RowMajorBit) };
+
+    
+    DstXprType temp(src.rows(), src.cols());
+
+    temp.reserve((std::max)(src.rows(),src.cols())*2);
+    for (Index j=0; j<outerEvaluationSize; ++j)
+    {
+      temp.startVec(j);
+      for (typename SrcEvaluatorType::InnerIterator it(srcEvaluator, j); it; ++it)
+      {
+        Scalar v = it.value();
+        temp.insertBackByOuterInner(Flip?it.index():j,Flip?j:it.index()) = v;
+      }
+    }
+    temp.finalize();
+
+    dst = temp.markAsRValue();
+  }
+}
+
+// Generic Sparse to Sparse assignment
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Sparse2Sparse>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    assign_sparse_to_sparse(dst.derived(), src.derived());
+  }
+};
+
+// Generic Sparse to Dense assignment
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Sparse2Dense>
+{
+  static void run(DstXprType &dst, const SrcXprType &src, const Functor &func)
+  {
+    if(internal::is_same<Functor,internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> >::value)
+      dst.setZero();
+    
+    internal::evaluator<SrcXprType> srcEval(src);
+    resize_if_allowed(dst, src, func);
+    internal::evaluator<DstXprType> dstEval(dst);
+    
+    const Index outerEvaluationSize = (internal::evaluator<SrcXprType>::Flags&RowMajorBit) ? src.rows() : src.cols();
+    for (Index j=0; j<outerEvaluationSize; ++j)
+      for (typename internal::evaluator<SrcXprType>::InnerIterator i(srcEval,j); i; ++i)
+        func.assignCoeff(dstEval.coeffRef(i.row(),i.col()), i.value());
+  }
+};
+
+// Specialization for "dst = dec.solve(rhs)"
+// NOTE we need to specialize it for Sparse2Sparse to avoid ambiguous specialization error
+template<typename DstXprType, typename DecType, typename RhsType, typename Scalar>
+struct Assignment<DstXprType, Solve<DecType,RhsType>, internal::assign_op<Scalar,Scalar>, Sparse2Sparse>
+{
+  typedef Solve<DecType,RhsType> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    src.dec()._solve_impl(src.rhs(), dst);
+  }
+};
+
+struct Diagonal2Sparse {};
+
+template<> struct AssignmentKind<SparseShape,DiagonalShape> { typedef Diagonal2Sparse Kind; };
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, Diagonal2Sparse>
+{
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  typedef typename DstXprType::Scalar Scalar;
+  typedef Array<StorageIndex,Dynamic,1> ArrayXI;
+  typedef Array<Scalar,Dynamic,1> ArrayXS;
+  template<int Options>
+  static void run(SparseMatrix<Scalar,Options,StorageIndex> &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+
+    Index size = src.diagonal().size();
+    dst.makeCompressed();
+    dst.resizeNonZeros(size);
+    Map<ArrayXI>(dst.innerIndexPtr(), size).setLinSpaced(0,StorageIndex(size)-1);
+    Map<ArrayXI>(dst.outerIndexPtr(), size+1).setLinSpaced(0,StorageIndex(size));
+    Map<ArrayXS>(dst.valuePtr(), size) = src.diagonal();
+  }
+  
+  template<typename DstDerived>
+  static void run(SparseMatrixBase<DstDerived> &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  {
+    dst.diagonal() = src.diagonal();
+  }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() += src.diagonal(); }
+  
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &/*func*/)
+  { dst.diagonal() -= src.diagonal(); }
+};
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSEASSIGN_H
diff --git a/Eigen/src/SparseCore/SparseBlock.h b/Eigen/src/SparseCore/SparseBlock.h
index 0c90baf..511e92b 100644
--- a/Eigen/src/SparseCore/SparseBlock.h
+++ b/Eigen/src/SparseCore/SparseBlock.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,8 +10,9 @@
 #ifndef EIGEN_SPARSE_BLOCK_H
 #define EIGEN_SPARSE_BLOCK_H
 
-namespace Eigen { 
+namespace Eigen {
 
+// Subset of columns or rows
 template<typename XprType, int BlockRows, int BlockCols>
 class BlockImpl<XprType,BlockRows,BlockCols,true,Sparse>
   : public SparseMatrixBase<Block<XprType,BlockRows,BlockCols,true> >
@@ -22,185 +23,189 @@
     enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
 protected:
     enum { OuterSize = IsRowMajor ? BlockRows : BlockCols };
+    typedef SparseMatrixBase<BlockType> Base;
+    using Base::convert_index;
 public:
     EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
-    
-    class InnerIterator: public XprType::InnerIterator
-    {
-        typedef typename BlockImpl::Index Index;
-      public:
-        inline InnerIterator(const BlockType& xpr, Index outer)
-          : XprType::InnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
-    class ReverseInnerIterator: public XprType::ReverseInnerIterator
-    {
-        typedef typename BlockImpl::Index Index;
-      public:
-        inline ReverseInnerIterator(const BlockType& xpr, Index outer)
-          : XprType::ReverseInnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
 
-    inline BlockImpl(const XprType& xpr, int i)
-      : m_matrix(xpr), m_outerStart(i), m_outerSize(OuterSize)
+    inline BlockImpl(XprType& xpr, Index i)
+      : m_matrix(xpr), m_outerStart(convert_index(i)), m_outerSize(OuterSize)
     {}
 
-    inline BlockImpl(const XprType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_outerStart(IsRowMajor ? startRow : startCol), m_outerSize(IsRowMajor ? blockRows : blockCols)
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : m_matrix(xpr), m_outerStart(convert_index(IsRowMajor ? startRow : startCol)), m_outerSize(convert_index(IsRowMajor ? blockRows : blockCols))
     {}
-    
-    inline const Scalar coeff(int row, int col) const
-    {
-      return m_matrix.coeff(row + IsRowMajor ? m_outerStart : 0, col +IsRowMajor ? 0 :  m_outerStart);
-    }
-    
-    inline const Scalar coeff(int index) const
-    {
-      return m_matrix.coeff(IsRowMajor ? m_outerStart : index, IsRowMajor ? index :  m_outerStart);
-    }
 
     EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
     EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
 
+    Index nonZeros() const
+    {
+      typedef internal::evaluator<XprType> EvaluatorType;
+      EvaluatorType matEval(m_matrix);
+      Index nnz = 0;
+      Index end = m_outerStart + m_outerSize.value();
+      for(Index j=m_outerStart; j<end; ++j)
+        for(typename EvaluatorType::InnerIterator it(matEval, j); it; ++it)
+          ++nnz;
+      return nnz;
+    }
+
+    inline const Scalar coeff(Index row, Index col) const
+    {
+      return m_matrix.coeff(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
+    }
+
+    inline const Scalar coeff(Index index) const
+    {
+      return m_matrix.coeff(IsRowMajor ? m_outerStart : index, IsRowMajor ? index :  m_outerStart);
+    }
+
+    inline const XprType& nestedExpression() const { return m_matrix; }
+    inline XprType& nestedExpression() { return m_matrix; }
+    Index startRow() const { return IsRowMajor ? m_outerStart : 0; }
+    Index startCol() const { return IsRowMajor ? 0 : m_outerStart; }
+    Index blockRows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
+    Index blockCols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
+
   protected:
 
-    typename XprType::Nested m_matrix;
+    typename internal::ref_selector<XprType>::non_const_type m_matrix;
     Index m_outerStart;
     const internal::variable_if_dynamic<Index, OuterSize> m_outerSize;
 
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
-  private:
-    Index nonZeros() const;
+  protected:
+    // Disable assignment with clear error message.
+    // Note that simply removing operator= yields compilation errors with ICC+MSVC
+    template<typename T>
+    BlockImpl& operator=(const T&)
+    {
+      EIGEN_STATIC_ASSERT(sizeof(T)==0, THIS_SPARSE_BLOCK_SUBEXPRESSION_IS_READ_ONLY);
+      return *this;
+    }
 };
 
 
 /***************************************************************************
-* specialisation for SparseMatrix
+* specialization for SparseMatrix
 ***************************************************************************/
 
-template<typename _Scalar, int _Options, typename _Index, int BlockRows, int BlockCols>
-class BlockImpl<SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true,Sparse>
-  : public SparseMatrixBase<Block<SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true> >
+namespace internal {
+
+template<typename SparseMatrixType, int BlockRows, int BlockCols>
+class sparse_matrix_block_impl
+  : public SparseCompressedBase<Block<SparseMatrixType,BlockRows,BlockCols,true> >
 {
-    typedef SparseMatrix<_Scalar, _Options, _Index> SparseMatrixType;
     typedef typename internal::remove_all<typename SparseMatrixType::Nested>::type _MatrixTypeNested;
     typedef Block<SparseMatrixType, BlockRows, BlockCols, true> BlockType;
-    typedef Block<const SparseMatrixType, BlockRows, BlockCols, true> ConstBlockType;
+    typedef SparseCompressedBase<Block<SparseMatrixType,BlockRows,BlockCols,true> > Base;
+    using Base::convert_index;
 public:
     enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
     EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
 protected:
+    typedef typename Base::IndexVector IndexVector;
     enum { OuterSize = IsRowMajor ? BlockRows : BlockCols };
 public:
-    
-    class InnerIterator: public SparseMatrixType::InnerIterator
-    {
-      public:
-        inline InnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::InnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
-    class ReverseInnerIterator: public SparseMatrixType::ReverseInnerIterator
-    {
-      public:
-        inline ReverseInnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::ReverseInnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
 
-    inline BlockImpl(const SparseMatrixType& xpr, int i)
-      : m_matrix(xpr), m_outerStart(i), m_outerSize(OuterSize)
+    inline sparse_matrix_block_impl(SparseMatrixType& xpr, Index i)
+      : m_matrix(xpr), m_outerStart(convert_index(i)), m_outerSize(OuterSize)
     {}
 
-    inline BlockImpl(const SparseMatrixType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_outerStart(IsRowMajor ? startRow : startCol), m_outerSize(IsRowMajor ? blockRows : blockCols)
+    inline sparse_matrix_block_impl(SparseMatrixType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : m_matrix(xpr), m_outerStart(convert_index(IsRowMajor ? startRow : startCol)), m_outerSize(convert_index(IsRowMajor ? blockRows : blockCols))
     {}
 
     template<typename OtherDerived>
     inline BlockType& operator=(const SparseMatrixBase<OtherDerived>& other)
     {
       typedef typename internal::remove_all<typename SparseMatrixType::Nested>::type _NestedMatrixType;
-      _NestedMatrixType& matrix = const_cast<_NestedMatrixType&>(m_matrix);;
-      // This assignement is slow if this vector set is not empty
+      _NestedMatrixType& matrix = m_matrix;
+      // This assignment is slow if this vector set is not empty
       // and/or it is not at the end of the nonzeros of the underlying matrix.
 
       // 1 - eval to a temporary to avoid transposition and/or aliasing issues
-      SparseMatrix<Scalar, IsRowMajor ? RowMajor : ColMajor, Index> tmp(other);
+      Ref<const SparseMatrix<Scalar, IsRowMajor ? RowMajor : ColMajor, StorageIndex> > tmp(other.derived());
+      eigen_internal_assert(tmp.outerSize()==m_outerSize.value());
 
       // 2 - let's check whether there is enough allocated memory
       Index nnz           = tmp.nonZeros();
-      Index start         = m_outerStart==0 ? 0 : matrix.outerIndexPtr()[m_outerStart]; // starting position of the current block
-      Index end           = m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()]; // ending posiiton of the current block
+      Index start         = m_outerStart==0 ? 0 : m_matrix.outerIndexPtr()[m_outerStart]; // starting position of the current block
+      Index end           = m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()]; // ending position of the current block
       Index block_size    = end - start;                                                // available room in the current block
       Index tail_size     = m_matrix.outerIndexPtr()[m_matrix.outerSize()] - end;
-      
+
       Index free_size     = m_matrix.isCompressed()
                           ? Index(matrix.data().allocatedSize()) + block_size
                           : block_size;
 
-      if(nnz>free_size) 
+      Index tmp_start = tmp.outerIndexPtr()[0];
+
+      bool update_trailing_pointers = false;
+      if(nnz>free_size)
       {
         // realloc manually to reduce copies
         typename SparseMatrixType::Storage newdata(m_matrix.data().allocatedSize() - block_size + nnz);
 
-        std::memcpy(&newdata.value(0), &m_matrix.data().value(0), start*sizeof(Scalar));
-        std::memcpy(&newdata.index(0), &m_matrix.data().index(0), start*sizeof(Index));
+        internal::smart_copy(m_matrix.valuePtr(),       m_matrix.valuePtr() + start,      newdata.valuePtr());
+        internal::smart_copy(m_matrix.innerIndexPtr(),  m_matrix.innerIndexPtr() + start, newdata.indexPtr());
 
-        std::memcpy(&newdata.value(start), &tmp.data().value(0), nnz*sizeof(Scalar));
-        std::memcpy(&newdata.index(start), &tmp.data().index(0), nnz*sizeof(Index));
+        internal::smart_copy(tmp.valuePtr() + tmp_start,      tmp.valuePtr() + tmp_start + nnz,       newdata.valuePtr() + start);
+        internal::smart_copy(tmp.innerIndexPtr() + tmp_start, tmp.innerIndexPtr() + tmp_start + nnz,  newdata.indexPtr() + start);
 
-        std::memcpy(&newdata.value(start+nnz), &matrix.data().value(end), tail_size*sizeof(Scalar));
-        std::memcpy(&newdata.index(start+nnz), &matrix.data().index(end), tail_size*sizeof(Index));
-        
+        internal::smart_copy(matrix.valuePtr()+end,       matrix.valuePtr()+end + tail_size,      newdata.valuePtr()+start+nnz);
+        internal::smart_copy(matrix.innerIndexPtr()+end,  matrix.innerIndexPtr()+end + tail_size, newdata.indexPtr()+start+nnz);
+
         newdata.resize(m_matrix.outerIndexPtr()[m_matrix.outerSize()] - block_size + nnz);
 
         matrix.data().swap(newdata);
+
+        update_trailing_pointers = true;
       }
       else
       {
-        // no need to realloc, simply copy the tail at its respective position and insert tmp
-        matrix.data().resize(start + nnz + tail_size);
+        if(m_matrix.isCompressed())
+        {
+          // no need to realloc, simply copy the tail at its respective position and insert tmp
+          matrix.data().resize(start + nnz + tail_size);
 
-        std::memmove(&matrix.data().value(start+nnz), &matrix.data().value(end), tail_size*sizeof(Scalar));
-        std::memmove(&matrix.data().index(start+nnz), &matrix.data().index(end), tail_size*sizeof(Index));
+          internal::smart_memmove(matrix.valuePtr()+end,      matrix.valuePtr() + end+tail_size,      matrix.valuePtr() + start+nnz);
+          internal::smart_memmove(matrix.innerIndexPtr()+end, matrix.innerIndexPtr() + end+tail_size, matrix.innerIndexPtr() + start+nnz);
 
-        std::memcpy(&matrix.data().value(start), &tmp.data().value(0), nnz*sizeof(Scalar));
-        std::memcpy(&matrix.data().index(start), &tmp.data().index(0), nnz*sizeof(Index));
+          update_trailing_pointers = true;
+        }
+
+        internal::smart_copy(tmp.valuePtr() + tmp_start,      tmp.valuePtr() + tmp_start + nnz,       matrix.valuePtr() + start);
+        internal::smart_copy(tmp.innerIndexPtr() + tmp_start, tmp.innerIndexPtr() + tmp_start + nnz,  matrix.innerIndexPtr() + start);
       }
-      
-      // update innerNonZeros
-      if(!m_matrix.isCompressed())
-        for(Index j=0; j<m_outerSize.value(); ++j)
-          matrix.innerNonZeroPtr()[m_outerStart+j] = tmp.innerVector(j).nonZeros();
 
-      // update outer index pointers
-      Index p = start;
-      for(Index k=0; k<m_outerSize.value(); ++k)
+      // update outer index pointers and innerNonZeros
+      if(IsVectorAtCompileTime)
       {
-        matrix.outerIndexPtr()[m_outerStart+k] = p;
-        p += tmp.innerVector(k).nonZeros();
+        if(!m_matrix.isCompressed())
+          matrix.innerNonZeroPtr()[m_outerStart] = StorageIndex(nnz);
+        matrix.outerIndexPtr()[m_outerStart] = StorageIndex(start);
       }
-      std::ptrdiff_t offset = nnz - block_size;
-      for(Index k = m_outerStart + m_outerSize.value(); k<=matrix.outerSize(); ++k)
+      else
       {
-        matrix.outerIndexPtr()[k] += offset;
+        StorageIndex p = StorageIndex(start);
+        for(Index k=0; k<m_outerSize.value(); ++k)
+        {
+          StorageIndex nnz_k = internal::convert_index<StorageIndex>(tmp.innerVector(k).nonZeros());
+          if(!m_matrix.isCompressed())
+            matrix.innerNonZeroPtr()[m_outerStart+k] = nnz_k;
+          matrix.outerIndexPtr()[m_outerStart+k] = p;
+          p += nnz_k;
+        }
+      }
+
+      if(update_trailing_pointers)
+      {
+        StorageIndex offset = internal::convert_index<StorageIndex>(nnz - block_size);
+        for(Index k = m_outerStart + m_outerSize.value(); k<=matrix.outerSize(); ++k)
+        {
+          matrix.outerIndexPtr()[k] += offset;
+        }
       }
 
       return derived();
@@ -212,50 +217,46 @@
     }
 
     inline const Scalar* valuePtr() const
-    { return m_matrix.valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
+    { return m_matrix.valuePtr(); }
     inline Scalar* valuePtr()
-    { return m_matrix.const_cast_derived().valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
+    { return m_matrix.valuePtr(); }
 
-    inline const Index* innerIndexPtr() const
-    { return m_matrix.innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
-    inline Index* innerIndexPtr()
-    { return m_matrix.const_cast_derived().innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
+    inline const StorageIndex* innerIndexPtr() const
+    { return m_matrix.innerIndexPtr(); }
+    inline StorageIndex* innerIndexPtr()
+    { return m_matrix.innerIndexPtr(); }
 
-    inline const Index* outerIndexPtr() const
+    inline const StorageIndex* outerIndexPtr() const
     { return m_matrix.outerIndexPtr() + m_outerStart; }
-    inline Index* outerIndexPtr()
-    { return m_matrix.const_cast_derived().outerIndexPtr() + m_outerStart; }
+    inline StorageIndex* outerIndexPtr()
+    { return m_matrix.outerIndexPtr() + m_outerStart; }
 
-    Index nonZeros() const
+    inline const StorageIndex* innerNonZeroPtr() const
+    { return isCompressed() ? 0 : (m_matrix.innerNonZeroPtr()+m_outerStart); }
+    inline StorageIndex* innerNonZeroPtr()
+    { return isCompressed() ? 0 : (m_matrix.innerNonZeroPtr()+m_outerStart); }
+
+    bool isCompressed() const { return m_matrix.innerNonZeroPtr()==0; }
+
+    inline Scalar& coeffRef(Index row, Index col)
     {
-      if(m_matrix.isCompressed())
-        return  std::size_t(m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()])
-              - std::size_t(m_matrix.outerIndexPtr()[m_outerStart]);
-      else if(m_outerSize.value()==0)
-        return 0;
-      else
-        return Map<const Matrix<Index,OuterSize,1> >(m_matrix.innerNonZeroPtr()+m_outerStart, m_outerSize.value()).sum();
+      return m_matrix.coeffRef(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
     }
-    
-    inline Scalar& coeffRef(int row, int col)
-    {
-      return m_matrix.const_cast_derived().coeffRef(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
-    }
-    
-    inline const Scalar coeff(int row, int col) const
+
+    inline const Scalar coeff(Index row, Index col) const
     {
       return m_matrix.coeff(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
     }
-    
-    inline const Scalar coeff(int index) const
+
+    inline const Scalar coeff(Index index) const
     {
       return m_matrix.coeff(IsRowMajor ? m_outerStart : index, IsRowMajor ? index :  m_outerStart);
     }
 
     const Scalar& lastCoeff() const
     {
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(BlockImpl);
-      eigen_assert(nonZeros()>0);
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(sparse_matrix_block_impl);
+      eigen_assert(Base::nonZeros()>0);
       if(m_matrix.isCompressed())
         return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart+1]-1];
       else
@@ -265,109 +266,62 @@
     EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
     EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
 
+    inline const SparseMatrixType& nestedExpression() const { return m_matrix; }
+    inline SparseMatrixType& nestedExpression() { return m_matrix; }
+    Index startRow() const { return IsRowMajor ? m_outerStart : 0; }
+    Index startCol() const { return IsRowMajor ? 0 : m_outerStart; }
+    Index blockRows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
+    Index blockCols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
+
   protected:
 
-    typename SparseMatrixType::Nested m_matrix;
+    typename internal::ref_selector<SparseMatrixType>::non_const_type m_matrix;
     Index m_outerStart;
     const internal::variable_if_dynamic<Index, OuterSize> m_outerSize;
 
 };
 
+} // namespace internal
 
-template<typename _Scalar, int _Options, typename _Index, int BlockRows, int BlockCols>
-class BlockImpl<const SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true,Sparse>
-  : public SparseMatrixBase<Block<const SparseMatrix<_Scalar, _Options, _Index>,BlockRows,BlockCols,true> >
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+class BlockImpl<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true,Sparse>
+  : public internal::sparse_matrix_block_impl<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols>
 {
-    typedef SparseMatrix<_Scalar, _Options, _Index> SparseMatrixType;
-    typedef typename internal::remove_all<typename SparseMatrixType::Nested>::type _MatrixTypeNested;
-    typedef Block<const SparseMatrixType, BlockRows, BlockCols, true> BlockType;
 public:
-    enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
-    EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
-protected:
-    enum { OuterSize = IsRowMajor ? BlockRows : BlockCols };
+  typedef _StorageIndex StorageIndex;
+  typedef SparseMatrix<_Scalar, _Options, _StorageIndex> SparseMatrixType;
+  typedef internal::sparse_matrix_block_impl<SparseMatrixType,BlockRows,BlockCols> Base;
+  inline BlockImpl(SparseMatrixType& xpr, Index i)
+    : Base(xpr, i)
+  {}
+
+  inline BlockImpl(SparseMatrixType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+    : Base(xpr, startRow, startCol, blockRows, blockCols)
+  {}
+
+  using Base::operator=;
+};
+
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+class BlockImpl<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true,Sparse>
+  : public internal::sparse_matrix_block_impl<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols>
+{
 public:
-    
-    class InnerIterator: public SparseMatrixType::InnerIterator
-    {
-      public:
-        inline InnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::InnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
-    class ReverseInnerIterator: public SparseMatrixType::ReverseInnerIterator
-    {
-      public:
-        inline ReverseInnerIterator(const BlockType& xpr, Index outer)
-          : SparseMatrixType::ReverseInnerIterator(xpr.m_matrix, xpr.m_outerStart + outer), m_outer(outer)
-        {}
-        inline Index row() const { return IsRowMajor ? m_outer : this->index(); }
-        inline Index col() const { return IsRowMajor ? this->index() : m_outer; }
-      protected:
-        Index m_outer;
-    };
+  typedef _StorageIndex StorageIndex;
+  typedef const SparseMatrix<_Scalar, _Options, _StorageIndex> SparseMatrixType;
+  typedef internal::sparse_matrix_block_impl<SparseMatrixType,BlockRows,BlockCols> Base;
+  inline BlockImpl(SparseMatrixType& xpr, Index i)
+    : Base(xpr, i)
+  {}
 
-    inline BlockImpl(const SparseMatrixType& xpr, int i)
-      : m_matrix(xpr), m_outerStart(i), m_outerSize(OuterSize)
-    {}
+  inline BlockImpl(SparseMatrixType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+    : Base(xpr, startRow, startCol, blockRows, blockCols)
+  {}
 
-    inline BlockImpl(const SparseMatrixType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_outerStart(IsRowMajor ? startRow : startCol), m_outerSize(IsRowMajor ? blockRows : blockCols)
-    {}
-
-    inline const Scalar* valuePtr() const
-    { return m_matrix.valuePtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
-
-    inline const Index* innerIndexPtr() const
-    { return m_matrix.innerIndexPtr() + m_matrix.outerIndexPtr()[m_outerStart]; }
-
-    inline const Index* outerIndexPtr() const
-    { return m_matrix.outerIndexPtr() + m_outerStart; }
-
-    Index nonZeros() const
-    {
-      if(m_matrix.isCompressed())
-        return  std::size_t(m_matrix.outerIndexPtr()[m_outerStart+m_outerSize.value()])
-              - std::size_t(m_matrix.outerIndexPtr()[m_outerStart]);
-      else if(m_outerSize.value()==0)
-        return 0;
-      else
-        return Map<const Matrix<Index,OuterSize,1> >(m_matrix.innerNonZeroPtr()+m_outerStart, m_outerSize.value()).sum();
-    }
-    
-    inline const Scalar coeff(int row, int col) const
-    {
-      return m_matrix.coeff(row + (IsRowMajor ? m_outerStart : 0), col + (IsRowMajor ? 0 :  m_outerStart));
-    }
-    
-    inline const Scalar coeff(int index) const
-    {
-      return m_matrix.coeff(IsRowMajor ? m_outerStart : index, IsRowMajor ? index :  m_outerStart);
-    }
-
-    const Scalar& lastCoeff() const
-    {
-      EIGEN_STATIC_ASSERT_VECTOR_ONLY(BlockImpl);
-      eigen_assert(nonZeros()>0);
-      if(m_matrix.isCompressed())
-        return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart+1]-1];
-      else
-        return m_matrix.valuePtr()[m_matrix.outerIndexPtr()[m_outerStart]+m_matrix.innerNonZeroPtr()[m_outerStart]-1];
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return IsRowMajor ? m_outerSize.value() : m_matrix.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return IsRowMajor ? m_matrix.cols() : m_outerSize.value(); }
-
-  protected:
-
-    typename SparseMatrixType::Nested m_matrix;
-    Index m_outerStart;
-    const internal::variable_if_dynamic<Index, OuterSize> m_outerSize;
-
+  using Base::operator=;
+private:
+  template<typename Derived> BlockImpl(const SparseMatrixBase<Derived>& xpr, Index i);
+  template<typename Derived> BlockImpl(const SparseMatrixBase<Derived>& xpr);
 };
 
 //----------
@@ -396,7 +350,7 @@
   return Block<Derived,Dynamic,Dynamic,true>(derived(),
                                              IsRowMajor ? outerStart : 0, IsRowMajor ? 0 : outerStart,
                                              IsRowMajor ? outerSize : rows(), IsRowMajor ? cols() : outerSize);
-  
+
 }
 
 /** \returns the \a outer -th column (resp. row) of the matrix \c *this if \c *this
@@ -409,129 +363,241 @@
   return Block<const Derived,Dynamic,Dynamic,true>(derived(),
                                                   IsRowMajor ? outerStart : 0, IsRowMajor ? 0 : outerStart,
                                                   IsRowMajor ? outerSize : rows(), IsRowMajor ? cols() : outerSize);
-  
+
 }
 
 /** Generic implementation of sparse Block expression.
-  * Real-only. 
+  * Real-only.
   */
 template<typename XprType, int BlockRows, int BlockCols, bool InnerPanel>
 class BlockImpl<XprType,BlockRows,BlockCols,InnerPanel,Sparse>
   : public SparseMatrixBase<Block<XprType,BlockRows,BlockCols,InnerPanel> >, internal::no_assignment_operator
 {
-  typedef typename internal::remove_all<typename XprType::Nested>::type _MatrixTypeNested;
-  typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef Block<XprType, BlockRows, BlockCols, InnerPanel> BlockType;
+    typedef SparseMatrixBase<BlockType> Base;
+    using Base::convert_index;
 public:
     enum { IsRowMajor = internal::traits<BlockType>::IsRowMajor };
     EIGEN_SPARSE_PUBLIC_INTERFACE(BlockType)
 
+    typedef typename internal::remove_all<typename XprType::Nested>::type _MatrixTypeNested;
+
     /** Column or Row constructor
       */
-    inline BlockImpl(const XprType& xpr, int i)
+    inline BlockImpl(XprType& xpr, Index i)
       : m_matrix(xpr),
-        m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0),
-        m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0),
+        m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? convert_index(i) : 0),
+        m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? convert_index(i) : 0),
         m_blockRows(BlockRows==1 ? 1 : xpr.rows()),
         m_blockCols(BlockCols==1 ? 1 : xpr.cols())
     {}
 
     /** Dynamic-size constructor
       */
-    inline BlockImpl(const XprType& xpr, int startRow, int startCol, int blockRows, int blockCols)
-      : m_matrix(xpr), m_startRow(startRow), m_startCol(startCol), m_blockRows(blockRows), m_blockCols(blockCols)
+    inline BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols)
+      : m_matrix(xpr), m_startRow(convert_index(startRow)), m_startCol(convert_index(startCol)), m_blockRows(convert_index(blockRows)), m_blockCols(convert_index(blockCols))
     {}
 
-    inline int rows() const { return m_blockRows.value(); }
-    inline int cols() const { return m_blockCols.value(); }
+    inline Index rows() const { return m_blockRows.value(); }
+    inline Index cols() const { return m_blockCols.value(); }
 
-    inline Scalar& coeffRef(int row, int col)
+    inline Scalar& coeffRef(Index row, Index col)
     {
-      return m_matrix.const_cast_derived()
-               .coeffRef(row + m_startRow.value(), col + m_startCol.value());
+      return m_matrix.coeffRef(row + m_startRow.value(), col + m_startCol.value());
     }
 
-    inline const Scalar coeff(int row, int col) const
+    inline const Scalar coeff(Index row, Index col) const
     {
       return m_matrix.coeff(row + m_startRow.value(), col + m_startCol.value());
     }
 
-    inline Scalar& coeffRef(int index)
+    inline Scalar& coeffRef(Index index)
     {
-      return m_matrix.const_cast_derived()
-             .coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                       m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_matrix.coeffRef(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                               m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
 
-    inline const Scalar coeff(int index) const
+    inline const Scalar coeff(Index index) const
     {
-      return m_matrix
-             .coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
-                    m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
+      return m_matrix.coeff(m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index),
+                            m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0));
     }
-    
-    inline const _MatrixTypeNested& nestedExpression() const { return m_matrix; }
-    
-    class InnerIterator : public _MatrixTypeNested::InnerIterator
-    {
-      typedef typename _MatrixTypeNested::InnerIterator Base;
-      const BlockType& m_block;
-      Index m_end;
-    public:
 
-      EIGEN_STRONG_INLINE InnerIterator(const BlockType& block, Index outer)
-        : Base(block.derived().nestedExpression(), outer + (IsRowMajor ? block.m_startRow.value() : block.m_startCol.value())),
-          m_block(block),
-          m_end(IsRowMajor ? block.m_startCol.value()+block.m_blockCols.value() : block.m_startRow.value()+block.m_blockRows.value())
-      {
-        while( (Base::operator bool()) && (Base::index() < (IsRowMajor ? m_block.m_startCol.value() : m_block.m_startRow.value())) )
-          Base::operator++();
-      }
+    inline const XprType& nestedExpression() const { return m_matrix; }
+    inline XprType& nestedExpression() { return m_matrix; }
+    Index startRow() const { return m_startRow.value(); }
+    Index startCol() const { return m_startCol.value(); }
+    Index blockRows() const { return m_blockRows.value(); }
+    Index blockCols() const { return m_blockCols.value(); }
 
-      inline Index index()  const { return Base::index() - (IsRowMajor ? m_block.m_startCol.value() : m_block.m_startRow.value()); }
-      inline Index outer()  const { return Base::outer() - (IsRowMajor ? m_block.m_startRow.value() : m_block.m_startCol.value()); }
-      inline Index row()    const { return Base::row()   - m_block.m_startRow.value(); }
-      inline Index col()    const { return Base::col()   - m_block.m_startCol.value(); }
-      
-      inline operator bool() const { return Base::operator bool() && Base::index() < m_end; }
-    };
-    class ReverseInnerIterator : public _MatrixTypeNested::ReverseInnerIterator
-    {
-      typedef typename _MatrixTypeNested::ReverseInnerIterator Base;
-      const BlockType& m_block;
-      Index m_begin;
-    public:
-
-      EIGEN_STRONG_INLINE ReverseInnerIterator(const BlockType& block, Index outer)
-        : Base(block.derived().nestedExpression(), outer + (IsRowMajor ? block.m_startRow.value() : block.m_startCol.value())),
-          m_block(block),
-          m_begin(IsRowMajor ? block.m_startCol.value() : block.m_startRow.value())
-      {
-        while( (Base::operator bool()) && (Base::index() >= (IsRowMajor ? m_block.m_startCol.value()+block.m_blockCols.value() : m_block.m_startRow.value()+block.m_blockRows.value())) )
-          Base::operator--();
-      }
-
-      inline Index index()  const { return Base::index() - (IsRowMajor ? m_block.m_startCol.value() : m_block.m_startRow.value()); }
-      inline Index outer()  const { return Base::outer() - (IsRowMajor ? m_block.m_startRow.value() : m_block.m_startCol.value()); }
-      inline Index row()    const { return Base::row()   - m_block.m_startRow.value(); }
-      inline Index col()    const { return Base::col()   - m_block.m_startCol.value(); }
-      
-      inline operator bool() const { return Base::operator bool() && Base::index() >= m_begin; }
-    };
   protected:
-    friend class InnerIterator;
-    friend class ReverseInnerIterator;
-    
-    EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl)
+//     friend class internal::GenericSparseBlockInnerIteratorImpl<XprType,BlockRows,BlockCols,InnerPanel>;
+    friend struct internal::unary_evaluator<Block<XprType,BlockRows,BlockCols,InnerPanel>, internal::IteratorBased, Scalar >;
 
-    typename XprType::Nested m_matrix;
+    Index nonZeros() const { return Dynamic; }
+
+    typename internal::ref_selector<XprType>::non_const_type m_matrix;
     const internal::variable_if_dynamic<Index, XprType::RowsAtCompileTime == 1 ? 0 : Dynamic> m_startRow;
     const internal::variable_if_dynamic<Index, XprType::ColsAtCompileTime == 1 ? 0 : Dynamic> m_startCol;
     const internal::variable_if_dynamic<Index, RowsAtCompileTime> m_blockRows;
     const internal::variable_if_dynamic<Index, ColsAtCompileTime> m_blockCols;
 
+  protected:
+    // Disable assignment with clear error message.
+    // Note that simply removing operator= yields compilation errors with ICC+MSVC
+    template<typename T>
+    BlockImpl& operator=(const T&)
+    {
+      EIGEN_STATIC_ASSERT(sizeof(T)==0, THIS_SPARSE_BLOCK_SUBEXPRESSION_IS_READ_ONLY);
+      return *this;
+    }
+
 };
 
+namespace internal {
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+struct unary_evaluator<Block<ArgType,BlockRows,BlockCols,InnerPanel>, IteratorBased >
+ : public evaluator_base<Block<ArgType,BlockRows,BlockCols,InnerPanel> >
+{
+    class InnerVectorInnerIterator;
+    class OuterVectorInnerIterator;
+  public:
+    typedef Block<ArgType,BlockRows,BlockCols,InnerPanel> XprType;
+    typedef typename XprType::StorageIndex StorageIndex;
+    typedef typename XprType::Scalar Scalar;
+
+    enum {
+      IsRowMajor = XprType::IsRowMajor,
+
+      OuterVector =  (BlockCols==1 && ArgType::IsRowMajor)
+                    | // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&".
+                      // revert to || as soon as not needed anymore.
+                     (BlockRows==1 && !ArgType::IsRowMajor),
+
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+
+    typedef typename internal::conditional<OuterVector,OuterVectorInnerIterator,InnerVectorInnerIterator>::type InnerIterator;
+
+    explicit unary_evaluator(const XprType& op)
+      : m_argImpl(op.nestedExpression()), m_block(op)
+    {}
+
+    inline Index nonZerosEstimate() const {
+      Index nnz = m_block.nonZeros();
+      if(nnz<0)
+        return m_argImpl.nonZerosEstimate() * m_block.size() / m_block.nestedExpression().size();
+      return nnz;
+    }
+
+  protected:
+    typedef typename evaluator<ArgType>::InnerIterator EvalIterator;
+
+    evaluator<ArgType> m_argImpl;
+    const XprType &m_block;
+};
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+class unary_evaluator<Block<ArgType,BlockRows,BlockCols,InnerPanel>, IteratorBased>::InnerVectorInnerIterator
+ : public EvalIterator
+{
+  enum { IsRowMajor = unary_evaluator::IsRowMajor };
+  const XprType& m_block;
+  Index m_end;
+public:
+
+  EIGEN_STRONG_INLINE InnerVectorInnerIterator(const unary_evaluator& aEval, Index outer)
+    : EvalIterator(aEval.m_argImpl, outer + (IsRowMajor ? aEval.m_block.startRow() : aEval.m_block.startCol())),
+      m_block(aEval.m_block),
+      m_end(IsRowMajor ? aEval.m_block.startCol()+aEval.m_block.blockCols() : aEval.m_block.startRow()+aEval.m_block.blockRows())
+  {
+    while( (EvalIterator::operator bool()) && (EvalIterator::index() < (IsRowMajor ? m_block.startCol() : m_block.startRow())) )
+      EvalIterator::operator++();
+  }
+
+  inline StorageIndex index() const { return EvalIterator::index() - convert_index<StorageIndex>(IsRowMajor ? m_block.startCol() : m_block.startRow()); }
+  inline Index outer()  const { return EvalIterator::outer() - (IsRowMajor ? m_block.startRow() : m_block.startCol()); }
+  inline Index row()    const { return EvalIterator::row()   - m_block.startRow(); }
+  inline Index col()    const { return EvalIterator::col()   - m_block.startCol(); }
+
+  inline operator bool() const { return EvalIterator::operator bool() && EvalIterator::index() < m_end; }
+};
+
+template<typename ArgType, int BlockRows, int BlockCols, bool InnerPanel>
+class unary_evaluator<Block<ArgType,BlockRows,BlockCols,InnerPanel>, IteratorBased>::OuterVectorInnerIterator
+{
+  enum { IsRowMajor = unary_evaluator::IsRowMajor };
+  const unary_evaluator& m_eval;
+  Index m_outerPos;
+  const Index m_innerIndex;
+  Index m_end;
+  EvalIterator m_it;
+public:
+
+  EIGEN_STRONG_INLINE OuterVectorInnerIterator(const unary_evaluator& aEval, Index outer)
+    : m_eval(aEval),
+      m_outerPos( (IsRowMajor ? aEval.m_block.startCol() : aEval.m_block.startRow()) ),
+      m_innerIndex(IsRowMajor ? aEval.m_block.startRow() : aEval.m_block.startCol()),
+      m_end(IsRowMajor ? aEval.m_block.startCol()+aEval.m_block.blockCols() : aEval.m_block.startRow()+aEval.m_block.blockRows()),
+      m_it(m_eval.m_argImpl, m_outerPos)
+  {
+    EIGEN_UNUSED_VARIABLE(outer);
+    eigen_assert(outer==0);
+
+    while(m_it && m_it.index() < m_innerIndex) ++m_it;
+    if((!m_it) || (m_it.index()!=m_innerIndex))
+      ++(*this);
+  }
+
+  inline StorageIndex index() const { return convert_index<StorageIndex>(m_outerPos - (IsRowMajor ? m_eval.m_block.startCol() : m_eval.m_block.startRow())); }
+  inline Index outer()  const { return 0; }
+  inline Index row()    const { return IsRowMajor ? 0 : index(); }
+  inline Index col()    const { return IsRowMajor ? index() : 0; }
+
+  inline Scalar value() const { return m_it.value(); }
+  inline Scalar& valueRef() { return m_it.valueRef(); }
+
+  inline OuterVectorInnerIterator& operator++()
+  {
+    // search next non-zero entry
+    while(++m_outerPos<m_end)
+    {
+      // Restart iterator at the next inner-vector:
+      m_it.~EvalIterator();
+      ::new (&m_it) EvalIterator(m_eval.m_argImpl, m_outerPos);
+      // search for the key m_innerIndex in the current outer-vector
+      while(m_it && m_it.index() < m_innerIndex) ++m_it;
+      if(m_it && m_it.index()==m_innerIndex) break;
+    }
+    return *this;
+  }
+
+  inline operator bool() const { return m_outerPos < m_end; }
+};
+
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+struct unary_evaluator<Block<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true>, IteratorBased>
+  : evaluator<SparseCompressedBase<Block<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> > >
+{
+  typedef Block<SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> XprType;
+  typedef evaluator<SparseCompressedBase<XprType> > Base;
+  explicit unary_evaluator(const XprType &xpr) : Base(xpr) {}
+};
+
+template<typename _Scalar, int _Options, typename _StorageIndex, int BlockRows, int BlockCols>
+struct unary_evaluator<Block<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true>, IteratorBased>
+  : evaluator<SparseCompressedBase<Block<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> > >
+{
+  typedef Block<const SparseMatrix<_Scalar, _Options, _StorageIndex>,BlockRows,BlockCols,true> XprType;
+  typedef evaluator<SparseCompressedBase<XprType> > Base;
+  explicit unary_evaluator(const XprType &xpr) : Base(xpr) {}
+};
+
+} // end namespace internal
+
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSE_BLOCK_H
-
diff --git a/Eigen/src/SparseCore/SparseColEtree.h b/Eigen/src/SparseCore/SparseColEtree.h
index f8745f4..ebe02d1 100644
--- a/Eigen/src/SparseCore/SparseColEtree.h
+++ b/Eigen/src/SparseCore/SparseColEtree.h
@@ -58,30 +58,29 @@
   * \param perm The permutation to apply to the column of \b mat
   */
 template <typename MatrixType, typename IndexVector>
-int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowElt, typename MatrixType::Index *perm=0)
+int coletree(const MatrixType& mat, IndexVector& parent, IndexVector& firstRowElt, typename MatrixType::StorageIndex *perm=0)
 {
-  typedef typename MatrixType::Index Index;
-  Index nc = mat.cols(); // Number of columns 
-  Index m = mat.rows();
-  Index diagSize = (std::min)(nc,m);
+  typedef typename MatrixType::StorageIndex StorageIndex;
+  StorageIndex nc = convert_index<StorageIndex>(mat.cols()); // Number of columns
+  StorageIndex m = convert_index<StorageIndex>(mat.rows());
+  StorageIndex diagSize = (std::min)(nc,m);
   IndexVector root(nc); // root of subtree of etree 
   root.setZero();
   IndexVector pp(nc); // disjoint sets 
   pp.setZero(); // Initialize disjoint sets 
   parent.resize(mat.cols());
   //Compute first nonzero column in each row 
-  Index row,col; 
   firstRowElt.resize(m);
   firstRowElt.setConstant(nc);
   firstRowElt.segment(0, diagSize).setLinSpaced(diagSize, 0, diagSize-1);
   bool found_diag;
-  for (col = 0; col < nc; col++)
+  for (StorageIndex col = 0; col < nc; col++)
   {
-    Index pcol = col;
+    StorageIndex pcol = col;
     if(perm) pcol  = perm[col];
     for (typename MatrixType::InnerIterator it(mat, pcol); it; ++it)
     { 
-      row = it.row();
+      Index row = it.row();
       firstRowElt(row) = (std::min)(firstRowElt(row), col);
     }
   }
@@ -89,8 +88,8 @@
           except use (firstRowElt[r],c) in place of an edge (r,c) of A.
     Thus each row clique in A'*A is replaced by a star
     centered at its first vertex, which has the same fill. */
-  Index rset, cset, rroot; 
-  for (col = 0; col < nc; col++) 
+  StorageIndex rset, cset, rroot;
+  for (StorageIndex col = 0; col < nc; col++) 
   {
     found_diag = col>=m;
     pp(col) = col; 
@@ -99,7 +98,7 @@
     parent(col) = nc; 
     /* The diagonal element is treated here even if it does not exist in the matrix
      * hence the loop is executed once more */ 
-    Index pcol = col;
+    StorageIndex pcol = col;
     if(perm) pcol  = perm[col];
     for (typename MatrixType::InnerIterator it(mat, pcol); it||!found_diag; ++it)
     { //  A sequence of interleaved find and union is performed 
@@ -107,7 +106,7 @@
       if(it) i = it.index();
       if (i == col) found_diag = true;
       
-      row = firstRowElt(i);
+      StorageIndex row = firstRowElt(i);
       if (row >= col) continue; 
       rset = internal::etree_find(row, pp); // Find the name of the set containing row
       rroot = root(rset);
@@ -127,10 +126,11 @@
   * Depth-first search from vertex n.  No recursion.
   * This routine was contributed by Cédric Doucet, CEDRAT Group, Meylan, France.
 */
-template <typename Index, typename IndexVector>
-void nr_etdfs (Index n, IndexVector& parent, IndexVector& first_kid, IndexVector& next_kid, IndexVector& post, Index postnum)
+template <typename IndexVector>
+void nr_etdfs (typename IndexVector::Scalar n, IndexVector& parent, IndexVector& first_kid, IndexVector& next_kid, IndexVector& post, typename IndexVector::Scalar postnum)
 {
-  Index current = n, first, next;
+  typedef typename IndexVector::Scalar StorageIndex;
+  StorageIndex current = n, first, next;
   while (postnum != n) 
   {
     // No kid for the current node
@@ -174,22 +174,22 @@
   * \param parent Input tree
   * \param post postordered tree
   */
-template <typename Index, typename IndexVector>
-void treePostorder(Index n, IndexVector& parent, IndexVector& post)
+template <typename IndexVector>
+void treePostorder(typename IndexVector::Scalar n, IndexVector& parent, IndexVector& post)
 {
+  typedef typename IndexVector::Scalar StorageIndex;
   IndexVector first_kid, next_kid; // Linked list of children 
-  Index postnum; 
+  StorageIndex postnum; 
   // Allocate storage for working arrays and results 
   first_kid.resize(n+1); 
   next_kid.setZero(n+1);
   post.setZero(n+1);
   
   // Set up structure describing children
-  Index v, dad; 
   first_kid.setConstant(-1); 
-  for (v = n-1; v >= 0; v--) 
+  for (StorageIndex v = n-1; v >= 0; v--) 
   {
-    dad = parent(v);
+    StorageIndex dad = parent(v);
     next_kid(v) = first_kid(dad); 
     first_kid(dad) = v; 
   }
diff --git a/Eigen/src/SparseCore/SparseCompressedBase.h b/Eigen/src/SparseCore/SparseCompressedBase.h
new file mode 100644
index 0000000..5ccb466
--- /dev/null
+++ b/Eigen/src/SparseCore/SparseCompressedBase.h
@@ -0,0 +1,341 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSE_COMPRESSED_BASE_H
+#define EIGEN_SPARSE_COMPRESSED_BASE_H
+
+namespace Eigen { 
+
+template<typename Derived> class SparseCompressedBase;
+  
+namespace internal {
+
+template<typename Derived>
+struct traits<SparseCompressedBase<Derived> > : traits<Derived>
+{};
+
+} // end namespace internal
+
+/** \ingroup SparseCore_Module
+  * \class SparseCompressedBase
+  * \brief Common base class for sparse [compressed]-{row|column}-storage format.
+  *
+  * This class defines the common interface for all derived classes implementing the compressed sparse storage format, such as:
+  *  - SparseMatrix
+  *  - Ref<SparseMatrixType,Options>
+  *  - Map<SparseMatrixType>
+  *
+  */
+template<typename Derived>
+class SparseCompressedBase
+  : public SparseMatrixBase<Derived>
+{
+  public:
+    typedef SparseMatrixBase<Derived> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(SparseCompressedBase)
+    using Base::operator=;
+    using Base::IsRowMajor;
+    
+    class InnerIterator;
+    class ReverseInnerIterator;
+    
+  protected:
+    typedef typename Base::IndexVector IndexVector;
+    Eigen::Map<IndexVector> innerNonZeros() { return Eigen::Map<IndexVector>(innerNonZeroPtr(), isCompressed()?0:derived().outerSize()); }
+    const  Eigen::Map<const IndexVector> innerNonZeros() const { return Eigen::Map<const IndexVector>(innerNonZeroPtr(), isCompressed()?0:derived().outerSize()); }
+        
+  public:
+    
+    /** \returns the number of non zero coefficients */
+    inline Index nonZeros() const
+    {
+      if(Derived::IsVectorAtCompileTime && outerIndexPtr()==0)
+        return derived().nonZeros();
+      else if(isCompressed())
+        return outerIndexPtr()[derived().outerSize()]-outerIndexPtr()[0];
+      else if(derived().outerSize()==0)
+        return 0;
+      else
+        return innerNonZeros().sum();
+    }
+    
+    /** \returns a const pointer to the array of values.
+      * This function is aimed at interoperability with other libraries.
+      * \sa innerIndexPtr(), outerIndexPtr() */
+    inline const Scalar* valuePtr() const { return derived().valuePtr(); }
+    /** \returns a non-const pointer to the array of values.
+      * This function is aimed at interoperability with other libraries.
+      * \sa innerIndexPtr(), outerIndexPtr() */
+    inline Scalar* valuePtr() { return derived().valuePtr(); }
+
+    /** \returns a const pointer to the array of inner indices.
+      * This function is aimed at interoperability with other libraries.
+      * \sa valuePtr(), outerIndexPtr() */
+    inline const StorageIndex* innerIndexPtr() const { return derived().innerIndexPtr(); }
+    /** \returns a non-const pointer to the array of inner indices.
+      * This function is aimed at interoperability with other libraries.
+      * \sa valuePtr(), outerIndexPtr() */
+    inline StorageIndex* innerIndexPtr() { return derived().innerIndexPtr(); }
+
+    /** \returns a const pointer to the array of the starting positions of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 for SparseVector
+      * \sa valuePtr(), innerIndexPtr() */
+    inline const StorageIndex* outerIndexPtr() const { return derived().outerIndexPtr(); }
+    /** \returns a non-const pointer to the array of the starting positions of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 for SparseVector
+      * \sa valuePtr(), innerIndexPtr() */
+    inline StorageIndex* outerIndexPtr() { return derived().outerIndexPtr(); }
+
+    /** \returns a const pointer to the array of the number of non zeros of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 in compressed mode */
+    inline const StorageIndex* innerNonZeroPtr() const { return derived().innerNonZeroPtr(); }
+    /** \returns a non-const pointer to the array of the number of non zeros of the inner vectors.
+      * This function is aimed at interoperability with other libraries.
+      * \warning it returns the null pointer 0 in compressed mode */
+    inline StorageIndex* innerNonZeroPtr() { return derived().innerNonZeroPtr(); }
+    
+    /** \returns whether \c *this is in compressed form. */
+    inline bool isCompressed() const { return innerNonZeroPtr()==0; }
+
+    /** \returns a read-only view of the stored coefficients as a 1D array expression.
+      *
+      * \warning this method is for \b compressed \b storage \b only, and it will trigger an assertion otherwise.
+      *
+      * \sa valuePtr(), isCompressed() */
+    const Map<const Array<Scalar,Dynamic,1> > coeffs() const { eigen_assert(isCompressed()); return Array<Scalar,Dynamic,1>::Map(valuePtr(),nonZeros()); }
+
+    /** \returns a read-write view of the stored coefficients as a 1D array expression
+      *
+      * \warning this method is for \b compressed \b storage \b only, and it will trigger an assertion otherwise.
+      *
+      * Here is an example:
+      * \include SparseMatrix_coeffs.cpp
+      * and the output is:
+      * \include SparseMatrix_coeffs.out
+      *
+      * \sa valuePtr(), isCompressed() */
+    Map<Array<Scalar,Dynamic,1> > coeffs() { eigen_assert(isCompressed()); return Array<Scalar,Dynamic,1>::Map(valuePtr(),nonZeros()); }
+
+  protected:
+    /** Default constructor. Do nothing. */
+    SparseCompressedBase() {}
+  private:
+    template<typename OtherDerived> explicit SparseCompressedBase(const SparseCompressedBase<OtherDerived>&);
+};
+
+template<typename Derived>
+class SparseCompressedBase<Derived>::InnerIterator
+{
+  public:
+    InnerIterator()
+      : m_values(0), m_indices(0), m_outer(0), m_id(0), m_end(0)
+    {}
+
+    InnerIterator(const InnerIterator& other)
+      : m_values(other.m_values), m_indices(other.m_indices), m_outer(other.m_outer), m_id(other.m_id), m_end(other.m_end)
+    {}
+
+    InnerIterator& operator=(const InnerIterator& other)
+    {
+      m_values = other.m_values;
+      m_indices = other.m_indices;
+      const_cast<OuterType&>(m_outer).setValue(other.m_outer.value());
+      m_id = other.m_id;
+      m_end = other.m_end;
+      return *this;
+    }
+
+    InnerIterator(const SparseCompressedBase& mat, Index outer)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer)
+    {
+      if(Derived::IsVectorAtCompileTime && mat.outerIndexPtr()==0)
+      {
+        m_id = 0;
+        m_end = mat.nonZeros();
+      }
+      else
+      {
+        m_id = mat.outerIndexPtr()[outer];
+        if(mat.isCompressed())
+          m_end = mat.outerIndexPtr()[outer+1];
+        else
+          m_end = m_id + mat.innerNonZeroPtr()[outer];
+      }
+    }
+
+    explicit InnerIterator(const SparseCompressedBase& mat)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(0), m_id(0), m_end(mat.nonZeros())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    explicit InnerIterator(const internal::CompressedStorage<Scalar,StorageIndex>& data)
+      : m_values(data.valuePtr()), m_indices(data.indexPtr()), m_outer(0), m_id(0), m_end(data.size())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    inline InnerIterator& operator++() { m_id++; return *this; }
+
+    inline const Scalar& value() const { return m_values[m_id]; }
+    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id]); }
+
+    inline StorageIndex index() const { return m_indices[m_id]; }
+    inline Index outer() const { return m_outer.value(); }
+    inline Index row() const { return IsRowMajor ? m_outer.value() : index(); }
+    inline Index col() const { return IsRowMajor ? index() : m_outer.value(); }
+
+    inline operator bool() const { return (m_id < m_end); }
+
+  protected:
+    const Scalar* m_values;
+    const StorageIndex* m_indices;
+    typedef internal::variable_if_dynamic<Index,Derived::IsVectorAtCompileTime?0:Dynamic> OuterType;
+    const OuterType m_outer;
+    Index m_id;
+    Index m_end;
+  private:
+    // If you get here, then you're not using the right InnerIterator type, e.g.:
+    //   SparseMatrix<double,RowMajor> A;
+    //   SparseMatrix<double>::InnerIterator it(A,0);
+    template<typename T> InnerIterator(const SparseMatrixBase<T>&, Index outer);
+};
+
+template<typename Derived>
+class SparseCompressedBase<Derived>::ReverseInnerIterator
+{
+  public:
+    ReverseInnerIterator(const SparseCompressedBase& mat, Index outer)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer)
+    {
+      if(Derived::IsVectorAtCompileTime && mat.outerIndexPtr()==0)
+      {
+        m_start = 0;
+        m_id = mat.nonZeros();
+      }
+      else
+      {
+        m_start = mat.outerIndexPtr()[outer];
+        if(mat.isCompressed())
+          m_id = mat.outerIndexPtr()[outer+1];
+        else
+          m_id = m_start + mat.innerNonZeroPtr()[outer];
+      }
+    }
+
+    explicit ReverseInnerIterator(const SparseCompressedBase& mat)
+      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(0), m_start(0), m_id(mat.nonZeros())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    explicit ReverseInnerIterator(const internal::CompressedStorage<Scalar,StorageIndex>& data)
+      : m_values(data.valuePtr()), m_indices(data.indexPtr()), m_outer(0), m_start(0), m_id(data.size())
+    {
+      EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived);
+    }
+
+    inline ReverseInnerIterator& operator--() { --m_id; return *this; }
+
+    inline const Scalar& value() const { return m_values[m_id-1]; }
+    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id-1]); }
+
+    inline StorageIndex index() const { return m_indices[m_id-1]; }
+    inline Index outer() const { return m_outer.value(); }
+    inline Index row() const { return IsRowMajor ? m_outer.value() : index(); }
+    inline Index col() const { return IsRowMajor ? index() : m_outer.value(); }
+
+    inline operator bool() const { return (m_id > m_start); }
+
+  protected:
+    const Scalar* m_values;
+    const StorageIndex* m_indices;
+    typedef internal::variable_if_dynamic<Index,Derived::IsVectorAtCompileTime?0:Dynamic> OuterType;
+    const OuterType m_outer;
+    Index m_start;
+    Index m_id;
+};
+
+namespace internal {
+
+template<typename Derived>
+struct evaluator<SparseCompressedBase<Derived> >
+  : evaluator_base<Derived>
+{
+  typedef typename Derived::Scalar Scalar;
+  typedef typename Derived::InnerIterator InnerIterator;
+  
+  enum {
+    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = Derived::Flags
+  };
+  
+  evaluator() : m_matrix(0), m_zero(0)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  explicit evaluator(const Derived &mat) : m_matrix(&mat), m_zero(0)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_matrix->nonZeros();
+  }
+  
+  operator Derived&() { return m_matrix->const_cast_derived(); }
+  operator const Derived&() const { return *m_matrix; }
+  
+  typedef typename DenseCoeffsBase<Derived,ReadOnlyAccessors>::CoeffReturnType CoeffReturnType;
+  const Scalar& coeff(Index row, Index col) const
+  {
+    Index p = find(row,col);
+
+    if(p==Dynamic)
+      return m_zero;
+    else
+      return m_matrix->const_cast_derived().valuePtr()[p];
+  }
+
+  Scalar& coeffRef(Index row, Index col)
+  {
+    Index p = find(row,col);
+    eigen_assert(p!=Dynamic && "written coefficient does not exist");
+    return m_matrix->const_cast_derived().valuePtr()[p];
+  }
+
+protected:
+
+  Index find(Index row, Index col) const
+  {
+    eigen_internal_assert(row>=0 && row<m_matrix->rows() && col>=0 && col<m_matrix->cols());
+
+    const Index outer = Derived::IsRowMajor ? row : col;
+    const Index inner = Derived::IsRowMajor ? col : row;
+
+    Index start = m_matrix->outerIndexPtr()[outer];
+    Index end = m_matrix->isCompressed() ? m_matrix->outerIndexPtr()[outer+1] : m_matrix->outerIndexPtr()[outer] + m_matrix->innerNonZeroPtr()[outer];
+    eigen_assert(end>=start && "you are using a non finalized sparse matrix or written coefficient does not exist");
+    const Index p = std::lower_bound(m_matrix->innerIndexPtr()+start, m_matrix->innerIndexPtr()+end,inner) - m_matrix->innerIndexPtr();
+
+    return ((p<end) && (m_matrix->innerIndexPtr()[p]==inner)) ? p : Dynamic;
+  }
+
+  const Derived *m_matrix;
+  const Scalar m_zero;
+};
+
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSE_COMPRESSED_BASE_H
diff --git a/Eigen/src/SparseCore/SparseCwiseBinaryOp.h b/Eigen/src/SparseCore/SparseCwiseBinaryOp.h
index 60ca769..e315e35 100644
--- a/Eigen/src/SparseCore/SparseCwiseBinaryOp.h
+++ b/Eigen/src/SparseCore/SparseCwiseBinaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -28,89 +28,57 @@
 //                         generic      sparse
 //  4 - dense op dense     product      dense
 //                         generic      dense
-
-namespace internal {
-
-template<> struct promote_storage_type<Dense,Sparse>
-{ typedef Sparse ret; };
-
-template<> struct promote_storage_type<Sparse,Dense>
-{ typedef Sparse ret; };
-
-template<typename BinaryOp, typename Lhs, typename Rhs, typename Derived,
-  typename _LhsStorageMode = typename traits<Lhs>::StorageKind,
-  typename _RhsStorageMode = typename traits<Rhs>::StorageKind>
-class sparse_cwise_binary_op_inner_iterator_selector;
-
-} // end namespace internal
+//
+// TODO to ease compiler job, we could specialize product/quotient with a scalar
+//      and fallback to cwise-unary evaluator using bind1st_op and bind2nd_op.
 
 template<typename BinaryOp, typename Lhs, typename Rhs>
 class CwiseBinaryOpImpl<BinaryOp, Lhs, Rhs, Sparse>
   : public SparseMatrixBase<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
 {
   public:
-    class InnerIterator;
-    class ReverseInnerIterator;
     typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> Derived;
+    typedef SparseMatrixBase<Derived> Base;
     EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
     CwiseBinaryOpImpl()
     {
-      typedef typename internal::traits<Lhs>::StorageKind LhsStorageKind;
-      typedef typename internal::traits<Rhs>::StorageKind RhsStorageKind;
       EIGEN_STATIC_ASSERT((
-                (!internal::is_same<LhsStorageKind,RhsStorageKind>::value)
-            ||  ((Lhs::Flags&RowMajorBit) == (Rhs::Flags&RowMajorBit))),
+                (!internal::is_same<typename internal::traits<Lhs>::StorageKind,
+                                    typename internal::traits<Rhs>::StorageKind>::value)
+            ||  ((internal::evaluator<Lhs>::Flags&RowMajorBit) == (internal::evaluator<Rhs>::Flags&RowMajorBit))),
             THE_STORAGE_ORDER_OF_BOTH_SIDES_MUST_MATCH);
     }
 };
 
-template<typename BinaryOp, typename Lhs, typename Rhs>
-class CwiseBinaryOpImpl<BinaryOp,Lhs,Rhs,Sparse>::InnerIterator
-  : public internal::sparse_cwise_binary_op_inner_iterator_selector<BinaryOp,Lhs,Rhs,typename CwiseBinaryOpImpl<BinaryOp,Lhs,Rhs,Sparse>::InnerIterator>
-{
-  public:
-    typedef typename Lhs::Index Index;
-    typedef internal::sparse_cwise_binary_op_inner_iterator_selector<
-      BinaryOp,Lhs,Rhs, InnerIterator> Base;
-
-    // NOTE: we have to prefix Index by "typename Lhs::" to avoid an ICE with VC11
-    EIGEN_STRONG_INLINE InnerIterator(const CwiseBinaryOpImpl& binOp, typename Lhs::Index outer)
-      : Base(binOp.derived(),outer)
-    {}
-};
-
-/***************************************************************************
-* Implementation of inner-iterators
-***************************************************************************/
-
-// template<typename T> struct internal::func_is_conjunction { enum { ret = false }; };
-// template<typename T> struct internal::func_is_conjunction<internal::scalar_product_op<T> > { enum { ret = true }; };
-
-// TODO generalize the internal::scalar_product_op specialization to all conjunctions if any !
-
 namespace internal {
 
-// sparse - sparse  (generic)
-template<typename BinaryOp, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<BinaryOp, Lhs, Rhs, Derived, Sparse, Sparse>
+  
+// Generic "sparse OP sparse"
+template<typename XprType> struct binary_sparse_evaluator;
+
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IteratorBased, IteratorBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
 {
-    typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename traits<CwiseBinaryXpr>::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_LhsNested _LhsNested;
-    typedef typename traits<CwiseBinaryXpr>::_RhsNested _RhsNested;
-    typedef typename _LhsNested::InnerIterator LhsIterator;
-    typedef typename _RhsNested::InnerIterator RhsIterator;
-    typedef typename Lhs::Index Index;
+protected:
+  typedef typename evaluator<Lhs>::InnerIterator  LhsIterator;
+  typedef typename evaluator<Rhs>::InnerIterator  RhsIterator;
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+public:
 
+  class InnerIterator
+  {
   public:
-
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_lhsIter(xpr.lhs(),outer), m_rhsIter(xpr.rhs(),outer), m_functor(xpr.functor())
+    
+    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor)
     {
       this->operator++();
     }
 
-    EIGEN_STRONG_INLINE Derived& operator++()
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
     {
       if (m_lhsIter && m_rhsIter && (m_lhsIter.index() == m_rhsIter.index()))
       {
@@ -136,12 +104,13 @@
         m_value = 0; // this is to avoid a compilation warning
         m_id = -1;
       }
-      return *static_cast<Derived*>(this);
+      return *this;
     }
 
     EIGEN_STRONG_INLINE Scalar value() const { return m_value; }
 
-    EIGEN_STRONG_INLINE Index index() const { return m_id; }
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return Lhs::IsRowMajor ? m_lhsIter.row() : index(); }
     EIGEN_STRONG_INLINE Index col() const { return Lhs::IsRowMajor ? index() : m_lhsIter.col(); }
 
@@ -152,25 +121,303 @@
     RhsIterator m_rhsIter;
     const BinaryOp& m_functor;
     Scalar m_value;
-    Index m_id;
+    StorageIndex m_id;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    Flags = XprType::Flags
+  };
+  
+  explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_lhsImpl.nonZerosEstimate() + m_rhsImpl.nonZerosEstimate();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
 };
 
-// sparse - sparse  (product)
-template<typename T, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs, Rhs, Derived, Sparse, Sparse>
+// dense op sparse
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IndexBased, IteratorBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
 {
-    typedef scalar_product_op<T> BinaryFunc;
-    typedef CwiseBinaryOp<BinaryFunc, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename CwiseBinaryXpr::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_LhsNested _LhsNested;
-    typedef typename _LhsNested::InnerIterator LhsIterator;
-    typedef typename traits<CwiseBinaryXpr>::_RhsNested _RhsNested;
-    typedef typename _RhsNested::InnerIterator RhsIterator;
-    typedef typename Lhs::Index Index;
+protected:
+  typedef typename evaluator<Rhs>::InnerIterator  RhsIterator;
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+public:
+
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(Rhs::Flags)&RowMajorBit)==RowMajorBit };
   public:
 
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_lhsIter(xpr.lhs(),outer), m_rhsIter(xpr.rhs(),outer), m_functor(xpr.functor())
+    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+      : m_lhsEval(aEval.m_lhsImpl), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor), m_value(0), m_id(-1), m_innerSize(aEval.m_expr.rhs().innerSize())
+    {
+      this->operator++();
+    }
+
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
+    {
+      ++m_id;
+      if(m_id<m_innerSize)
+      {
+        Scalar lhsVal = m_lhsEval.coeff(IsRowMajor?m_rhsIter.outer():m_id,
+                                        IsRowMajor?m_id:m_rhsIter.outer());
+        if(m_rhsIter && m_rhsIter.index()==m_id)
+        {
+          m_value = m_functor(lhsVal, m_rhsIter.value());
+          ++m_rhsIter;
+        }
+        else
+          m_value = m_functor(lhsVal, Scalar(0));
+      }
+
+      return *this;
+    }
+
+    EIGEN_STRONG_INLINE Scalar value() const { eigen_internal_assert(m_id<m_innerSize); return m_value; }
+
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_rhsIter.outer(); }
+    EIGEN_STRONG_INLINE Index row() const { return IsRowMajor ? m_rhsIter.outer() : m_id; }
+    EIGEN_STRONG_INLINE Index col() const { return IsRowMajor ? m_id : m_rhsIter.outer(); }
+
+    EIGEN_STRONG_INLINE operator bool() const { return m_id<m_innerSize; }
+
+  protected:
+    const evaluator<Lhs> &m_lhsEval;
+    RhsIterator m_rhsIter;
+    const BinaryOp& m_functor;
+    Scalar m_value;
+    StorageIndex m_id;
+    StorageIndex m_innerSize;
+  };
+
+
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(Rhs::Flags)&RowMajorBit)
+  };
+
+  explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()),
+      m_rhsImpl(xpr.rhs()),
+      m_expr(xpr)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  inline Index nonZerosEstimate() const {
+    return m_expr.size();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
+  const XprType &m_expr;
+};
+
+// sparse op dense
+template<typename BinaryOp, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<BinaryOp, Lhs, Rhs>, IteratorBased, IndexBased>
+  : evaluator_base<CwiseBinaryOp<BinaryOp, Lhs, Rhs> >
+{
+protected:
+  typedef typename evaluator<Lhs>::InnerIterator  LhsIterator;
+  typedef CwiseBinaryOp<BinaryOp, Lhs, Rhs> XprType;
+  typedef typename traits<XprType>::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+public:
+
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(Lhs::Flags)&RowMajorBit)==RowMajorBit };
+  public:
+
+    EIGEN_STRONG_INLINE InnerIterator(const binary_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsEval(aEval.m_rhsImpl), m_functor(aEval.m_functor), m_value(0), m_id(-1), m_innerSize(aEval.m_expr.lhs().innerSize())
+    {
+      this->operator++();
+    }
+
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
+    {
+      ++m_id;
+      if(m_id<m_innerSize)
+      {
+        Scalar rhsVal = m_rhsEval.coeff(IsRowMajor?m_lhsIter.outer():m_id,
+                                        IsRowMajor?m_id:m_lhsIter.outer());
+        if(m_lhsIter && m_lhsIter.index()==m_id)
+        {
+          m_value = m_functor(m_lhsIter.value(), rhsVal);
+          ++m_lhsIter;
+        }
+        else
+          m_value = m_functor(Scalar(0),rhsVal);
+      }
+
+      return *this;
+    }
+
+    EIGEN_STRONG_INLINE Scalar value() const { eigen_internal_assert(m_id<m_innerSize); return m_value; }
+
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_id; }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
+    EIGEN_STRONG_INLINE Index row() const { return IsRowMajor ? m_lhsIter.outer() : m_id; }
+    EIGEN_STRONG_INLINE Index col() const { return IsRowMajor ? m_id : m_lhsIter.outer(); }
+
+    EIGEN_STRONG_INLINE operator bool() const { return m_id<m_innerSize; }
+
+  protected:
+    LhsIterator m_lhsIter;
+    const evaluator<Rhs> &m_rhsEval;
+    const BinaryOp& m_functor;
+    Scalar m_value;
+    StorageIndex m_id;
+    StorageIndex m_innerSize;
+  };
+
+
+  enum {
+    CoeffReadCost = evaluator<Lhs>::CoeffReadCost + evaluator<Rhs>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(Lhs::Flags)&RowMajorBit)
+  };
+
+  explicit binary_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()),
+      m_rhsImpl(xpr.rhs()),
+      m_expr(xpr)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+
+  inline Index nonZerosEstimate() const {
+    return m_expr.size();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<Lhs> m_lhsImpl;
+  evaluator<Rhs> m_rhsImpl;
+  const XprType &m_expr;
+};
+
+template<typename T,
+         typename LhsKind   = typename evaluator_traits<typename T::Lhs>::Kind,
+         typename RhsKind   = typename evaluator_traits<typename T::Rhs>::Kind,
+         typename LhsScalar = typename traits<typename T::Lhs>::Scalar,
+         typename RhsScalar = typename traits<typename T::Rhs>::Scalar> struct sparse_conjunction_evaluator;
+
+// "sparse .* sparse"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IteratorBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "dense .* sparse"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IndexBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "sparse .* dense"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_product_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse ./ dense"
+template<typename T1, typename T2, typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_quotient_op<T1,T2>, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse && sparse"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IteratorBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "dense && sparse"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IndexBased, IteratorBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+// "sparse && dense"
+template<typename Lhs, typename Rhs>
+struct binary_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs>, IteratorBased, IndexBased>
+  : sparse_conjunction_evaluator<CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> >
+{
+  typedef CwiseBinaryOp<scalar_boolean_and_op, Lhs, Rhs> XprType;
+  typedef sparse_conjunction_evaluator<XprType> Base;
+  explicit binary_evaluator(const XprType& xpr) : Base(xpr) {}
+};
+
+// "sparse ^ sparse"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IteratorBased, IteratorBased>
+  : evaluator_base<XprType>
+{
+protected:
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef typename evaluator<LhsArg>::InnerIterator  LhsIterator;
+  typedef typename evaluator<RhsArg>::InnerIterator  RhsIterator;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename traits<XprType>::Scalar Scalar;
+public:
+
+  class InnerIterator
+  {
+  public:
+    
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor)
     {
       while (m_lhsIter && m_rhsIter && (m_lhsIter.index() != m_rhsIter.index()))
       {
@@ -181,7 +428,7 @@
       }
     }
 
-    EIGEN_STRONG_INLINE Derived& operator++()
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
     {
       ++m_lhsIter;
       ++m_rhsIter;
@@ -192,12 +439,13 @@
         else
           ++m_rhsIter;
       }
-      return *static_cast<Derived*>(this);
+      return *this;
     }
-
+    
     EIGEN_STRONG_INLINE Scalar value() const { return m_functor(m_lhsIter.value(), m_rhsIter.value()); }
 
-    EIGEN_STRONG_INLINE Index index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
     EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
 
@@ -206,91 +454,184 @@
   protected:
     LhsIterator m_lhsIter;
     RhsIterator m_rhsIter;
-    const BinaryFunc& m_functor;
+    const BinaryOp& m_functor;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    Flags = XprType::Flags
+  };
+  
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return (std::min)(m_lhsImpl.nonZerosEstimate(), m_rhsImpl.nonZerosEstimate());
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
-// sparse - dense  (product)
-template<typename T, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs, Rhs, Derived, Sparse, Dense>
+// "dense ^ sparse"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IndexBased, IteratorBased>
+  : evaluator_base<XprType>
 {
-    typedef scalar_product_op<T> BinaryFunc;
-    typedef CwiseBinaryOp<BinaryFunc, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename CwiseBinaryXpr::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_LhsNested _LhsNested;
-    typedef typename traits<CwiseBinaryXpr>::RhsNested RhsNested;
-    typedef typename _LhsNested::InnerIterator LhsIterator;
-    typedef typename Lhs::Index Index;
-    enum { IsRowMajor = (int(Lhs::Flags)&RowMajorBit)==RowMajorBit };
-  public:
+protected:
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef evaluator<LhsArg> LhsEvaluator;
+  typedef typename evaluator<RhsArg>::InnerIterator  RhsIterator;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename traits<XprType>::Scalar Scalar;
+public:
 
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_rhs(xpr.rhs()), m_lhsIter(xpr.lhs(),outer), m_functor(xpr.functor()), m_outer(outer)
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(RhsArg::Flags)&RowMajorBit)==RowMajorBit };
+
+  public:
+    
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
+      : m_lhsEval(aEval.m_lhsImpl), m_rhsIter(aEval.m_rhsImpl,outer), m_functor(aEval.m_functor), m_outer(outer)
     {}
 
-    EIGEN_STRONG_INLINE Derived& operator++()
-    {
-      ++m_lhsIter;
-      return *static_cast<Derived*>(this);
-    }
-
-    EIGEN_STRONG_INLINE Scalar value() const
-    { return m_functor(m_lhsIter.value(),
-                       m_rhs.coeff(IsRowMajor?m_outer:m_lhsIter.index(),IsRowMajor?m_lhsIter.index():m_outer)); }
-
-    EIGEN_STRONG_INLINE Index index() const { return m_lhsIter.index(); }
-    EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
-    EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
-
-    EIGEN_STRONG_INLINE operator bool() const { return m_lhsIter; }
-
-  protected:
-    RhsNested m_rhs;
-    LhsIterator m_lhsIter;
-    const BinaryFunc m_functor;
-    const Index m_outer;
-};
-
-// sparse - dense  (product)
-template<typename T, typename Lhs, typename Rhs, typename Derived>
-class sparse_cwise_binary_op_inner_iterator_selector<scalar_product_op<T>, Lhs, Rhs, Derived, Dense, Sparse>
-{
-    typedef scalar_product_op<T> BinaryFunc;
-    typedef CwiseBinaryOp<BinaryFunc, Lhs, Rhs> CwiseBinaryXpr;
-    typedef typename CwiseBinaryXpr::Scalar Scalar;
-    typedef typename traits<CwiseBinaryXpr>::_RhsNested _RhsNested;
-    typedef typename _RhsNested::InnerIterator RhsIterator;
-    typedef typename Lhs::Index Index;
-
-    enum { IsRowMajor = (int(Rhs::Flags)&RowMajorBit)==RowMajorBit };
-  public:
-
-    EIGEN_STRONG_INLINE sparse_cwise_binary_op_inner_iterator_selector(const CwiseBinaryXpr& xpr, Index outer)
-      : m_xpr(xpr), m_rhsIter(xpr.rhs(),outer), m_functor(xpr.functor()), m_outer(outer)
-    {}
-
-    EIGEN_STRONG_INLINE Derived& operator++()
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
     {
       ++m_rhsIter;
-      return *static_cast<Derived*>(this);
+      return *this;
     }
 
     EIGEN_STRONG_INLINE Scalar value() const
-    { return m_functor(m_xpr.lhs().coeff(IsRowMajor?m_outer:m_rhsIter.index(),IsRowMajor?m_rhsIter.index():m_outer), m_rhsIter.value()); }
+    { return m_functor(m_lhsEval.coeff(IsRowMajor?m_outer:m_rhsIter.index(),IsRowMajor?m_rhsIter.index():m_outer), m_rhsIter.value()); }
 
-    EIGEN_STRONG_INLINE Index index() const { return m_rhsIter.index(); }
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_rhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_rhsIter.outer(); }
     EIGEN_STRONG_INLINE Index row() const { return m_rhsIter.row(); }
     EIGEN_STRONG_INLINE Index col() const { return m_rhsIter.col(); }
 
     EIGEN_STRONG_INLINE operator bool() const { return m_rhsIter; }
-
+    
   protected:
-    const CwiseBinaryXpr& m_xpr;
+    const LhsEvaluator &m_lhsEval;
     RhsIterator m_rhsIter;
-    const BinaryFunc& m_functor;
+    const BinaryOp& m_functor;
     const Index m_outer;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(RhsArg::Flags)&RowMajorBit)
+  };
+  
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_rhsImpl.nonZerosEstimate();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
 };
 
-} // end namespace internal
+// "sparse ^ dense"
+template<typename XprType>
+struct sparse_conjunction_evaluator<XprType, IteratorBased, IndexBased>
+  : evaluator_base<XprType>
+{
+protected:
+  typedef typename XprType::Functor BinaryOp;
+  typedef typename XprType::Lhs LhsArg;
+  typedef typename XprType::Rhs RhsArg;
+  typedef typename evaluator<LhsArg>::InnerIterator LhsIterator;
+  typedef evaluator<RhsArg> RhsEvaluator;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename traits<XprType>::Scalar Scalar;
+public:
+
+  class InnerIterator
+  {
+    enum { IsRowMajor = (int(LhsArg::Flags)&RowMajorBit)==RowMajorBit };
+
+  public:
+    
+    EIGEN_STRONG_INLINE InnerIterator(const sparse_conjunction_evaluator& aEval, Index outer)
+      : m_lhsIter(aEval.m_lhsImpl,outer), m_rhsEval(aEval.m_rhsImpl), m_functor(aEval.m_functor), m_outer(outer)
+    {}
+
+    EIGEN_STRONG_INLINE InnerIterator& operator++()
+    {
+      ++m_lhsIter;
+      return *this;
+    }
+
+    EIGEN_STRONG_INLINE Scalar value() const
+    { return m_functor(m_lhsIter.value(),
+                       m_rhsEval.coeff(IsRowMajor?m_outer:m_lhsIter.index(),IsRowMajor?m_lhsIter.index():m_outer)); }
+
+    EIGEN_STRONG_INLINE StorageIndex index() const { return m_lhsIter.index(); }
+    EIGEN_STRONG_INLINE Index outer() const { return m_lhsIter.outer(); }
+    EIGEN_STRONG_INLINE Index row() const { return m_lhsIter.row(); }
+    EIGEN_STRONG_INLINE Index col() const { return m_lhsIter.col(); }
+
+    EIGEN_STRONG_INLINE operator bool() const { return m_lhsIter; }
+    
+  protected:
+    LhsIterator m_lhsIter;
+    const evaluator<RhsArg> &m_rhsEval;
+    const BinaryOp& m_functor;
+    const Index m_outer;
+  };
+  
+  
+  enum {
+    CoeffReadCost = evaluator<LhsArg>::CoeffReadCost + evaluator<RhsArg>::CoeffReadCost + functor_traits<BinaryOp>::Cost,
+    // Expose storage order of the sparse expression
+    Flags = (XprType::Flags & ~RowMajorBit) | (int(LhsArg::Flags)&RowMajorBit)
+  };
+  
+  explicit sparse_conjunction_evaluator(const XprType& xpr)
+    : m_functor(xpr.functor()),
+      m_lhsImpl(xpr.lhs()), 
+      m_rhsImpl(xpr.rhs())  
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<BinaryOp>::Cost);
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_lhsImpl.nonZerosEstimate();
+  }
+
+protected:
+  const BinaryOp m_functor;
+  evaluator<LhsArg> m_lhsImpl;
+  evaluator<RhsArg> m_rhsImpl;
+};
+
+}
 
 /***************************************************************************
 * Implementation of SparseMatrixBase and SparseCwise functions/operators
@@ -298,6 +639,22 @@
 
 template<typename Derived>
 template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator+=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator-=(const EigenBase<OtherDerived> &other)
+{
+  call_assignment(derived(), other.derived(), internal::assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
 EIGEN_STRONG_INLINE Derived &
 SparseMatrixBase<Derived>::operator-=(const SparseMatrixBase<OtherDerived> &other)
 {
@@ -314,10 +671,54 @@
 
 template<typename Derived>
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE
+Derived& SparseMatrixBase<Derived>::operator+=(const DiagonalBase<OtherDerived>& other)
+{
+  call_assignment_no_alias(derived(), other.derived(), internal::add_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+
+template<typename Derived>
+template<typename OtherDerived>
+Derived& SparseMatrixBase<Derived>::operator-=(const DiagonalBase<OtherDerived>& other)
+{
+  call_assignment_no_alias(derived(), other.derived(), internal::sub_assign_op<Scalar,typename OtherDerived::Scalar>());
+  return derived();
+}
+    
+template<typename Derived>
+template<typename OtherDerived>
+EIGEN_STRONG_INLINE const typename SparseMatrixBase<Derived>::template CwiseProductDenseReturnType<OtherDerived>::Type
 SparseMatrixBase<Derived>::cwiseProduct(const MatrixBase<OtherDerived> &other) const
 {
-  return EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE(derived(), other.derived());
+  return typename CwiseProductDenseReturnType<OtherDerived>::Type(derived(), other.derived());
+}
+
+template<typename DenseDerived, typename SparseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_sum_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>
+operator+(const MatrixBase<DenseDerived> &a, const SparseMatrixBase<SparseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_sum_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>(a.derived(), b.derived());
+}
+
+template<typename SparseDerived, typename DenseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_sum_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>
+operator+(const SparseMatrixBase<SparseDerived> &a, const MatrixBase<DenseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_sum_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>(a.derived(), b.derived());
+}
+
+template<typename DenseDerived, typename SparseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_difference_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>
+operator-(const MatrixBase<DenseDerived> &a, const SparseMatrixBase<SparseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_difference_op<typename DenseDerived::Scalar,typename SparseDerived::Scalar>, const DenseDerived, const SparseDerived>(a.derived(), b.derived());
+}
+
+template<typename SparseDerived, typename DenseDerived>
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_difference_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>
+operator-(const SparseMatrixBase<SparseDerived> &a, const MatrixBase<DenseDerived> &b)
+{
+  return CwiseBinaryOp<internal::scalar_difference_op<typename SparseDerived::Scalar,typename DenseDerived::Scalar>, const SparseDerived, const DenseDerived>(a.derived(), b.derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/SparseCore/SparseCwiseUnaryOp.h b/Eigen/src/SparseCore/SparseCwiseUnaryOp.h
index 5a50c78..ea79737 100644
--- a/Eigen/src/SparseCore/SparseCwiseUnaryOp.h
+++ b/Eigen/src/SparseCore/SparseCwiseUnaryOp.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,138 +12,121 @@
 
 namespace Eigen { 
 
-template<typename UnaryOp, typename MatrixType>
-class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>
-  : public SparseMatrixBase<CwiseUnaryOp<UnaryOp, MatrixType> >
+namespace internal {
+  
+template<typename UnaryOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>
+  : public evaluator_base<CwiseUnaryOp<UnaryOp,ArgType> >
 {
   public:
+    typedef CwiseUnaryOp<UnaryOp, ArgType> XprType;
 
     class InnerIterator;
-    class ReverseInnerIterator;
-
-    typedef CwiseUnaryOp<UnaryOp, MatrixType> Derived;
-    EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<UnaryOp>::Cost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& op) : m_functor(op.functor()), m_argImpl(op.nestedExpression())
+    {
+      EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<UnaryOp>::Cost);
+      EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+    }
+    
+    inline Index nonZerosEstimate() const {
+      return m_argImpl.nonZerosEstimate();
+    }
 
   protected:
-    typedef typename internal::traits<Derived>::_XprTypeNested _MatrixTypeNested;
-    typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
-    typedef typename _MatrixTypeNested::ReverseInnerIterator MatrixTypeReverseIterator;
+    typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
+    
+    const UnaryOp m_functor;
+    evaluator<ArgType> m_argImpl;
 };
 
-template<typename UnaryOp, typename MatrixType>
-class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::InnerIterator
-    : public CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeIterator
+template<typename UnaryOp, typename ArgType>
+class unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::InnerIterator
+    : public unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::EvalIterator
 {
-    typedef typename CwiseUnaryOpImpl::Scalar Scalar;
-    typedef typename CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeIterator Base;
+    typedef typename XprType::Scalar Scalar;
+    typedef typename unary_evaluator<CwiseUnaryOp<UnaryOp,ArgType>, IteratorBased>::EvalIterator Base;
   public:
 
-    EIGEN_STRONG_INLINE InnerIterator(const CwiseUnaryOpImpl& unaryOp, typename CwiseUnaryOpImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
+    EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& unaryOp, Index outer)
+      : Base(unaryOp.m_argImpl,outer), m_functor(unaryOp.m_functor)
     {}
 
     EIGEN_STRONG_INLINE InnerIterator& operator++()
     { Base::operator++(); return *this; }
 
-    EIGEN_STRONG_INLINE typename CwiseUnaryOpImpl::Scalar value() const { return m_functor(Base::value()); }
+    EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
 
   protected:
     const UnaryOp m_functor;
   private:
-    typename CwiseUnaryOpImpl::Scalar& valueRef();
+    Scalar& valueRef();
 };
 
-template<typename UnaryOp, typename MatrixType>
-class CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::ReverseInnerIterator
-    : public CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeReverseIterator
-{
-    typedef typename CwiseUnaryOpImpl::Scalar Scalar;
-    typedef typename CwiseUnaryOpImpl<UnaryOp,MatrixType,Sparse>::MatrixTypeReverseIterator Base;
-  public:
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const CwiseUnaryOpImpl& unaryOp, typename CwiseUnaryOpImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
-    {}
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-    { Base::operator--(); return *this; }
-
-    EIGEN_STRONG_INLINE typename CwiseUnaryOpImpl::Scalar value() const { return m_functor(Base::value()); }
-
-  protected:
-    const UnaryOp m_functor;
-  private:
-    typename CwiseUnaryOpImpl::Scalar& valueRef();
-};
-
-template<typename ViewOp, typename MatrixType>
-class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>
-  : public SparseMatrixBase<CwiseUnaryView<ViewOp, MatrixType> >
+template<typename ViewOp, typename ArgType>
+struct unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>
+  : public evaluator_base<CwiseUnaryView<ViewOp,ArgType> >
 {
   public:
+    typedef CwiseUnaryView<ViewOp, ArgType> XprType;
 
     class InnerIterator;
-    class ReverseInnerIterator;
-
-    typedef CwiseUnaryView<ViewOp, MatrixType> Derived;
-    EIGEN_SPARSE_PUBLIC_INTERFACE(Derived)
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost + functor_traits<ViewOp>::Cost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& op) : m_functor(op.functor()), m_argImpl(op.nestedExpression())
+    {
+      EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits<ViewOp>::Cost);
+      EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+    }
 
   protected:
-    typedef typename internal::traits<Derived>::_MatrixTypeNested _MatrixTypeNested;
-    typedef typename _MatrixTypeNested::InnerIterator MatrixTypeIterator;
-    typedef typename _MatrixTypeNested::ReverseInnerIterator MatrixTypeReverseIterator;
+    typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
+    
+    const ViewOp m_functor;
+    evaluator<ArgType> m_argImpl;
 };
 
-template<typename ViewOp, typename MatrixType>
-class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::InnerIterator
-    : public CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeIterator
+template<typename ViewOp, typename ArgType>
+class unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::InnerIterator
+    : public unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::EvalIterator
 {
-    typedef typename CwiseUnaryViewImpl::Scalar Scalar;
-    typedef typename CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeIterator Base;
+    typedef typename XprType::Scalar Scalar;
+    typedef typename unary_evaluator<CwiseUnaryView<ViewOp,ArgType>, IteratorBased>::EvalIterator Base;
   public:
 
-    EIGEN_STRONG_INLINE InnerIterator(const CwiseUnaryViewImpl& unaryOp, typename CwiseUnaryViewImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
+    EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& unaryOp, Index outer)
+      : Base(unaryOp.m_argImpl,outer), m_functor(unaryOp.m_functor)
     {}
 
     EIGEN_STRONG_INLINE InnerIterator& operator++()
     { Base::operator++(); return *this; }
 
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar value() const { return m_functor(Base::value()); }
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar& valueRef() { return m_functor(Base::valueRef()); }
+    EIGEN_STRONG_INLINE Scalar value() const { return m_functor(Base::value()); }
+    EIGEN_STRONG_INLINE Scalar& valueRef() { return m_functor(Base::valueRef()); }
 
   protected:
     const ViewOp m_functor;
 };
 
-template<typename ViewOp, typename MatrixType>
-class CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::ReverseInnerIterator
-    : public CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeReverseIterator
-{
-    typedef typename CwiseUnaryViewImpl::Scalar Scalar;
-    typedef typename CwiseUnaryViewImpl<ViewOp,MatrixType,Sparse>::MatrixTypeReverseIterator Base;
-  public:
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const CwiseUnaryViewImpl& unaryOp, typename CwiseUnaryViewImpl::Index outer)
-      : Base(unaryOp.derived().nestedExpression(),outer), m_functor(unaryOp.derived().functor())
-    {}
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-    { Base::operator--(); return *this; }
-
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar value() const { return m_functor(Base::value()); }
-    EIGEN_STRONG_INLINE typename CwiseUnaryViewImpl::Scalar& valueRef() { return m_functor(Base::valueRef()); }
-
-  protected:
-    const ViewOp m_functor;
-};
+} // end namespace internal
 
 template<typename Derived>
 EIGEN_STRONG_INLINE Derived&
 SparseMatrixBase<Derived>::operator*=(const Scalar& other)
 {
+  typedef typename internal::evaluator<Derived>::InnerIterator EvalIterator;
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator i(derived(),j); i; ++i)
+    for (EvalIterator i(thisEval,j); i; ++i)
       i.valueRef() *= other;
   return derived();
 }
@@ -152,8 +135,10 @@
 EIGEN_STRONG_INLINE Derived&
 SparseMatrixBase<Derived>::operator/=(const Scalar& other)
 {
+  typedef typename internal::evaluator<Derived>::InnerIterator EvalIterator;
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator i(derived(),j); i; ++i)
+    for (EvalIterator i(thisEval,j); i; ++i)
       i.valueRef() /= other;
   return derived();
 }
diff --git a/Eigen/src/SparseCore/SparseDenseProduct.h b/Eigen/src/SparseCore/SparseDenseProduct.h
index ccb6ae7..0547db5 100644
--- a/Eigen/src/SparseCore/SparseDenseProduct.h
+++ b/Eigen/src/SparseCore/SparseDenseProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,196 +12,93 @@
 
 namespace Eigen { 
 
-template<typename Lhs, typename Rhs, int InnerSize> struct SparseDenseProductReturnType
-{
-  typedef SparseTimeDenseProduct<Lhs,Rhs> Type;
-};
-
-template<typename Lhs, typename Rhs> struct SparseDenseProductReturnType<Lhs,Rhs,1>
-{
-  typedef typename internal::conditional<
-    Lhs::IsRowMajor,
-    SparseDenseOuterProduct<Rhs,Lhs,true>,
-    SparseDenseOuterProduct<Lhs,Rhs,false> >::type Type;
-};
-
-template<typename Lhs, typename Rhs, int InnerSize> struct DenseSparseProductReturnType
-{
-  typedef DenseTimeSparseProduct<Lhs,Rhs> Type;
-};
-
-template<typename Lhs, typename Rhs> struct DenseSparseProductReturnType<Lhs,Rhs,1>
-{
-  typedef typename internal::conditional<
-    Rhs::IsRowMajor,
-    SparseDenseOuterProduct<Rhs,Lhs,true>,
-    SparseDenseOuterProduct<Lhs,Rhs,false> >::type Type;
-};
-
 namespace internal {
 
-template<typename Lhs, typename Rhs, bool Tr>
-struct traits<SparseDenseOuterProduct<Lhs,Rhs,Tr> >
-{
-  typedef Sparse StorageKind;
-  typedef typename scalar_product_traits<typename traits<Lhs>::Scalar,
-                                         typename traits<Rhs>::Scalar>::ReturnType Scalar;
-  typedef typename Lhs::Index Index;
-  typedef typename Lhs::Nested LhsNested;
-  typedef typename Rhs::Nested RhsNested;
-  typedef typename remove_all<LhsNested>::type _LhsNested;
-  typedef typename remove_all<RhsNested>::type _RhsNested;
-
-  enum {
-    LhsCoeffReadCost = traits<_LhsNested>::CoeffReadCost,
-    RhsCoeffReadCost = traits<_RhsNested>::CoeffReadCost,
-
-    RowsAtCompileTime    = Tr ? int(traits<Rhs>::RowsAtCompileTime)     : int(traits<Lhs>::RowsAtCompileTime),
-    ColsAtCompileTime    = Tr ? int(traits<Lhs>::ColsAtCompileTime)     : int(traits<Rhs>::ColsAtCompileTime),
-    MaxRowsAtCompileTime = Tr ? int(traits<Rhs>::MaxRowsAtCompileTime)  : int(traits<Lhs>::MaxRowsAtCompileTime),
-    MaxColsAtCompileTime = Tr ? int(traits<Lhs>::MaxColsAtCompileTime)  : int(traits<Rhs>::MaxColsAtCompileTime),
-
-    Flags = Tr ? RowMajorBit : 0,
-
-    CoeffReadCost = LhsCoeffReadCost + RhsCoeffReadCost + NumTraits<Scalar>::MulCost
-  };
-};
-
-} // end namespace internal
-
-template<typename Lhs, typename Rhs, bool Tr>
-class SparseDenseOuterProduct
- : public SparseMatrixBase<SparseDenseOuterProduct<Lhs,Rhs,Tr> >
-{
-  public:
-
-    typedef SparseMatrixBase<SparseDenseOuterProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SparseDenseOuterProduct)
-    typedef internal::traits<SparseDenseOuterProduct> Traits;
-
-  private:
-
-    typedef typename Traits::LhsNested LhsNested;
-    typedef typename Traits::RhsNested RhsNested;
-    typedef typename Traits::_LhsNested _LhsNested;
-    typedef typename Traits::_RhsNested _RhsNested;
-
-  public:
-
-    class InnerIterator;
-
-    EIGEN_STRONG_INLINE SparseDenseOuterProduct(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      EIGEN_STATIC_ASSERT(!Tr,YOU_MADE_A_PROGRAMMING_MISTAKE);
-    }
-
-    EIGEN_STRONG_INLINE SparseDenseOuterProduct(const Rhs& rhs, const Lhs& lhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      EIGEN_STATIC_ASSERT(Tr,YOU_MADE_A_PROGRAMMING_MISTAKE);
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return Tr ? m_rhs.rows() : m_lhs.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return Tr ? m_lhs.cols() : m_rhs.cols(); }
-
-    EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
-    EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
-
-  protected:
-    LhsNested m_lhs;
-    RhsNested m_rhs;
-};
-
-template<typename Lhs, typename Rhs, bool Transpose>
-class SparseDenseOuterProduct<Lhs,Rhs,Transpose>::InnerIterator : public _LhsNested::InnerIterator
-{
-    typedef typename _LhsNested::InnerIterator Base;
-    typedef typename SparseDenseOuterProduct::Index Index;
-  public:
-    EIGEN_STRONG_INLINE InnerIterator(const SparseDenseOuterProduct& prod, Index outer)
-      : Base(prod.lhs(), 0), m_outer(outer), m_factor(get(prod.rhs(), outer, typename internal::traits<Rhs>::StorageKind() ))
-    { }
-
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return Transpose ? m_outer : Base::index(); }
-    inline Index col() const { return Transpose ? Base::index() : m_outer; }
-
-    inline Scalar value() const { return Base::value() * m_factor; }
-
-  protected:
-    static Scalar get(const _RhsNested &rhs, Index outer, Dense = Dense())
-    {
-      return rhs.coeff(outer);
-    }
-    
-    static Scalar get(const _RhsNested &rhs, Index outer, Sparse = Sparse())
-    {
-      typename Traits::_RhsNested::InnerIterator it(rhs, outer);
-      if (it && it.index()==0)
-        return it.value();
-      
-      return Scalar(0);
-    }
-    
-    Index m_outer;
-    Scalar m_factor;
-};
-
-namespace internal {
-template<typename Lhs, typename Rhs>
-struct traits<SparseTimeDenseProduct<Lhs,Rhs> >
- : traits<ProductBase<SparseTimeDenseProduct<Lhs,Rhs>, Lhs, Rhs> >
-{
-  typedef Dense StorageKind;
-  typedef MatrixXpr XprKind;
-};
+template <> struct product_promote_storage_type<Sparse,Dense, OuterProduct> { typedef Sparse ret; };
+template <> struct product_promote_storage_type<Dense,Sparse, OuterProduct> { typedef Sparse ret; };
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType,
+         typename AlphaType,
          int LhsStorageOrder = ((SparseLhsType::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor,
          bool ColPerCol = ((DenseRhsType::Flags&RowMajorBit)==0) || DenseRhsType::ColsAtCompileTime==1>
 struct sparse_time_dense_product_impl;
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, RowMajor, true>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, typename DenseResType::Scalar, RowMajor, true>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::Index Index;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
+  typedef evaluator<Lhs> LhsEval;
   static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
   {
+    LhsEval lhsEval(lhs);
+    
+    Index n = lhs.outerSize();
+#ifdef EIGEN_HAS_OPENMP
+    Eigen::initParallel();
+    Index threads = Eigen::nbThreads();
+#endif
+    
     for(Index c=0; c<rhs.cols(); ++c)
     {
-      Index n = lhs.outerSize();
-      for(Index j=0; j<n; ++j)
+#ifdef EIGEN_HAS_OPENMP
+      // This 20000 threshold has been found experimentally on 2D and 3D Poisson problems.
+      // It basically represents the minimal amount of work to be done to be worth it.
+      if(threads>1 && lhsEval.nonZerosEstimate() > 20000)
       {
-        typename Res::Scalar tmp(0);
-        for(LhsInnerIterator it(lhs,j); it ;++it)
-          tmp += it.value() * rhs.coeff(it.index(),c);
-        res.coeffRef(j,c) += alpha * tmp;
+        #pragma omp parallel for schedule(dynamic,(n+threads*4-1)/(threads*4)) num_threads(threads)
+        for(Index i=0; i<n; ++i)
+          processRow(lhsEval,rhs,res,alpha,i,c);
+      }
+      else
+#endif
+      {
+        for(Index i=0; i<n; ++i)
+          processRow(lhsEval,rhs,res,alpha,i,c);
       }
     }
   }
+  
+  static void processRow(const LhsEval& lhsEval, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha, Index i, Index col)
+  {
+    typename Res::Scalar tmp(0);
+    for(LhsInnerIterator it(lhsEval,i); it ;++it)
+      tmp += it.value() * rhs.coeff(it.index(),col);
+    res.coeffRef(i,col) += alpha * tmp;
+  }
+  
 };
 
-template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, ColMajor, true>
+// FIXME: what is the purpose of the following specialization? Is it for the BlockedSparse format?
+// -> let's disable it for now as it is conflicting with generic scalar*matrix and matrix*scalar operators
+// template<typename T1, typename T2/*, int _Options, typename _StrideType*/>
+// struct ScalarBinaryOpTraits<T1, Ref<T2/*, _Options, _StrideType*/> >
+// {
+//   enum {
+//     Defined = 1
+//   };
+//   typedef typename CwiseUnaryOp<scalar_multiple2_op<T1, typename T2::Scalar>, T2>::PlainObject ReturnType;
+// };
+
+template<typename SparseLhsType, typename DenseRhsType, typename DenseResType, typename AlphaType>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, AlphaType, ColMajor, true>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
-  typedef typename Lhs::Index Index;
-  static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
+  static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const AlphaType& alpha)
   {
+    evaluator<Lhs> lhsEval(lhs);
     for(Index c=0; c<rhs.cols(); ++c)
     {
       for(Index j=0; j<lhs.outerSize(); ++j)
       {
-        typename Res::Scalar rhs_j = alpha * rhs.coeff(j,c);
-        for(LhsInnerIterator it(lhs,j); it ;++it)
+//        typename Res::Scalar rhs_j = alpha * rhs.coeff(j,c);
+        typename ScalarBinaryOpTraits<AlphaType, typename Rhs::Scalar>::ReturnType rhs_j(alpha * rhs.coeff(j,c));
+        for(LhsInnerIterator it(lhsEval,j); it ;++it)
           res.coeffRef(it.index(),c) += it.value() * rhs_j;
       }
     }
@@ -209,38 +106,38 @@
 };
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, RowMajor, false>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, typename DenseResType::Scalar, RowMajor, false>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
-  typedef typename Lhs::Index Index;
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
   static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
   {
+    evaluator<Lhs> lhsEval(lhs);
     for(Index j=0; j<lhs.outerSize(); ++j)
     {
       typename Res::RowXpr res_j(res.row(j));
-      for(LhsInnerIterator it(lhs,j); it ;++it)
+      for(LhsInnerIterator it(lhsEval,j); it ;++it)
         res_j += (alpha*it.value()) * rhs.row(it.index());
     }
   }
 };
 
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType>
-struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, ColMajor, false>
+struct sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, typename DenseResType::Scalar, ColMajor, false>
 {
   typedef typename internal::remove_all<SparseLhsType>::type Lhs;
   typedef typename internal::remove_all<DenseRhsType>::type Rhs;
   typedef typename internal::remove_all<DenseResType>::type Res;
-  typedef typename Lhs::InnerIterator LhsInnerIterator;
-  typedef typename Lhs::Index Index;
+  typedef typename evaluator<Lhs>::InnerIterator LhsInnerIterator;
   static void run(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const typename Res::Scalar& alpha)
   {
+    evaluator<Lhs> lhsEval(lhs);
     for(Index j=0; j<lhs.outerSize(); ++j)
     {
       typename Rhs::ConstRowXpr rhs_j(rhs.row(j));
-      for(LhsInnerIterator it(lhs,j); it ;++it)
+      for(LhsInnerIterator it(lhsEval,j); it ;++it)
         res.row(it.index()) += (alpha*it.value()) * rhs_j;
     }
   }
@@ -249,63 +146,175 @@
 template<typename SparseLhsType, typename DenseRhsType, typename DenseResType,typename AlphaType>
 inline void sparse_time_dense_product(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const AlphaType& alpha)
 {
-  sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType>::run(lhs, rhs, res, alpha);
+  sparse_time_dense_product_impl<SparseLhsType,DenseRhsType,DenseResType, AlphaType>::run(lhs, rhs, res, alpha);
 }
 
 } // end namespace internal
 
-template<typename Lhs, typename Rhs>
-class SparseTimeDenseProduct
-  : public ProductBase<SparseTimeDenseProduct<Lhs,Rhs>, Lhs, Rhs>
-{
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(SparseTimeDenseProduct)
-
-    SparseTimeDenseProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {}
-
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
-    {
-      internal::sparse_time_dense_product(m_lhs, m_rhs, dest, alpha);
-    }
-
-  private:
-    SparseTimeDenseProduct& operator=(const SparseTimeDenseProduct&);
-};
-
-
-// dense = dense * sparse
 namespace internal {
-template<typename Lhs, typename Rhs>
-struct traits<DenseTimeSparseProduct<Lhs,Rhs> >
- : traits<ProductBase<DenseTimeSparseProduct<Lhs,Rhs>, Lhs, Rhs> >
-{
-  typedef Dense StorageKind;
-};
-} // end namespace internal
 
-template<typename Lhs, typename Rhs>
-class DenseTimeSparseProduct
-  : public ProductBase<DenseTimeSparseProduct<Lhs,Rhs>, Lhs, Rhs>
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseShape, DenseShape, ProductType>
+ : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,SparseShape,DenseShape,ProductType> >
 {
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    typedef typename nested_eval<Lhs,((Rhs::Flags&RowMajorBit)==0) ? 1 : Rhs::ColsAtCompileTime>::type LhsNested;
+    typedef typename nested_eval<Rhs,((Lhs::Flags&RowMajorBit)==0) ? 1 : Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    internal::sparse_time_dense_product(lhsNested, rhsNested, dst, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseTriangularShape, DenseShape, ProductType>
+  : generic_product_impl<Lhs, Rhs, SparseShape, DenseShape, ProductType>
+{};
+
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, DenseShape, SparseShape, ProductType>
+  : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,SparseShape,ProductType> >
+{
+  typedef typename Product<Lhs,Rhs>::Scalar Scalar;
+  
+  template<typename Dst>
+  static void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha)
+  {
+    typedef typename nested_eval<Lhs,((Rhs::Flags&RowMajorBit)==0) ? Dynamic : 1>::type LhsNested;
+    typedef typename nested_eval<Rhs,((Lhs::Flags&RowMajorBit)==RowMajorBit) ? 1 : Lhs::RowsAtCompileTime>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    
+    // transpose everything
+    Transpose<Dst> dstT(dst);
+    internal::sparse_time_dense_product(rhsNested.transpose(), lhsNested.transpose(), dstT, alpha);
+  }
+};
+
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, DenseShape, SparseTriangularShape, ProductType>
+  : generic_product_impl<Lhs, Rhs, DenseShape, SparseShape, ProductType>
+{};
+
+template<typename LhsT, typename RhsT, bool NeedToTranspose>
+struct sparse_dense_outer_product_evaluator
+{
+protected:
+  typedef typename conditional<NeedToTranspose,RhsT,LhsT>::type Lhs1;
+  typedef typename conditional<NeedToTranspose,LhsT,RhsT>::type ActualRhs;
+  typedef Product<LhsT,RhsT,DefaultProduct> ProdXprType;
+  
+  // if the actual left-hand side is a dense vector,
+  // then build a sparse-view so that we can seamlessly iterate over it.
+  typedef typename conditional<is_same<typename internal::traits<Lhs1>::StorageKind,Sparse>::value,
+            Lhs1, SparseView<Lhs1> >::type ActualLhs;
+  typedef typename conditional<is_same<typename internal::traits<Lhs1>::StorageKind,Sparse>::value,
+            Lhs1 const&, SparseView<Lhs1> >::type LhsArg;
+            
+  typedef evaluator<ActualLhs> LhsEval;
+  typedef evaluator<ActualRhs> RhsEval;
+  typedef typename evaluator<ActualLhs>::InnerIterator LhsIterator;
+  typedef typename ProdXprType::Scalar Scalar;
+  
+public:
+  enum {
+    Flags = NeedToTranspose ? RowMajorBit : 0,
+    CoeffReadCost = HugeCost
+  };
+  
+  class InnerIterator : public LhsIterator
+  {
   public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(DenseTimeSparseProduct)
-
-    DenseTimeSparseProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
+    InnerIterator(const sparse_dense_outer_product_evaluator &xprEval, Index outer)
+      : LhsIterator(xprEval.m_lhsXprImpl, 0),
+        m_outer(outer),
+        m_empty(false),
+        m_factor(get(xprEval.m_rhsXprImpl, outer, typename internal::traits<ActualRhs>::StorageKind() ))
     {}
+    
+    EIGEN_STRONG_INLINE Index outer() const { return m_outer; }
+    EIGEN_STRONG_INLINE Index row()   const { return NeedToTranspose ? m_outer : LhsIterator::index(); }
+    EIGEN_STRONG_INLINE Index col()   const { return NeedToTranspose ? LhsIterator::index() : m_outer; }
 
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+    EIGEN_STRONG_INLINE Scalar value() const { return LhsIterator::value() * m_factor; }
+    EIGEN_STRONG_INLINE operator bool() const { return LhsIterator::operator bool() && (!m_empty); }
+    
+  protected:
+    Scalar get(const RhsEval &rhs, Index outer, Dense = Dense()) const
     {
-      Transpose<const _LhsNested> lhs_t(m_lhs);
-      Transpose<const _RhsNested> rhs_t(m_rhs);
-      Transpose<Dest> dest_t(dest);
-      internal::sparse_time_dense_product(rhs_t, lhs_t, dest_t, alpha);
+      return rhs.coeff(outer);
     }
-
-  private:
-    DenseTimeSparseProduct& operator=(const DenseTimeSparseProduct&);
+    
+    Scalar get(const RhsEval &rhs, Index outer, Sparse = Sparse())
+    {
+      typename RhsEval::InnerIterator it(rhs, outer);
+      if (it && it.index()==0 && it.value()!=Scalar(0))
+        return it.value();
+      m_empty = true;
+      return Scalar(0);
+    }
+    
+    Index m_outer;
+    bool m_empty;
+    Scalar m_factor;
+  };
+  
+  sparse_dense_outer_product_evaluator(const Lhs1 &lhs, const ActualRhs &rhs)
+     : m_lhs(lhs), m_lhsXprImpl(m_lhs), m_rhsXprImpl(rhs)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  // transpose case
+  sparse_dense_outer_product_evaluator(const ActualRhs &rhs, const Lhs1 &lhs)
+     : m_lhs(lhs), m_lhsXprImpl(m_lhs), m_rhsXprImpl(rhs)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+    
+protected:
+  const LhsArg m_lhs;
+  evaluator<ActualLhs> m_lhsXprImpl;
+  evaluator<ActualRhs> m_rhsXprImpl;
 };
 
+// sparse * dense outer product
+template<typename Lhs, typename Rhs>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, OuterProduct, SparseShape, DenseShape>
+  : sparse_dense_outer_product_evaluator<Lhs,Rhs, Lhs::IsRowMajor>
+{
+  typedef sparse_dense_outer_product_evaluator<Lhs,Rhs, Lhs::IsRowMajor> Base;
+  
+  typedef Product<Lhs, Rhs> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+
+  explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.lhs(), xpr.rhs())
+  {}
+  
+};
+
+template<typename Lhs, typename Rhs>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, OuterProduct, DenseShape, SparseShape>
+  : sparse_dense_outer_product_evaluator<Lhs,Rhs, Rhs::IsRowMajor>
+{
+  typedef sparse_dense_outer_product_evaluator<Lhs,Rhs, Rhs::IsRowMajor> Base;
+  
+  typedef Product<Lhs, Rhs> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+
+  explicit product_evaluator(const XprType& xpr)
+    : Base(xpr.lhs(), xpr.rhs())
+  {}
+  
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSEDENSEPRODUCT_H
diff --git a/Eigen/src/SparseCore/SparseDiagonalProduct.h b/Eigen/src/SparseCore/SparseDiagonalProduct.h
index 1bb590e..941c03b 100644
--- a/Eigen/src/SparseCore/SparseDiagonalProduct.h
+++ b/Eigen/src/SparseCore/SparseDiagonalProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -26,171 +26,113 @@
 
 namespace internal {
 
-template<typename Lhs, typename Rhs>
-struct traits<SparseDiagonalProduct<Lhs, Rhs> >
+enum {
+  SDP_AsScalarProduct,
+  SDP_AsCwiseProduct
+};
+  
+template<typename SparseXprType, typename DiagonalCoeffType, int SDP_Tag>
+struct sparse_diagonal_product_evaluator;
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, ProductTag, DiagonalShape, SparseShape>
+  : public sparse_diagonal_product_evaluator<Rhs, typename Lhs::DiagonalVectorType, Rhs::Flags&RowMajorBit?SDP_AsScalarProduct:SDP_AsCwiseProduct>
 {
-  typedef typename remove_all<Lhs>::type _Lhs;
-  typedef typename remove_all<Rhs>::type _Rhs;
-  typedef typename _Lhs::Scalar Scalar;
-  typedef typename promote_index_type<typename traits<Lhs>::Index,
-                                         typename traits<Rhs>::Index>::type Index;
-  typedef Sparse StorageKind;
-  typedef MatrixXpr XprKind;
-  enum {
-    RowsAtCompileTime = _Lhs::RowsAtCompileTime,
-    ColsAtCompileTime = _Rhs::ColsAtCompileTime,
-
-    MaxRowsAtCompileTime = _Lhs::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = _Rhs::MaxColsAtCompileTime,
-
-    SparseFlags = is_diagonal<_Lhs>::ret ? int(_Rhs::Flags) : int(_Lhs::Flags),
-    Flags = (SparseFlags&RowMajorBit),
-    CoeffReadCost = Dynamic
-  };
+  typedef Product<Lhs, Rhs, DefaultProduct> XprType;
+  enum { CoeffReadCost = HugeCost, Flags = Rhs::Flags&RowMajorBit, Alignment = 0 }; // FIXME CoeffReadCost & Flags
+  
+  typedef sparse_diagonal_product_evaluator<Rhs, typename Lhs::DiagonalVectorType, Rhs::Flags&RowMajorBit?SDP_AsScalarProduct:SDP_AsCwiseProduct> Base;
+  explicit product_evaluator(const XprType& xpr) : Base(xpr.rhs(), xpr.lhs().diagonal()) {}
 };
 
-enum {SDP_IsDiagonal, SDP_IsSparseRowMajor, SDP_IsSparseColMajor};
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType, int RhsMode, int LhsMode>
-class sparse_diagonal_product_inner_iterator_selector;
-
-} // end namespace internal
-
-template<typename Lhs, typename Rhs>
-class SparseDiagonalProduct
-  : public SparseMatrixBase<SparseDiagonalProduct<Lhs,Rhs> >,
-    internal::no_assignment_operator
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, DefaultProduct>, ProductTag, SparseShape, DiagonalShape>
+  : public sparse_diagonal_product_evaluator<Lhs, Transpose<const typename Rhs::DiagonalVectorType>, Lhs::Flags&RowMajorBit?SDP_AsCwiseProduct:SDP_AsScalarProduct>
 {
-    typedef typename Lhs::Nested LhsNested;
-    typedef typename Rhs::Nested RhsNested;
+  typedef Product<Lhs, Rhs, DefaultProduct> XprType;
+  enum { CoeffReadCost = HugeCost, Flags = Lhs::Flags&RowMajorBit, Alignment = 0 }; // FIXME CoeffReadCost & Flags
+  
+  typedef sparse_diagonal_product_evaluator<Lhs, Transpose<const typename Rhs::DiagonalVectorType>, Lhs::Flags&RowMajorBit?SDP_AsCwiseProduct:SDP_AsScalarProduct> Base;
+  explicit product_evaluator(const XprType& xpr) : Base(xpr.lhs(), xpr.rhs().diagonal().transpose()) {}
+};
 
-    typedef typename internal::remove_all<LhsNested>::type _LhsNested;
-    typedef typename internal::remove_all<RhsNested>::type _RhsNested;
-
-    enum {
-      LhsMode = internal::is_diagonal<_LhsNested>::ret ? internal::SDP_IsDiagonal
-              : (_LhsNested::Flags&RowMajorBit) ? internal::SDP_IsSparseRowMajor : internal::SDP_IsSparseColMajor,
-      RhsMode = internal::is_diagonal<_RhsNested>::ret ? internal::SDP_IsDiagonal
-              : (_RhsNested::Flags&RowMajorBit) ? internal::SDP_IsSparseRowMajor : internal::SDP_IsSparseColMajor
-    };
-
+template<typename SparseXprType, typename DiagonalCoeffType>
+struct sparse_diagonal_product_evaluator<SparseXprType, DiagonalCoeffType, SDP_AsScalarProduct>
+{
+protected:
+  typedef typename evaluator<SparseXprType>::InnerIterator SparseXprInnerIterator;
+  typedef typename SparseXprType::Scalar Scalar;
+  
+public:
+  class InnerIterator : public SparseXprInnerIterator
+  {
   public:
-
-    EIGEN_SPARSE_PUBLIC_INTERFACE(SparseDiagonalProduct)
-
-    typedef internal::sparse_diagonal_product_inner_iterator_selector
-                      <_LhsNested,_RhsNested,SparseDiagonalProduct,LhsMode,RhsMode> InnerIterator;
+    InnerIterator(const sparse_diagonal_product_evaluator &xprEval, Index outer)
+      : SparseXprInnerIterator(xprEval.m_sparseXprImpl, outer),
+        m_coeff(xprEval.m_diagCoeffImpl.coeff(outer))
+    {}
     
-    // We do not want ReverseInnerIterator for diagonal-sparse products,
-    // but this dummy declaration is needed to make diag * sparse * diag compile.
-    class ReverseInnerIterator;
-
-    EIGEN_STRONG_INLINE SparseDiagonalProduct(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs)
-    {
-      eigen_assert(lhs.cols() == rhs.rows() && "invalid sparse matrix * diagonal matrix product");
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
-
-    EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
-    EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
-
+    EIGEN_STRONG_INLINE Scalar value() const { return m_coeff * SparseXprInnerIterator::value(); }
   protected:
-    LhsNested m_lhs;
-    RhsNested m_rhs;
+    typename DiagonalCoeffType::Scalar m_coeff;
+  };
+  
+  sparse_diagonal_product_evaluator(const SparseXprType &sparseXpr, const DiagonalCoeffType &diagCoeff)
+    : m_sparseXprImpl(sparseXpr), m_diagCoeffImpl(diagCoeff)
+  {}
+
+  Index nonZerosEstimate() const { return m_sparseXprImpl.nonZerosEstimate(); }
+    
+protected:
+  evaluator<SparseXprType> m_sparseXprImpl;
+  evaluator<DiagonalCoeffType> m_diagCoeffImpl;
 };
 
-namespace internal {
 
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsDiagonal,SDP_IsSparseRowMajor>
-  : public CwiseUnaryOp<scalar_multiple_op<typename Lhs::Scalar>,const Rhs>::InnerIterator
+template<typename SparseXprType, typename DiagCoeffType>
+struct sparse_diagonal_product_evaluator<SparseXprType, DiagCoeffType, SDP_AsCwiseProduct>
 {
-    typedef typename CwiseUnaryOp<scalar_multiple_op<typename Lhs::Scalar>,const Rhs>::InnerIterator Base;
-    typedef typename Lhs::Index Index;
+  typedef typename SparseXprType::Scalar Scalar;
+  typedef typename SparseXprType::StorageIndex StorageIndex;
+  
+  typedef typename nested_eval<DiagCoeffType,SparseXprType::IsRowMajor ? SparseXprType::RowsAtCompileTime
+                                                                       : SparseXprType::ColsAtCompileTime>::type DiagCoeffNested;
+  
+  class InnerIterator
+  {
+    typedef typename evaluator<SparseXprType>::InnerIterator SparseXprIter;
   public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.rhs()*(expr.lhs().diagonal().coeff(outer)), outer)
-    {}
-};
-
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsDiagonal,SDP_IsSparseColMajor>
-  : public CwiseBinaryOp<
-      scalar_product_op<typename Lhs::Scalar>,
-      const typename Rhs::ConstInnerVectorReturnType,
-      const typename Lhs::DiagonalVectorType>::InnerIterator
-{
-    typedef typename CwiseBinaryOp<
-      scalar_product_op<typename Lhs::Scalar>,
-      const typename Rhs::ConstInnerVectorReturnType,
-      const typename Lhs::DiagonalVectorType>::InnerIterator Base;
-    typedef typename Lhs::Index Index;
-    Index m_outer;
-  public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.rhs().innerVector(outer) .cwiseProduct(expr.lhs().diagonal()), 0), m_outer(outer)
+    InnerIterator(const sparse_diagonal_product_evaluator &xprEval, Index outer)
+      : m_sparseIter(xprEval.m_sparseXprEval, outer), m_diagCoeffNested(xprEval.m_diagCoeffNested)
     {}
     
-    inline Index outer() const { return m_outer; }
-    inline Index col() const { return m_outer; }
-};
-
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsSparseColMajor,SDP_IsDiagonal>
-  : public CwiseUnaryOp<scalar_multiple_op<typename Rhs::Scalar>,const Lhs>::InnerIterator
-{
-    typedef typename CwiseUnaryOp<scalar_multiple_op<typename Rhs::Scalar>,const Lhs>::InnerIterator Base;
-    typedef typename Lhs::Index Index;
-  public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.lhs()*expr.rhs().diagonal().coeff(outer), outer)
-    {}
-};
-
-template<typename Lhs, typename Rhs, typename SparseDiagonalProductType>
-class sparse_diagonal_product_inner_iterator_selector
-<Lhs,Rhs,SparseDiagonalProductType,SDP_IsSparseRowMajor,SDP_IsDiagonal>
-  : public CwiseBinaryOp<
-      scalar_product_op<typename Rhs::Scalar>,
-      const typename Lhs::ConstInnerVectorReturnType,
-      const Transpose<const typename Rhs::DiagonalVectorType> >::InnerIterator
-{
-    typedef typename CwiseBinaryOp<
-      scalar_product_op<typename Rhs::Scalar>,
-      const typename Lhs::ConstInnerVectorReturnType,
-      const Transpose<const typename Rhs::DiagonalVectorType> >::InnerIterator Base;
-    typedef typename Lhs::Index Index;
-    Index m_outer;
-  public:
-    inline sparse_diagonal_product_inner_iterator_selector(
-              const SparseDiagonalProductType& expr, Index outer)
-      : Base(expr.lhs().innerVector(outer) .cwiseProduct(expr.rhs().diagonal().transpose()), 0), m_outer(outer)
-    {}
+    inline Scalar value() const { return m_sparseIter.value() * m_diagCoeffNested.coeff(index()); }
+    inline StorageIndex index() const  { return m_sparseIter.index(); }
+    inline Index outer() const  { return m_sparseIter.outer(); }
+    inline Index col() const    { return SparseXprType::IsRowMajor ? m_sparseIter.index() : m_sparseIter.outer(); }
+    inline Index row() const    { return SparseXprType::IsRowMajor ? m_sparseIter.outer() : m_sparseIter.index(); }
     
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return m_outer; }
+    EIGEN_STRONG_INLINE InnerIterator& operator++() { ++m_sparseIter; return *this; }
+    inline operator bool() const  { return m_sparseIter; }
+    
+  protected:
+    SparseXprIter m_sparseIter;
+    DiagCoeffNested m_diagCoeffNested;
+  };
+  
+  sparse_diagonal_product_evaluator(const SparseXprType &sparseXpr, const DiagCoeffType &diagCoeff)
+    : m_sparseXprEval(sparseXpr), m_diagCoeffNested(diagCoeff)
+  {}
+
+  Index nonZerosEstimate() const { return m_sparseXprEval.nonZerosEstimate(); }
+    
+protected:
+  evaluator<SparseXprType> m_sparseXprEval;
+  DiagCoeffNested m_diagCoeffNested;
 };
 
 } // end namespace internal
 
-// SparseMatrixBase functions
-
-template<typename Derived>
-template<typename OtherDerived>
-const SparseDiagonalProduct<Derived,OtherDerived>
-SparseMatrixBase<Derived>::operator*(const DiagonalBase<OtherDerived> &other) const
-{
-  return SparseDiagonalProduct<Derived,OtherDerived>(this->derived(), other.derived());
-}
-
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSE_DIAGONAL_PRODUCT_H
diff --git a/Eigen/src/SparseCore/SparseDot.h b/Eigen/src/SparseCore/SparseDot.h
index db39c9a..38bc4aa 100644
--- a/Eigen/src/SparseCore/SparseDot.h
+++ b/Eigen/src/SparseCore/SparseDot.h
@@ -26,7 +26,8 @@
   eigen_assert(size() == other.size());
   eigen_assert(other.size()>0 && "you are using a non initialized vector");
 
-  typename Derived::InnerIterator i(derived(),0);
+  internal::evaluator<Derived> thisEval(derived());
+  typename internal::evaluator<Derived>::InnerIterator i(thisEval, 0);
   Scalar res(0);
   while (i)
   {
@@ -49,16 +50,12 @@
 
   eigen_assert(size() == other.size());
 
-  typedef typename Derived::Nested  Nested;
-  typedef typename OtherDerived::Nested  OtherNested;
-  typedef typename internal::remove_all<Nested>::type  NestedCleaned;
-  typedef typename internal::remove_all<OtherNested>::type  OtherNestedCleaned;
+  internal::evaluator<Derived> thisEval(derived());
+  typename internal::evaluator<Derived>::InnerIterator i(thisEval, 0);
+  
+  internal::evaluator<OtherDerived>  otherEval(other.derived());
+  typename internal::evaluator<OtherDerived>::InnerIterator j(otherEval, 0);
 
-  Nested nthis(derived());
-  OtherNested nother(other.derived());
-
-  typename NestedCleaned::InnerIterator i(nthis,0);
-  typename OtherNestedCleaned::InnerIterator j(nother,0);
   Scalar res(0);
   while (i && j)
   {
diff --git a/Eigen/src/SparseCore/SparseFuzzy.h b/Eigen/src/SparseCore/SparseFuzzy.h
index 45f36e9..7d47eb9 100644
--- a/Eigen/src/SparseCore/SparseFuzzy.h
+++ b/Eigen/src/SparseCore/SparseFuzzy.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,17 +10,20 @@
 #ifndef EIGEN_SPARSE_FUZZY_H
 #define EIGEN_SPARSE_FUZZY_H
 
-// template<typename Derived>
-// template<typename OtherDerived>
-// bool SparseMatrixBase<Derived>::isApprox(
-//   const OtherDerived& other,
-//   typename NumTraits<Scalar>::Real prec
-// ) const
-// {
-//   const typename internal::nested<Derived,2>::type nested(derived());
-//   const typename internal::nested<OtherDerived,2>::type otherNested(other.derived());
-//   return    (nested - otherNested).cwise().abs2().sum()
-//          <= prec * prec * (std::min)(nested.cwise().abs2().sum(), otherNested.cwise().abs2().sum());
-// }
+namespace Eigen {
+  
+template<typename Derived>
+template<typename OtherDerived>
+bool SparseMatrixBase<Derived>::isApprox(const SparseMatrixBase<OtherDerived>& other, const RealScalar &prec) const
+{
+  const typename internal::nested_eval<Derived,2,PlainObject>::type actualA(derived());
+  typename internal::conditional<bool(IsRowMajor)==bool(OtherDerived::IsRowMajor),
+    const typename internal::nested_eval<OtherDerived,2,PlainObject>::type,
+    const PlainObject>::type actualB(other.derived());
+
+  return (actualA - actualB).squaredNorm() <= prec * prec * numext::mini(actualA.squaredNorm(), actualB.squaredNorm());
+}
+
+} // end namespace Eigen
 
 #endif // EIGEN_SPARSE_FUZZY_H
diff --git a/Eigen/src/SparseCore/SparseMap.h b/Eigen/src/SparseCore/SparseMap.h
new file mode 100644
index 0000000..f99be33
--- /dev/null
+++ b/Eigen/src/SparseCore/SparseMap.h
@@ -0,0 +1,305 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSE_MAP_H
+#define EIGEN_SPARSE_MAP_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct traits<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : public traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  typedef traits<PlainObjectType> TraitsBase;
+  enum {
+    Flags = TraitsBase::Flags & (~NestByRefBit)
+  };
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct traits<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : public traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  typedef traits<PlainObjectType> TraitsBase;
+  enum {
+    Flags = TraitsBase::Flags & (~ (NestByRefBit | LvalueBit))
+  };
+};
+
+} // end namespace internal
+
+template<typename Derived,
+         int Level = internal::accessors_level<Derived>::has_write_access ? WriteAccessors : ReadOnlyAccessors
+> class SparseMapBase;
+
+/** \ingroup SparseCore_Module
+  * class SparseMapBase
+  * \brief Common base class for Map and Ref instance of sparse matrix and vector.
+  */
+template<typename Derived>
+class SparseMapBase<Derived,ReadOnlyAccessors>
+  : public SparseCompressedBase<Derived>
+{
+  public:
+    typedef SparseCompressedBase<Derived> Base;
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::StorageIndex StorageIndex;
+    enum { IsRowMajor = Base::IsRowMajor };
+    using Base::operator=;
+  protected:
+    
+    typedef typename internal::conditional<
+                         bool(internal::is_lvalue<Derived>::value),
+                         Scalar *, const Scalar *>::type ScalarPointer;
+    typedef typename internal::conditional<
+                         bool(internal::is_lvalue<Derived>::value),
+                         StorageIndex *, const StorageIndex *>::type IndexPointer;
+
+    Index   m_outerSize;
+    Index   m_innerSize;
+    Array<StorageIndex,2,1>  m_zero_nnz;
+    IndexPointer  m_outerIndex;
+    IndexPointer  m_innerIndices;
+    ScalarPointer m_values;
+    IndexPointer  m_innerNonZeros;
+
+  public:
+
+    /** \copydoc SparseMatrixBase::rows() */
+    inline Index rows() const { return IsRowMajor ? m_outerSize : m_innerSize; }
+    /** \copydoc SparseMatrixBase::cols() */
+    inline Index cols() const { return IsRowMajor ? m_innerSize : m_outerSize; }
+    /** \copydoc SparseMatrixBase::innerSize() */
+    inline Index innerSize() const { return m_innerSize; }
+    /** \copydoc SparseMatrixBase::outerSize() */
+    inline Index outerSize() const { return m_outerSize; }
+    /** \copydoc SparseCompressedBase::nonZeros */
+    inline Index nonZeros() const { return m_zero_nnz[1]; }
+    
+    /** \copydoc SparseCompressedBase::isCompressed */
+    bool isCompressed() const { return m_innerNonZeros==0; }
+
+    //----------------------------------------
+    // direct access interface
+    /** \copydoc SparseMatrix::valuePtr */
+    inline const Scalar* valuePtr() const { return m_values; }
+    /** \copydoc SparseMatrix::innerIndexPtr */
+    inline const StorageIndex* innerIndexPtr() const { return m_innerIndices; }
+    /** \copydoc SparseMatrix::outerIndexPtr */
+    inline const StorageIndex* outerIndexPtr() const { return m_outerIndex; }
+    /** \copydoc SparseMatrix::innerNonZeroPtr */
+    inline const StorageIndex* innerNonZeroPtr() const { return m_innerNonZeros; }
+    //----------------------------------------
+
+    /** \copydoc SparseMatrix::coeff */
+    inline Scalar coeff(Index row, Index col) const
+    {
+      const Index outer = IsRowMajor ? row : col;
+      const Index inner = IsRowMajor ? col : row;
+
+      Index start = m_outerIndex[outer];
+      Index end = isCompressed() ? m_outerIndex[outer+1] : start + m_innerNonZeros[outer];
+      if (start==end)
+        return Scalar(0);
+      else if (end>0 && inner==m_innerIndices[end-1])
+        return m_values[end-1];
+      // ^^  optimization: let's first check if it is the last coefficient
+      // (very common in high level algorithms)
+
+      const StorageIndex* r = std::lower_bound(&m_innerIndices[start],&m_innerIndices[end-1],inner);
+      const Index id = r-&m_innerIndices[0];
+      return ((*r==inner) && (id<end)) ? m_values[id] : Scalar(0);
+    }
+
+    inline SparseMapBase(Index rows, Index cols, Index nnz, IndexPointer outerIndexPtr, IndexPointer innerIndexPtr,
+                              ScalarPointer valuePtr, IndexPointer innerNonZerosPtr = 0)
+      : m_outerSize(IsRowMajor?rows:cols), m_innerSize(IsRowMajor?cols:rows), m_zero_nnz(0,internal::convert_index<StorageIndex>(nnz)), m_outerIndex(outerIndexPtr),
+        m_innerIndices(innerIndexPtr), m_values(valuePtr), m_innerNonZeros(innerNonZerosPtr)
+    {}
+
+    // for vectors
+    inline SparseMapBase(Index size, Index nnz, IndexPointer innerIndexPtr, ScalarPointer valuePtr)
+      : m_outerSize(1), m_innerSize(size), m_zero_nnz(0,internal::convert_index<StorageIndex>(nnz)), m_outerIndex(m_zero_nnz.data()),
+        m_innerIndices(innerIndexPtr), m_values(valuePtr), m_innerNonZeros(0)
+    {}
+
+    /** Empty destructor */
+    inline ~SparseMapBase() {}
+
+  protected:
+    inline SparseMapBase() {}
+};
+
+/** \ingroup SparseCore_Module
+  * class SparseMapBase
+  * \brief Common base class for writable Map and Ref instance of sparse matrix and vector.
+  */
+template<typename Derived>
+class SparseMapBase<Derived,WriteAccessors>
+  : public SparseMapBase<Derived,ReadOnlyAccessors>
+{
+    typedef MapBase<Derived, ReadOnlyAccessors> ReadOnlyMapBase;
+    
+  public:
+    typedef SparseMapBase<Derived, ReadOnlyAccessors> Base;
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::StorageIndex StorageIndex;
+    enum { IsRowMajor = Base::IsRowMajor };
+    
+    using Base::operator=;
+
+  public:
+    
+    //----------------------------------------
+    // direct access interface
+    using Base::valuePtr;
+    using Base::innerIndexPtr;
+    using Base::outerIndexPtr;
+    using Base::innerNonZeroPtr;
+    /** \copydoc SparseMatrix::valuePtr */
+    inline Scalar* valuePtr()              { return Base::m_values; }
+    /** \copydoc SparseMatrix::innerIndexPtr */
+    inline StorageIndex* innerIndexPtr()   { return Base::m_innerIndices; }
+    /** \copydoc SparseMatrix::outerIndexPtr */
+    inline StorageIndex* outerIndexPtr()   { return Base::m_outerIndex; }
+    /** \copydoc SparseMatrix::innerNonZeroPtr */
+    inline StorageIndex* innerNonZeroPtr() { return Base::m_innerNonZeros; }
+    //----------------------------------------
+
+    /** \copydoc SparseMatrix::coeffRef */
+    inline Scalar& coeffRef(Index row, Index col)
+    {
+      const Index outer = IsRowMajor ? row : col;
+      const Index inner = IsRowMajor ? col : row;
+
+      Index start = Base::m_outerIndex[outer];
+      Index end = Base::isCompressed() ? Base::m_outerIndex[outer+1] : start + Base::m_innerNonZeros[outer];
+      eigen_assert(end>=start && "you probably called coeffRef on a non finalized matrix");
+      eigen_assert(end>start && "coeffRef cannot be called on a zero coefficient");
+      StorageIndex* r = std::lower_bound(&Base::m_innerIndices[start],&Base::m_innerIndices[end],inner);
+      const Index id = r - &Base::m_innerIndices[0];
+      eigen_assert((*r==inner) && (id<end) && "coeffRef cannot be called on a zero coefficient");
+      return const_cast<Scalar*>(Base::m_values)[id];
+    }
+    
+    inline SparseMapBase(Index rows, Index cols, Index nnz, StorageIndex* outerIndexPtr, StorageIndex* innerIndexPtr,
+                         Scalar* valuePtr, StorageIndex* innerNonZerosPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZerosPtr)
+    {}
+
+    // for vectors
+    inline SparseMapBase(Index size, Index nnz, StorageIndex* innerIndexPtr, Scalar* valuePtr)
+      : Base(size, nnz, innerIndexPtr, valuePtr)
+    {}
+
+    /** Empty destructor */
+    inline ~SparseMapBase() {}
+
+  protected:
+    inline SparseMapBase() {}
+};
+
+/** \ingroup SparseCore_Module
+  *
+  * \brief Specialization of class Map for SparseMatrix-like storage.
+  *
+  * \tparam SparseMatrixType the equivalent sparse matrix type of the referenced data, it must be a template instance of class SparseMatrix.
+  *
+  * \sa class Map, class SparseMatrix, class Ref<SparseMatrixType,Options>
+  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public SparseMapBase<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+#else
+template<typename SparseMatrixType>
+class Map<SparseMatrixType>
+  : public SparseMapBase<Derived,WriteAccessors>
+#endif
+{
+  public:
+    typedef SparseMapBase<Map> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Map)
+    enum { IsRowMajor = Base::IsRowMajor };
+
+  public:
+
+    /** Constructs a read-write Map to a sparse matrix of size \a rows x \a cols, containing \a nnz non-zero coefficients,
+      * stored as a sparse format as defined by the pointers \a outerIndexPtr, \a innerIndexPtr, and \a valuePtr.
+      * If the optional parameter \a innerNonZerosPtr is the null pointer, then a standard compressed format is assumed.
+      *
+      * This constructor is available only if \c SparseMatrixType is non-const.
+      *
+      * More details on the expected storage schemes are given in the \ref TutorialSparse "manual pages".
+      */
+    inline Map(Index rows, Index cols, Index nnz, StorageIndex* outerIndexPtr,
+               StorageIndex* innerIndexPtr, Scalar* valuePtr, StorageIndex* innerNonZerosPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZerosPtr)
+    {}
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** Empty destructor */
+    inline ~Map() {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public SparseMapBase<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+{
+  public:
+    typedef SparseMapBase<Map> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Map)
+    enum { IsRowMajor = Base::IsRowMajor };
+
+  public:
+#endif
+    /** This is the const version of the above constructor.
+      *
+      * This constructor is available only if \c SparseMatrixType is const, e.g.:
+      * \code Map<const SparseMatrix<double> >  \endcode
+      */
+    inline Map(Index rows, Index cols, Index nnz, const StorageIndex* outerIndexPtr,
+               const StorageIndex* innerIndexPtr, const Scalar* valuePtr, const StorageIndex* innerNonZerosPtr = 0)
+      : Base(rows, cols, nnz, outerIndexPtr, innerIndexPtr, valuePtr, innerNonZerosPtr)
+    {}
+
+    /** Empty destructor */
+    inline ~Map() {}
+};
+
+namespace internal {
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Map<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Map<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSE_MAP_H
diff --git a/Eigen/src/SparseCore/SparseMatrix.h b/Eigen/src/SparseCore/SparseMatrix.h
index ba5e3a9..0a2490b 100644
--- a/Eigen/src/SparseCore/SparseMatrix.h
+++ b/Eigen/src/SparseCore/SparseMatrix.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -32,18 +32,22 @@
   * \tparam _Scalar the scalar type, i.e. the type of the coefficients
   * \tparam _Options Union of bit flags controlling the storage scheme. Currently the only possibility
   *                 is ColMajor or RowMajor. The default is 0 which means column-major.
-  * \tparam _Index the type of the indices. It has to be a \b signed type (e.g., short, int, std::ptrdiff_t). Default is \c int.
+  * \tparam _StorageIndex the type of the indices. It has to be a \b signed type (e.g., short, int, std::ptrdiff_t). Default is \c int.
+  *
+  * \warning In %Eigen 3.2, the undocumented type \c SparseMatrix::Index was improperly defined as the storage index type (e.g., int),
+  *          whereas it is now (starting from %Eigen 3.3) deprecated and always defined as Eigen::Index.
+  *          Codes making use of \c SparseMatrix::Index, might thus likely have to be changed to use \c SparseMatrix::StorageIndex instead.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_SPARSEMATRIX_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_SPARSEMATRIX_PLUGIN.
   */
 
 namespace internal {
-template<typename _Scalar, int _Options, typename _Index>
-struct traits<SparseMatrix<_Scalar, _Options, _Index> >
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct traits<SparseMatrix<_Scalar, _Options, _StorageIndex> >
 {
   typedef _Scalar Scalar;
-  typedef _Index Index;
+  typedef _StorageIndex StorageIndex;
   typedef Sparse StorageKind;
   typedef MatrixXpr XprKind;
   enum {
@@ -51,22 +55,21 @@
     ColsAtCompileTime = Dynamic,
     MaxRowsAtCompileTime = Dynamic,
     MaxColsAtCompileTime = Dynamic,
-    Flags = _Options | NestByRefBit | LvalueBit,
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = _Options | NestByRefBit | LvalueBit | CompressedAccessBit,
     SupportedAccessPatterns = InnerRandomAccessPattern
   };
 };
 
-template<typename _Scalar, int _Options, typename _Index, int DiagIndex>
-struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _Index>, DiagIndex> >
+template<typename _Scalar, int _Options, typename _StorageIndex, int DiagIndex>
+struct traits<Diagonal<SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
 {
-  typedef SparseMatrix<_Scalar, _Options, _Index> MatrixType;
-  typedef typename nested<MatrixType>::type MatrixTypeNested;
+  typedef SparseMatrix<_Scalar, _Options, _StorageIndex> MatrixType;
+  typedef typename ref_selector<MatrixType>::type MatrixTypeNested;
   typedef typename remove_reference<MatrixTypeNested>::type _MatrixTypeNested;
 
   typedef _Scalar Scalar;
   typedef Dense StorageKind;
-  typedef _Index Index;
+  typedef _StorageIndex StorageIndex;
   typedef MatrixXpr XprKind;
 
   enum {
@@ -74,47 +77,61 @@
     ColsAtCompileTime = 1,
     MaxRowsAtCompileTime = Dynamic,
     MaxColsAtCompileTime = 1,
-    Flags = 0,
-    CoeffReadCost = _MatrixTypeNested::CoeffReadCost*10
+    Flags = LvalueBit
+  };
+};
+
+template<typename _Scalar, int _Options, typename _StorageIndex, int DiagIndex>
+struct traits<Diagonal<const SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
+ : public traits<Diagonal<SparseMatrix<_Scalar, _Options, _StorageIndex>, DiagIndex> >
+{
+  enum {
+    Flags = 0
   };
 };
 
 } // end namespace internal
 
-template<typename _Scalar, int _Options, typename _Index>
+template<typename _Scalar, int _Options, typename _StorageIndex>
 class SparseMatrix
-  : public SparseMatrixBase<SparseMatrix<_Scalar, _Options, _Index> >
+  : public SparseCompressedBase<SparseMatrix<_Scalar, _Options, _StorageIndex> >
 {
+    typedef SparseCompressedBase<SparseMatrix> Base;
+    using Base::convert_index;
+    friend class SparseVector<_Scalar,0,_StorageIndex>;
   public:
+    using Base::isCompressed;
+    using Base::nonZeros;
     EIGEN_SPARSE_PUBLIC_INTERFACE(SparseMatrix)
-    EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseMatrix, +=)
-    EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseMatrix, -=)
+    using Base::operator+=;
+    using Base::operator-=;
 
     typedef MappedSparseMatrix<Scalar,Flags> Map;
+    typedef Diagonal<SparseMatrix> DiagonalReturnType;
+    typedef Diagonal<const SparseMatrix> ConstDiagonalReturnType;
+    typedef typename Base::InnerIterator InnerIterator;
+    typedef typename Base::ReverseInnerIterator ReverseInnerIterator;
+    
+
     using Base::IsRowMajor;
-    typedef internal::CompressedStorage<Scalar,Index> Storage;
+    typedef internal::CompressedStorage<Scalar,StorageIndex> Storage;
     enum {
       Options = _Options
     };
 
+    typedef typename Base::IndexVector IndexVector;
+    typedef typename Base::ScalarVector ScalarVector;
   protected:
-
     typedef SparseMatrix<Scalar,(Flags&~RowMajorBit)|(IsRowMajor?RowMajorBit:0)> TransposedSparseMatrix;
 
     Index m_outerSize;
     Index m_innerSize;
-    Index* m_outerIndex;
-    Index* m_innerNonZeros;     // optional, if null then the data is compressed
+    StorageIndex* m_outerIndex;
+    StorageIndex* m_innerNonZeros;     // optional, if null then the data is compressed
     Storage m_data;
-    
-    Eigen::Map<Matrix<Index,Dynamic,1> > innerNonZeros() { return Eigen::Map<Matrix<Index,Dynamic,1> >(m_innerNonZeros, m_innerNonZeros?m_outerSize:0); }
-    const  Eigen::Map<const Matrix<Index,Dynamic,1> > innerNonZeros() const { return Eigen::Map<const Matrix<Index,Dynamic,1> >(m_innerNonZeros, m_innerNonZeros?m_outerSize:0); }
 
   public:
     
-    /** \returns whether \c *this is in compressed form. */
-    inline bool isCompressed() const { return m_innerNonZeros==0; }
-
     /** \returns the number of rows of the matrix */
     inline Index rows() const { return IsRowMajor ? m_outerSize : m_innerSize; }
     /** \returns the number of columns of the matrix */
@@ -128,38 +145,38 @@
     /** \returns a const pointer to the array of values.
       * This function is aimed at interoperability with other libraries.
       * \sa innerIndexPtr(), outerIndexPtr() */
-    inline const Scalar* valuePtr() const { return &m_data.value(0); }
+    inline const Scalar* valuePtr() const { return m_data.valuePtr(); }
     /** \returns a non-const pointer to the array of values.
       * This function is aimed at interoperability with other libraries.
       * \sa innerIndexPtr(), outerIndexPtr() */
-    inline Scalar* valuePtr() { return &m_data.value(0); }
+    inline Scalar* valuePtr() { return m_data.valuePtr(); }
 
     /** \returns a const pointer to the array of inner indices.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), outerIndexPtr() */
-    inline const Index* innerIndexPtr() const { return &m_data.index(0); }
+    inline const StorageIndex* innerIndexPtr() const { return m_data.indexPtr(); }
     /** \returns a non-const pointer to the array of inner indices.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), outerIndexPtr() */
-    inline Index* innerIndexPtr() { return &m_data.index(0); }
+    inline StorageIndex* innerIndexPtr() { return m_data.indexPtr(); }
 
     /** \returns a const pointer to the array of the starting positions of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), innerIndexPtr() */
-    inline const Index* outerIndexPtr() const { return m_outerIndex; }
+    inline const StorageIndex* outerIndexPtr() const { return m_outerIndex; }
     /** \returns a non-const pointer to the array of the starting positions of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \sa valuePtr(), innerIndexPtr() */
-    inline Index* outerIndexPtr() { return m_outerIndex; }
+    inline StorageIndex* outerIndexPtr() { return m_outerIndex; }
 
     /** \returns a const pointer to the array of the number of non zeros of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \warning it returns the null pointer 0 in compressed mode */
-    inline const Index* innerNonZeroPtr() const { return m_innerNonZeros; }
+    inline const StorageIndex* innerNonZeroPtr() const { return m_innerNonZeros; }
     /** \returns a non-const pointer to the array of the number of non zeros of the inner vectors.
       * This function is aimed at interoperability with other libraries.
       * \warning it returns the null pointer 0 in compressed mode */
-    inline Index* innerNonZeroPtr() { return m_innerNonZeros; }
+    inline StorageIndex* innerNonZeroPtr() { return m_innerNonZeros; }
 
     /** \internal */
     inline Storage& data() { return m_data; }
@@ -175,7 +192,7 @@
       const Index outer = IsRowMajor ? row : col;
       const Index inner = IsRowMajor ? col : row;
       Index end = m_innerNonZeros ? m_outerIndex[outer] + m_innerNonZeros[outer] : m_outerIndex[outer+1];
-      return m_data.atInRange(m_outerIndex[outer], end, inner);
+      return m_data.atInRange(m_outerIndex[outer], end, StorageIndex(inner));
     }
 
     /** \returns a non-const reference to the value of the matrix at position \a i, \a j
@@ -198,7 +215,7 @@
       eigen_assert(end>=start && "you probably called coeffRef on a non finalized matrix");
       if(end<=start)
         return insert(row,col);
-      const Index p = m_data.searchLowerIndex(start,end-1,inner);
+      const Index p = m_data.searchLowerIndex(start,end-1,StorageIndex(inner));
       if((p<end) && (m_data.index(p)==inner))
         return m_data.value(p);
       else
@@ -209,45 +226,34 @@
       * The non zero coefficient must \b not already exist.
       *
       * If the matrix \c *this is in compressed mode, then \c *this is turned into uncompressed
-      * mode while reserving room for 2 non zeros per inner vector. It is strongly recommended to first
-      * call reserve(const SizesType &) to reserve a more appropriate number of elements per
-      * inner vector that better match your scenario.
+      * mode while reserving room for 2 x this->innerSize() non zeros if reserve(Index) has not been called earlier.
+      * In this case, the insertion procedure is optimized for a \e sequential insertion mode where elements are assumed to be
+      * inserted by increasing outer-indices.
+      * 
+      * If that's not the case, then it is strongly recommended to either use a triplet-list to assemble the matrix, or to first
+      * call reserve(const SizesType &) to reserve the appropriate number of non-zero elements per inner vector.
       *
-      * This function performs a sorted insertion in O(1) if the elements of each inner vector are
-      * inserted in increasing inner index order, and in O(nnz_j) for a random insertion.
+      * Assuming memory has been appropriately reserved, this function performs a sorted insertion in O(1)
+      * if the elements of each inner vector are inserted in increasing inner index order, and in O(nnz_j) for a random insertion.
       *
       */
-    Scalar& insert(Index row, Index col)
-    {
-      eigen_assert(row>=0 && row<rows() && col>=0 && col<cols());
-      
-      if(isCompressed())
-      {
-        reserve(Matrix<Index,Dynamic,1>::Constant(outerSize(), 2));
-      }
-      return insertUncompressed(row,col);
-    }
+    Scalar& insert(Index row, Index col);
 
   public:
 
-    class InnerIterator;
-    class ReverseInnerIterator;
-
-    /** Removes all non zeros but keep allocated memory */
+    /** Removes all non zeros but keep allocated memory
+      *
+      * This function does not free the currently allocated memory. To release as much as memory as possible,
+      * call \code mat.data().squeeze(); \endcode after resizing it.
+      * 
+      * \sa resize(Index,Index), data()
+      */
     inline void setZero()
     {
       m_data.clear();
-      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(Index));
+      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(StorageIndex));
       if(m_innerNonZeros)
-        memset(m_innerNonZeros, 0, (m_outerSize)*sizeof(Index));
-    }
-
-    /** \returns the number of non zero coefficients */
-    inline Index nonZeros() const
-    {
-      if(m_innerNonZeros)
-        return innerNonZeros().sum();
-      return static_cast<Index>(m_data.size());
+        memset(m_innerNonZeros, 0, (m_outerSize)*sizeof(StorageIndex));
     }
 
     /** Preallocates \a reserveSize non zeros.
@@ -262,22 +268,25 @@
     #ifdef EIGEN_PARSED_BY_DOXYGEN
     /** Preallocates \a reserveSize[\c j] non zeros for each column (resp. row) \c j.
       *
-      * This function turns the matrix in non-compressed mode */
+      * This function turns the matrix in non-compressed mode.
+      * 
+      * The type \c SizesType must expose the following interface:
+        \code
+        typedef value_type;
+        const value_type& operator[](i) const;
+        \endcode
+      * for \c i in the [0,this->outerSize()[ range.
+      * Typical choices include std::vector<int>, Eigen::VectorXi, Eigen::VectorXi::Constant, etc.
+      */
     template<class SizesType>
     inline void reserve(const SizesType& reserveSizes);
     #else
     template<class SizesType>
-    inline void reserve(const SizesType& reserveSizes, const typename SizesType::value_type& enableif = typename SizesType::value_type())
-    {
-      EIGEN_UNUSED_VARIABLE(enableif);
-      reserveInnerVectors(reserveSizes);
-    }
-    template<class SizesType>
-    inline void reserve(const SizesType& reserveSizes, const typename SizesType::Scalar& enableif =
-    #if (!defined(_MSC_VER)) || (_MSC_VER>=1500) // MSVC 2005 fails to compile with this typename
+    inline void reserve(const SizesType& reserveSizes, const typename SizesType::value_type& enableif =
+    #if (!EIGEN_COMP_MSVC) || (EIGEN_COMP_MSVC>=1500) // MSVC 2005 fails to compile with this typename
         typename
     #endif
-        SizesType::Scalar())
+        SizesType::value_type())
     {
       EIGEN_UNUSED_VARIABLE(enableif);
       reserveInnerVectors(reserveSizes);
@@ -289,15 +298,15 @@
     {
       if(isCompressed())
       {
-        std::size_t totalReserveSize = 0;
+        Index totalReserveSize = 0;
         // turn the matrix into non-compressed mode
-        m_innerNonZeros = static_cast<Index*>(std::malloc(m_outerSize * sizeof(Index)));
+        m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
         if (!m_innerNonZeros) internal::throw_std_bad_alloc();
         
         // temporarily use m_innerSizes to hold the new starting points.
-        Index* newOuterIndex = m_innerNonZeros;
+        StorageIndex* newOuterIndex = m_innerNonZeros;
         
-        Index count = 0;
+        StorageIndex count = 0;
         for(Index j=0; j<m_outerSize; ++j)
         {
           newOuterIndex[j] = count;
@@ -305,10 +314,10 @@
           totalReserveSize += reserveSizes[j];
         }
         m_data.reserve(totalReserveSize);
-        Index previousOuterIndex = m_outerIndex[m_outerSize];
+        StorageIndex previousOuterIndex = m_outerIndex[m_outerSize];
         for(Index j=m_outerSize-1; j>=0; --j)
         {
-          Index innerNNZ = previousOuterIndex - m_outerIndex[j];
+          StorageIndex innerNNZ = previousOuterIndex - m_outerIndex[j];
           for(Index i=innerNNZ-1; i>=0; --i)
           {
             m_data.index(newOuterIndex[j]+i) = m_data.index(m_outerIndex[j]+i);
@@ -324,15 +333,15 @@
       }
       else
       {
-        Index* newOuterIndex = static_cast<Index*>(std::malloc((m_outerSize+1)*sizeof(Index)));
+        StorageIndex* newOuterIndex = static_cast<StorageIndex*>(std::malloc((m_outerSize+1)*sizeof(StorageIndex)));
         if (!newOuterIndex) internal::throw_std_bad_alloc();
         
-        Index count = 0;
+        StorageIndex count = 0;
         for(Index j=0; j<m_outerSize; ++j)
         {
           newOuterIndex[j] = count;
-          Index alreadyReserved = (m_outerIndex[j+1]-m_outerIndex[j]) - m_innerNonZeros[j];
-          Index toReserve = std::max<Index>(reserveSizes[j], alreadyReserved);
+          StorageIndex alreadyReserved = (m_outerIndex[j+1]-m_outerIndex[j]) - m_innerNonZeros[j];
+          StorageIndex toReserve = std::max<StorageIndex>(reserveSizes[j], alreadyReserved);
           count += toReserve + m_innerNonZeros[j];
         }
         newOuterIndex[m_outerSize] = count;
@@ -343,7 +352,7 @@
           Index offset = newOuterIndex[j] - m_outerIndex[j];
           if(offset>0)
           {
-            Index innerNNZ = m_innerNonZeros[j];
+            StorageIndex innerNNZ = m_innerNonZeros[j];
             for(Index i=innerNNZ-1; i>=0; --i)
             {
               m_data.index(newOuterIndex[j]+i) = m_data.index(m_outerIndex[j]+i);
@@ -380,11 +389,11 @@
       * \sa insertBack, startVec */
     inline Scalar& insertBackByOuterInner(Index outer, Index inner)
     {
-      eigen_assert(size_t(m_outerIndex[outer+1]) == m_data.size() && "Invalid ordered insertion (invalid outer index)");
+      eigen_assert(Index(m_outerIndex[outer+1]) == m_data.size() && "Invalid ordered insertion (invalid outer index)");
       eigen_assert( (m_outerIndex[outer+1]-m_outerIndex[outer]==0 || m_data.index(m_data.size()-1)<inner) && "Invalid ordered insertion (invalid inner index)");
       Index p = m_outerIndex[outer+1];
       ++m_outerIndex[outer+1];
-      m_data.append(0, inner);
+      m_data.append(Scalar(0), inner);
       return m_data.value(p);
     }
 
@@ -394,7 +403,7 @@
     {
       Index p = m_outerIndex[outer+1];
       ++m_outerIndex[outer+1];
-      m_data.append(0, inner);
+      m_data.append(Scalar(0), inner);
       return m_data.value(p);
     }
 
@@ -414,7 +423,7 @@
     {
       if(isCompressed())
       {
-        Index size = static_cast<Index>(m_data.size());
+        StorageIndex size = internal::convert_index<StorageIndex>(m_data.size());
         Index i = m_outerSize;
         // find the last filled column
         while (i>=0 && m_outerIndex[i]==0)
@@ -433,7 +442,13 @@
     template<typename InputIterators>
     void setFromTriplets(const InputIterators& begin, const InputIterators& end);
 
-    void sumupDuplicates();
+    template<typename InputIterators,typename DupFunctor>
+    void setFromTriplets(const InputIterators& begin, const InputIterators& end, DupFunctor dup_func);
+
+    void sumupDuplicates() { collapseDuplicates(internal::scalar_sum_op<Scalar,Scalar>()); }
+
+    template<typename DupFunctor>
+    void collapseDuplicates(DupFunctor dup_func = DupFunctor());
 
     //---
     
@@ -451,6 +466,8 @@
       if(isCompressed())
         return;
       
+      eigen_internal_assert(m_outerIndex!=0 && m_outerSize>0);
+      
       Index oldStart = m_outerIndex[1];
       m_outerIndex[1] = m_innerNonZeros[0];
       for(Index j=1; j<m_outerSize; ++j)
@@ -479,7 +496,7 @@
     {
       if(m_innerNonZeros != 0)
         return; 
-      m_innerNonZeros = static_cast<Index*>(std::malloc(m_outerSize * sizeof(Index)));
+      m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
       for (Index i = 0; i < m_outerSize; i++)
       {
         m_innerNonZeros[i] = m_outerIndex[i+1] - m_outerIndex[i]; 
@@ -503,10 +520,9 @@
     void prune(const KeepFunc& keep = KeepFunc())
     {
       // TODO optimize the uncompressed mode to avoid moving and allocating the data twice
-      // TODO also implement a unit test
       makeCompressed();
 
-      Index k = 0;
+      StorageIndex k = 0;
       for(Index j=0; j<m_outerSize; ++j)
       {
         Index previousStart = m_outerIndex[j];
@@ -527,7 +543,12 @@
     }
 
     /** Resizes the matrix to a \a rows x \a cols matrix leaving old values untouched.
-      * \sa resizeNonZeros(Index), reserve(), setZero()
+      *
+      * If the sizes of the matrix are decreased, then the matrix is turned to \b uncompressed-mode
+      * and the storage of the out of bounds coefficients is kept and reserved.
+      * Call makeCompressed() to pack the entries and squeeze extra memory.
+      *
+      * \sa reserve(), setZero(), makeCompressed()
       */
     void conservativeResize(Index rows, Index cols) 
     {
@@ -539,13 +560,13 @@
 
       Index innerChange = IsRowMajor ? cols - this->cols() : rows - this->rows();
       Index outerChange = IsRowMajor ? rows - this->rows() : cols - this->cols();
-      Index newInnerSize = IsRowMajor ? cols : rows;
+      StorageIndex newInnerSize = convert_index(IsRowMajor ? cols : rows);
 
       // Deals with inner non zeros
       if (m_innerNonZeros)
       {
         // Resize m_innerNonZeros
-        Index *newInnerNonZeros = static_cast<Index*>(std::realloc(m_innerNonZeros, (m_outerSize + outerChange) * sizeof(Index)));
+        StorageIndex *newInnerNonZeros = static_cast<StorageIndex*>(std::realloc(m_innerNonZeros, (m_outerSize + outerChange) * sizeof(StorageIndex)));
         if (!newInnerNonZeros) internal::throw_std_bad_alloc();
         m_innerNonZeros = newInnerNonZeros;
         
@@ -555,7 +576,7 @@
       else if (innerChange < 0) 
       {
         // Inner size decreased: allocate a new m_innerNonZeros
-        m_innerNonZeros = static_cast<Index*>(std::malloc((m_outerSize+outerChange+1) * sizeof(Index)));
+        m_innerNonZeros = static_cast<StorageIndex*>(std::malloc((m_outerSize+outerChange+1) * sizeof(StorageIndex)));
         if (!m_innerNonZeros) internal::throw_std_bad_alloc();
         for(Index i = 0; i < m_outerSize; i++)
           m_innerNonZeros[i] = m_outerIndex[i+1] - m_outerIndex[i];
@@ -566,8 +587,8 @@
       {
         for(Index i = 0; i < m_outerSize + (std::min)(outerChange, Index(0)); i++)
         {
-          Index &n = m_innerNonZeros[i];
-          Index start = m_outerIndex[i];
+          StorageIndex &n = m_innerNonZeros[i];
+          StorageIndex start = m_outerIndex[i];
           while (n > 0 && m_data.index(start+n-1) >= newInnerSize) --n; 
         }
       }
@@ -578,12 +599,12 @@
       if (outerChange == 0)
         return;
           
-      Index *newOuterIndex = static_cast<Index*>(std::realloc(m_outerIndex, (m_outerSize + outerChange + 1) * sizeof(Index)));
+      StorageIndex *newOuterIndex = static_cast<StorageIndex*>(std::realloc(m_outerIndex, (m_outerSize + outerChange + 1) * sizeof(StorageIndex)));
       if (!newOuterIndex) internal::throw_std_bad_alloc();
       m_outerIndex = newOuterIndex;
       if (outerChange > 0)
       {
-        Index last = m_outerSize == 0 ? 0 : m_outerIndex[m_outerSize];
+        StorageIndex last = m_outerSize == 0 ? 0 : m_outerIndex[m_outerSize];
         for(Index i=m_outerSize; i<m_outerSize+outerChange+1; i++)          
           m_outerIndex[i] = last; 
       }
@@ -591,7 +612,11 @@
     }
     
     /** Resizes the matrix to a \a rows x \a cols matrix and initializes it to zero.
-      * \sa resizeNonZeros(Index), reserve(), setZero()
+      * 
+      * This function does not free the currently allocated memory. To release as much as memory as possible,
+      * call \code mat.data().squeeze(); \endcode after resizing it.
+      * 
+      * \sa reserve(), setZero()
       */
     void resize(Index rows, Index cols)
     {
@@ -601,7 +626,7 @@
       if (m_outerSize != outerSize || m_outerSize==0)
       {
         std::free(m_outerIndex);
-        m_outerIndex = static_cast<Index*>(std::malloc((outerSize + 1) * sizeof(Index)));
+        m_outerIndex = static_cast<StorageIndex*>(std::malloc((outerSize + 1) * sizeof(StorageIndex)));
         if (!m_outerIndex) internal::throw_std_bad_alloc();
         
         m_outerSize = outerSize;
@@ -611,19 +636,24 @@
         std::free(m_innerNonZeros);
         m_innerNonZeros = 0;
       }
-      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(Index));
+      memset(m_outerIndex, 0, (m_outerSize+1)*sizeof(StorageIndex));
     }
 
     /** \internal
       * Resize the nonzero vector to \a size */
     void resizeNonZeros(Index size)
     {
-      // TODO remove this function
       m_data.resize(size);
     }
 
-    /** \returns a const expression of the diagonal coefficients */
-    const Diagonal<const SparseMatrix> diagonal() const { return *this; }
+    /** \returns a const expression of the diagonal coefficients. */
+    const ConstDiagonalReturnType diagonal() const { return ConstDiagonalReturnType(*this); }
+    
+    /** \returns a read-write expression of the diagonal coefficients.
+      * \warning If the diagonal entries are written, then all diagonal
+      * entries \b must already exist, otherwise an assertion will be raised.
+      */
+    DiagonalReturnType diagonal() { return DiagonalReturnType(*this); }
 
     /** Default constructor yielding an empty \c 0 \c x \c 0 matrix */
     inline SparseMatrix()
@@ -649,7 +679,16 @@
       EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
       check_template_parameters();
-      *this = other.derived();
+      const bool needToTranspose = (Flags & RowMajorBit) != (internal::evaluator<OtherDerived>::Flags & RowMajorBit);
+      if (needToTranspose)
+        *this = other.derived();
+      else
+      {
+        #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+          EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+        #endif
+        internal::call_assignment_no_alias(*this, other.derived());
+      }
     }
     
     /** Constructs a sparse matrix from the sparse selfadjoint view \a other */
@@ -658,7 +697,7 @@
       : m_outerSize(0), m_innerSize(0), m_outerIndex(0), m_innerNonZeros(0)
     {
       check_template_parameters();
-      *this = other;
+      Base::operator=(other);
     }
 
     /** Copy constructor (it performs a deep copy) */
@@ -678,6 +717,15 @@
       initAssignment(other);
       other.evalTo(*this);
     }
+    
+    /** \brief Copy constructor with in-place evaluation */
+    template<typename OtherDerived>
+    explicit SparseMatrix(const DiagonalBase<OtherDerived>& other)
+      : Base(), m_outerSize(0), m_innerSize(0), m_outerIndex(0), m_innerNonZeros(0)
+    {
+      check_template_parameters();
+      *this = other.derived();
+    }
 
     /** Swaps the content of two sparse matrices of the same type.
       * This is a fast operation that simply swaps the underlying pointers and parameters. */
@@ -691,14 +739,17 @@
       m_data.swap(other.m_data);
     }
 
-    /** Sets *this to the identity matrix */
+    /** Sets *this to the identity matrix.
+      * This function also turns the matrix into compressed mode, and drop any reserved memory. */
     inline void setIdentity()
     {
       eigen_assert(rows() == cols() && "ONLY FOR SQUARED MATRICES");
       this->m_data.resize(rows());
-      Eigen::Map<Matrix<Index, Dynamic, 1> >(&this->m_data.index(0), rows()).setLinSpaced(0, rows()-1);
-      Eigen::Map<Matrix<Scalar, Dynamic, 1> >(&this->m_data.value(0), rows()).setOnes();
-      Eigen::Map<Matrix<Index, Dynamic, 1> >(this->m_outerIndex, rows()+1).setLinSpaced(0, rows());
+      Eigen::Map<IndexVector>(this->m_data.indexPtr(), rows()).setLinSpaced(0, StorageIndex(rows()-1));
+      Eigen::Map<ScalarVector>(this->m_data.valuePtr(), rows()).setOnes();
+      Eigen::Map<IndexVector>(this->m_outerIndex, rows()+1).setLinSpaced(0, StorageIndex(rows()));
+      std::free(m_innerNonZeros);
+      m_innerNonZeros = 0;
     }
     inline SparseMatrix& operator=(const SparseMatrix& other)
     {
@@ -708,10 +759,13 @@
       }
       else if(this!=&other)
       {
+        #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+          EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+        #endif
         initAssignment(other);
         if(other.isCompressed())
         {
-          memcpy(m_outerIndex, other.m_outerIndex, (m_outerSize+1)*sizeof(Index));
+          internal::smart_copy(other.m_outerIndex, other.m_outerIndex + m_outerSize + 1, m_outerIndex);
           m_data = other.m_data;
         }
         else
@@ -722,22 +776,11 @@
       return *this;
     }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
-    template<typename Lhs, typename Rhs>
-    inline SparseMatrix& operator=(const SparseSparseProduct<Lhs,Rhs>& product)
-    { return Base::operator=(product); }
-    
-    template<typename OtherDerived>
-    inline SparseMatrix& operator=(const ReturnByValue<OtherDerived>& other)
-    {
-      initAssignment(other);
-      return Base::operator=(other.derived());
-    }
-    
+#ifndef EIGEN_PARSED_BY_DOXYGEN
     template<typename OtherDerived>
     inline SparseMatrix& operator=(const EigenBase<OtherDerived>& other)
     { return Base::operator=(other.derived()); }
-    #endif
+#endif // EIGEN_PARSED_BY_DOXYGEN
 
     template<typename OtherDerived>
     EIGEN_DONT_INLINE SparseMatrix& operator=(const SparseMatrixBase<OtherDerived>& other);
@@ -747,30 +790,38 @@
       EIGEN_DBG_SPARSE(
         s << "Nonzero entries:\n";
         if(m.isCompressed())
+        {
           for (Index i=0; i<m.nonZeros(); ++i)
             s << "(" << m.m_data.value(i) << "," << m.m_data.index(i) << ") ";
+        }
         else
+        {
           for (Index i=0; i<m.outerSize(); ++i)
           {
             Index p = m.m_outerIndex[i];
             Index pe = m.m_outerIndex[i]+m.m_innerNonZeros[i];
             Index k=p;
-            for (; k<pe; ++k)
+            for (; k<pe; ++k) {
               s << "(" << m.m_data.value(k) << "," << m.m_data.index(k) << ") ";
-            for (; k<m.m_outerIndex[i+1]; ++k)
+            }
+            for (; k<m.m_outerIndex[i+1]; ++k) {
               s << "(_,_) ";
+            }
           }
+        }
         s << std::endl;
         s << std::endl;
         s << "Outer pointers:\n";
-        for (Index i=0; i<m.outerSize(); ++i)
+        for (Index i=0; i<m.outerSize(); ++i) {
           s << m.m_outerIndex[i] << " ";
+        }
         s << " $" << std::endl;
         if(!m.isCompressed())
         {
           s << "Inner non zeros:\n";
-          for (Index i=0; i<m.outerSize(); ++i)
+          for (Index i=0; i<m.outerSize(); ++i) {
             s << m.m_innerNonZeros[i] << " ";
+          }
           s << " $" << std::endl;
         }
         s << std::endl;
@@ -786,10 +837,8 @@
       std::free(m_innerNonZeros);
     }
 
-#ifndef EIGEN_PARSED_BY_DOXYGEN
     /** Overloaded for performance */
     Scalar sum() const;
-#endif
     
 #   ifdef EIGEN_SPARSEMATRIX_PLUGIN
 #     include EIGEN_SPARSEMATRIX_PLUGIN
@@ -816,15 +865,15 @@
       * A vector object that is equal to 0 everywhere but v at the position i */
     class SingletonVector
     {
-        Index m_index;
-        Index m_value;
+        StorageIndex m_index;
+        StorageIndex m_value;
       public:
-        typedef Index value_type;
+        typedef StorageIndex value_type;
         SingletonVector(Index i, Index v)
-          : m_index(i), m_value(v)
+          : m_index(convert_index(i)), m_value(convert_index(v))
         {}
 
-        Index operator[](Index i) const { return i==m_index ? m_value : 0; }
+        StorageIndex operator[](Index i) const { return i==m_index ? m_value : 0; }
     };
 
     /** \internal
@@ -843,14 +892,14 @@
       eigen_assert(m_innerNonZeros[outer]<=(m_outerIndex[outer+1] - m_outerIndex[outer]));
 
       Index p = m_outerIndex[outer] + m_innerNonZeros[outer]++;
-      m_data.index(p) = inner;
-      return (m_data.value(p) = 0);
+      m_data.index(p) = convert_index(inner);
+      return (m_data.value(p) = Scalar(0));
     }
 
 private:
   static void check_template_parameters()
   {
-    EIGEN_STATIC_ASSERT(NumTraits<Index>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
+    EIGEN_STATIC_ASSERT(NumTraits<StorageIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
     EIGEN_STATIC_ASSERT((Options&(ColMajor|RowMajor))==Options,INVALID_MATRIX_TEMPLATE_PARAMETERS);
   }
 
@@ -865,87 +914,20 @@
   };
 };
 
-template<typename Scalar, int _Options, typename _Index>
-class SparseMatrix<Scalar,_Options,_Index>::InnerIterator
-{
-  public:
-    InnerIterator(const SparseMatrix& mat, Index outer)
-      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer), m_id(mat.m_outerIndex[outer])
-    {
-      if(mat.isCompressed())
-        m_end = mat.m_outerIndex[outer+1];
-      else
-        m_end = m_id + mat.m_innerNonZeros[outer];
-    }
-
-    inline InnerIterator& operator++() { m_id++; return *this; }
-
-    inline const Scalar& value() const { return m_values[m_id]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id]); }
-
-    inline Index index() const { return m_indices[m_id]; }
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id < m_end); }
-
-  protected:
-    const Scalar* m_values;
-    const Index* m_indices;
-    const Index m_outer;
-    Index m_id;
-    Index m_end;
-};
-
-template<typename Scalar, int _Options, typename _Index>
-class SparseMatrix<Scalar,_Options,_Index>::ReverseInnerIterator
-{
-  public:
-    ReverseInnerIterator(const SparseMatrix& mat, Index outer)
-      : m_values(mat.valuePtr()), m_indices(mat.innerIndexPtr()), m_outer(outer), m_start(mat.m_outerIndex[outer])
-    {
-      if(mat.isCompressed())
-        m_id = mat.m_outerIndex[outer+1];
-      else
-        m_id = m_start + mat.m_innerNonZeros[outer];
-    }
-
-    inline ReverseInnerIterator& operator--() { --m_id; return *this; }
-
-    inline const Scalar& value() const { return m_values[m_id-1]; }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_values[m_id-1]); }
-
-    inline Index index() const { return m_indices[m_id-1]; }
-    inline Index outer() const { return m_outer; }
-    inline Index row() const { return IsRowMajor ? m_outer : index(); }
-    inline Index col() const { return IsRowMajor ? index() : m_outer; }
-
-    inline operator bool() const { return (m_id > m_start); }
-
-  protected:
-    const Scalar* m_values;
-    const Index* m_indices;
-    const Index m_outer;
-    Index m_id;
-    const Index m_start;
-};
-
 namespace internal {
 
-template<typename InputIterator, typename SparseMatrixType>
-void set_from_triplets(const InputIterator& begin, const InputIterator& end, SparseMatrixType& mat, int Options = 0)
+template<typename InputIterator, typename SparseMatrixType, typename DupFunctor>
+void set_from_triplets(const InputIterator& begin, const InputIterator& end, SparseMatrixType& mat, DupFunctor dup_func)
 {
-  EIGEN_UNUSED_VARIABLE(Options);
   enum { IsRowMajor = SparseMatrixType::IsRowMajor };
   typedef typename SparseMatrixType::Scalar Scalar;
-  typedef typename SparseMatrixType::Index Index;
-  SparseMatrix<Scalar,IsRowMajor?ColMajor:RowMajor,Index> trMat(mat.rows(),mat.cols());
+  typedef typename SparseMatrixType::StorageIndex StorageIndex;
+  SparseMatrix<Scalar,IsRowMajor?ColMajor:RowMajor,StorageIndex> trMat(mat.rows(),mat.cols());
 
   if(begin!=end)
   {
     // pass 1: count the nnz per inner-vector
-    Matrix<Index,Dynamic,1> wi(trMat.outerSize());
+    typename SparseMatrixType::IndexVector wi(trMat.outerSize());
     wi.setZero();
     for(InputIterator it(begin); it!=end; ++it)
     {
@@ -959,7 +941,7 @@
       trMat.insertBackUncompressed(it->row(),it->col()) = it->value();
 
     // pass 3:
-    trMat.sumupDuplicates();
+    trMat.collapseDuplicates(dup_func);
   }
 
   // pass 4: transposed copy -> implicit sorting
@@ -1006,26 +988,43 @@
   * an abstract iterator over a complex data-structure that would be expensive to evaluate. The triplets should rather
   * be explicitely stored into a std::vector for instance.
   */
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename InputIterators>
-void SparseMatrix<Scalar,_Options,_Index>::setFromTriplets(const InputIterators& begin, const InputIterators& end)
+void SparseMatrix<Scalar,_Options,_StorageIndex>::setFromTriplets(const InputIterators& begin, const InputIterators& end)
 {
-  internal::set_from_triplets(begin, end, *this);
+  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_StorageIndex> >(begin, end, *this, internal::scalar_sum_op<Scalar,Scalar>());
+}
+
+/** The same as setFromTriplets but when duplicates are met the functor \a dup_func is applied:
+  * \code
+  * value = dup_func(OldValue, NewValue)
+  * \endcode 
+  * Here is a C++11 example keeping the latest entry only:
+  * \code
+  * mat.setFromTriplets(triplets.begin(), triplets.end(), [] (const Scalar&,const Scalar &b) { return b; });
+  * \endcode
+  */
+template<typename Scalar, int _Options, typename _StorageIndex>
+template<typename InputIterators,typename DupFunctor>
+void SparseMatrix<Scalar,_Options,_StorageIndex>::setFromTriplets(const InputIterators& begin, const InputIterators& end, DupFunctor dup_func)
+{
+  internal::set_from_triplets<InputIterators, SparseMatrix<Scalar,_Options,_StorageIndex>, DupFunctor>(begin, end, *this, dup_func);
 }
 
 /** \internal */
-template<typename Scalar, int _Options, typename _Index>
-void SparseMatrix<Scalar,_Options,_Index>::sumupDuplicates()
+template<typename Scalar, int _Options, typename _StorageIndex>
+template<typename DupFunctor>
+void SparseMatrix<Scalar,_Options,_StorageIndex>::collapseDuplicates(DupFunctor dup_func)
 {
   eigen_assert(!isCompressed());
   // TODO, in practice we should be able to use m_innerNonZeros for that task
-  Matrix<Index,Dynamic,1> wi(innerSize());
+  IndexVector wi(innerSize());
   wi.fill(-1);
-  Index count = 0;
+  StorageIndex count = 0;
   // for each inner-vector, wi[inner_index] will hold the position of first element into the index/value buffers
   for(Index j=0; j<outerSize(); ++j)
   {
-    Index start   = count;
+    StorageIndex start   = count;
     Index oldEnd  = m_outerIndex[j]+m_innerNonZeros[j];
     for(Index k=m_outerIndex[j]; k<oldEnd; ++k)
     {
@@ -1033,7 +1032,7 @@
       if(wi(i)>=start)
       {
         // we already meet this entry => accumulate it
-        m_data.value(wi(i)) += m_data.value(k);
+        m_data.value(wi(i)) = dup_func(m_data.value(wi(i)), m_data.value(k));
       }
       else
       {
@@ -1053,39 +1052,48 @@
   m_data.resize(m_outerIndex[m_outerSize]);
 }
 
-template<typename Scalar, int _Options, typename _Index>
+template<typename Scalar, int _Options, typename _StorageIndex>
 template<typename OtherDerived>
-EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_Index>& SparseMatrix<Scalar,_Options,_Index>::operator=(const SparseMatrixBase<OtherDerived>& other)
+EIGEN_DONT_INLINE SparseMatrix<Scalar,_Options,_StorageIndex>& SparseMatrix<Scalar,_Options,_StorageIndex>::operator=(const SparseMatrixBase<OtherDerived>& other)
 {
   EIGEN_STATIC_ASSERT((internal::is_same<Scalar, typename OtherDerived::Scalar>::value),
         YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
-  
-  const bool needToTranspose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
+
+  #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+    EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+  #endif
+      
+  const bool needToTranspose = (Flags & RowMajorBit) != (internal::evaluator<OtherDerived>::Flags & RowMajorBit);
   if (needToTranspose)
   {
+    #ifdef EIGEN_SPARSE_TRANSPOSED_COPY_PLUGIN
+      EIGEN_SPARSE_TRANSPOSED_COPY_PLUGIN
+    #endif
     // two passes algorithm:
     //  1 - compute the number of coeffs per dest inner vector
     //  2 - do the actual copy/eval
     // Since each coeff of the rhs has to be evaluated twice, let's evaluate it if needed
-    typedef typename internal::nested<OtherDerived,2>::type OtherCopy;
+    typedef typename internal::nested_eval<OtherDerived,2,typename internal::plain_matrix_type<OtherDerived>::type >::type OtherCopy;
     typedef typename internal::remove_all<OtherCopy>::type _OtherCopy;
+    typedef internal::evaluator<_OtherCopy> OtherCopyEval;
     OtherCopy otherCopy(other.derived());
+    OtherCopyEval otherCopyEval(otherCopy);
 
     SparseMatrix dest(other.rows(),other.cols());
-    Eigen::Map<Matrix<Index, Dynamic, 1> > (dest.m_outerIndex,dest.outerSize()).setZero();
+    Eigen::Map<IndexVector> (dest.m_outerIndex,dest.outerSize()).setZero();
 
     // pass 1
     // FIXME the above copy could be merged with that pass
     for (Index j=0; j<otherCopy.outerSize(); ++j)
-      for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
+      for (typename OtherCopyEval::InnerIterator it(otherCopyEval, j); it; ++it)
         ++dest.m_outerIndex[it.index()];
 
     // prefix sum
-    Index count = 0;
-    Matrix<Index,Dynamic,1> positions(dest.outerSize());
+    StorageIndex count = 0;
+    IndexVector positions(dest.outerSize());
     for (Index j=0; j<dest.outerSize(); ++j)
     {
-      Index tmp = dest.m_outerIndex[j];
+      StorageIndex tmp = dest.m_outerIndex[j];
       dest.m_outerIndex[j] = count;
       positions[j] = count;
       count += tmp;
@@ -1094,9 +1102,9 @@
     // alloc
     dest.m_data.resize(count);
     // pass 2
-    for (Index j=0; j<otherCopy.outerSize(); ++j)
+    for (StorageIndex j=0; j<otherCopy.outerSize(); ++j)
     {
-      for (typename _OtherCopy::InnerIterator it(otherCopy, j); it; ++it)
+      for (typename OtherCopyEval::InnerIterator it(otherCopyEval, j); it; ++it)
       {
         Index pos = positions[it.index()]++;
         dest.m_data.index(pos) = j;
@@ -1109,26 +1117,148 @@
   else
   {
     if(other.isRValue())
+    {
       initAssignment(other.derived());
+    }
     // there is no special optimization
     return Base::operator=(other.derived());
   }
 }
 
-template<typename _Scalar, int _Options, typename _Index>
-EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Options,_Index>::insertUncompressed(Index row, Index col)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insert(Index row, Index col)
+{
+  eigen_assert(row>=0 && row<rows() && col>=0 && col<cols());
+  
+  const Index outer = IsRowMajor ? row : col;
+  const Index inner = IsRowMajor ? col : row;
+  
+  if(isCompressed())
+  {
+    if(nonZeros()==0)
+    {
+      // reserve space if not already done
+      if(m_data.allocatedSize()==0)
+        m_data.reserve(2*m_innerSize);
+      
+      // turn the matrix into non-compressed mode
+      m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
+      if(!m_innerNonZeros) internal::throw_std_bad_alloc();
+      
+      memset(m_innerNonZeros, 0, (m_outerSize)*sizeof(StorageIndex));
+      
+      // pack all inner-vectors to the end of the pre-allocated space
+      // and allocate the entire free-space to the first inner-vector
+      StorageIndex end = convert_index(m_data.allocatedSize());
+      for(Index j=1; j<=m_outerSize; ++j)
+        m_outerIndex[j] = end;
+    }
+    else
+    {
+      // turn the matrix into non-compressed mode
+      m_innerNonZeros = static_cast<StorageIndex*>(std::malloc(m_outerSize * sizeof(StorageIndex)));
+      if(!m_innerNonZeros) internal::throw_std_bad_alloc();
+      for(Index j=0; j<m_outerSize; ++j)
+        m_innerNonZeros[j] = m_outerIndex[j+1]-m_outerIndex[j];
+    }
+  }
+  
+  // check whether we can do a fast "push back" insertion
+  Index data_end = m_data.allocatedSize();
+  
+  // First case: we are filling a new inner vector which is packed at the end.
+  // We assume that all remaining inner-vectors are also empty and packed to the end.
+  if(m_outerIndex[outer]==data_end)
+  {
+    eigen_internal_assert(m_innerNonZeros[outer]==0);
+    
+    // pack previous empty inner-vectors to end of the used-space
+    // and allocate the entire free-space to the current inner-vector.
+    StorageIndex p = convert_index(m_data.size());
+    Index j = outer;
+    while(j>=0 && m_innerNonZeros[j]==0)
+      m_outerIndex[j--] = p;
+    
+    // push back the new element
+    ++m_innerNonZeros[outer];
+    m_data.append(Scalar(0), inner);
+    
+    // check for reallocation
+    if(data_end != m_data.allocatedSize())
+    {
+      // m_data has been reallocated
+      //  -> move remaining inner-vectors back to the end of the free-space
+      //     so that the entire free-space is allocated to the current inner-vector.
+      eigen_internal_assert(data_end < m_data.allocatedSize());
+      StorageIndex new_end = convert_index(m_data.allocatedSize());
+      for(Index k=outer+1; k<=m_outerSize; ++k)
+        if(m_outerIndex[k]==data_end)
+          m_outerIndex[k] = new_end;
+    }
+    return m_data.value(p);
+  }
+  
+  // Second case: the next inner-vector is packed to the end
+  // and the current inner-vector end match the used-space.
+  if(m_outerIndex[outer+1]==data_end && m_outerIndex[outer]+m_innerNonZeros[outer]==m_data.size())
+  {
+    eigen_internal_assert(outer+1==m_outerSize || m_innerNonZeros[outer+1]==0);
+    
+    // add space for the new element
+    ++m_innerNonZeros[outer];
+    m_data.resize(m_data.size()+1);
+    
+    // check for reallocation
+    if(data_end != m_data.allocatedSize())
+    {
+      // m_data has been reallocated
+      //  -> move remaining inner-vectors back to the end of the free-space
+      //     so that the entire free-space is allocated to the current inner-vector.
+      eigen_internal_assert(data_end < m_data.allocatedSize());
+      StorageIndex new_end = convert_index(m_data.allocatedSize());
+      for(Index k=outer+1; k<=m_outerSize; ++k)
+        if(m_outerIndex[k]==data_end)
+          m_outerIndex[k] = new_end;
+    }
+    
+    // and insert it at the right position (sorted insertion)
+    Index startId = m_outerIndex[outer];
+    Index p = m_outerIndex[outer]+m_innerNonZeros[outer]-1;
+    while ( (p > startId) && (m_data.index(p-1) > inner) )
+    {
+      m_data.index(p) = m_data.index(p-1);
+      m_data.value(p) = m_data.value(p-1);
+      --p;
+    }
+    
+    m_data.index(p) = convert_index(inner);
+    return (m_data.value(p) = 0);
+  }
+  
+  if(m_data.size() != m_data.allocatedSize())
+  {
+    // make sure the matrix is compatible to random un-compressed insertion:
+    m_data.resize(m_data.allocatedSize());
+    this->reserveInnerVectors(Array<StorageIndex,Dynamic,1>::Constant(m_outerSize, 2));
+  }
+  
+  return insertUncompressed(row,col);
+}
+    
+template<typename _Scalar, int _Options, typename _StorageIndex>
+EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insertUncompressed(Index row, Index col)
 {
   eigen_assert(!isCompressed());
 
   const Index outer = IsRowMajor ? row : col;
-  const Index inner = IsRowMajor ? col : row;
+  const StorageIndex inner = convert_index(IsRowMajor ? col : row);
 
   Index room = m_outerIndex[outer+1] - m_outerIndex[outer];
-  Index innerNNZ = m_innerNonZeros[outer];
+  StorageIndex innerNNZ = m_innerNonZeros[outer];
   if(innerNNZ>=room)
   {
     // this inner vector is full, we need to reallocate the whole buffer :(
-    reserve(SingletonVector(outer,std::max<Index>(2,innerNNZ)));
+    reserve(SingletonVector(outer,std::max<StorageIndex>(2,innerNNZ)));
   }
 
   Index startId = m_outerIndex[outer];
@@ -1139,16 +1269,16 @@
     m_data.value(p) = m_data.value(p-1);
     --p;
   }
-  eigen_assert((p<=startId || m_data.index(p-1)!=inner) && "you cannot insert an element that already exist, you must call coeffRef to this end");
+  eigen_assert((p<=startId || m_data.index(p-1)!=inner) && "you cannot insert an element that already exists, you must call coeffRef to this end");
 
   m_innerNonZeros[outer]++;
 
   m_data.index(p) = inner;
-  return (m_data.value(p) = 0);
+  return (m_data.value(p) = Scalar(0));
 }
 
-template<typename _Scalar, int _Options, typename _Index>
-EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_Index>::Scalar& SparseMatrix<_Scalar,_Options,_Index>::insertCompressed(Index row, Index col)
+template<typename _Scalar, int _Options, typename _StorageIndex>
+EIGEN_DONT_INLINE typename SparseMatrix<_Scalar,_Options,_StorageIndex>::Scalar& SparseMatrix<_Scalar,_Options,_StorageIndex>::insertCompressed(Index row, Index col)
 {
   eigen_assert(isCompressed());
 
@@ -1161,7 +1291,7 @@
     // we start a new inner vector
     while (previousOuter>=0 && m_outerIndex[previousOuter]==0)
     {
-      m_outerIndex[previousOuter] = static_cast<Index>(m_data.size());
+      m_outerIndex[previousOuter] = convert_index(m_data.size());
       --previousOuter;
     }
     m_outerIndex[outer+1] = m_outerIndex[outer];
@@ -1171,11 +1301,11 @@
   // starts with: [ 0 0 0 0 0 1 ...] and we are inserted in, e.g.,
   // the 2nd inner vector...
   bool isLastVec = (!(previousOuter==-1 && m_data.size()!=0))
-                && (size_t(m_outerIndex[outer+1]) == m_data.size());
+                && (std::size_t(m_outerIndex[outer+1]) == m_data.size());
 
-  size_t startId = m_outerIndex[outer];
-  // FIXME let's make sure sizeof(long int) == sizeof(size_t)
-  size_t p = m_outerIndex[outer+1];
+  std::size_t startId = m_outerIndex[outer];
+  // FIXME let's make sure sizeof(long int) == sizeof(std::size_t)
+  std::size_t p = m_outerIndex[outer+1];
   ++m_outerIndex[outer+1];
 
   double reallocRatio = 1;
@@ -1251,7 +1381,21 @@
   }
 
   m_data.index(p) = inner;
-  return (m_data.value(p) = 0);
+  return (m_data.value(p) = Scalar(0));
+}
+
+namespace internal {
+
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct evaluator<SparseMatrix<_Scalar,_Options,_StorageIndex> >
+  : evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_StorageIndex> > >
+{
+  typedef evaluator<SparseCompressedBase<SparseMatrix<_Scalar,_Options,_StorageIndex> > > Base;
+  typedef SparseMatrix<_Scalar,_Options,_StorageIndex> SparseMatrixType;
+  evaluator() : Base() {}
+  explicit evaluator(const SparseMatrixType &mat) : Base(mat) {}
+};
+
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/SparseCore/SparseMatrixBase.h b/Eigen/src/SparseCore/SparseMatrixBase.h
index 6b2169a..c6b548f 100644
--- a/Eigen/src/SparseCore/SparseMatrixBase.h
+++ b/Eigen/src/SparseCore/SparseMatrixBase.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,32 +18,41 @@
   *
   * \brief Base class of any sparse matrices or sparse expressions
   *
-  * \tparam Derived
+  * \tparam Derived is the derived type, e.g. a sparse matrix type, or an expression, etc.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_SPARSEMATRIXBASE_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_SPARSEMATRIXBASE_PLUGIN.
   */
-template<typename Derived> class SparseMatrixBase : public EigenBase<Derived>
+template<typename Derived> class SparseMatrixBase
+  : public EigenBase<Derived>
 {
   public:
 
     typedef typename internal::traits<Derived>::Scalar Scalar;
+    
+    /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex<float>, etc.
+      *
+      * It is an alias for the Scalar type */
+    typedef Scalar value_type;
+    
     typedef typename internal::packet_traits<Scalar>::type PacketScalar;
     typedef typename internal::traits<Derived>::StorageKind StorageKind;
-    typedef typename internal::traits<Derived>::Index Index;
+
+    /** The integer type used to \b store indices within a SparseMatrix.
+      * For a \c SparseMatrix<Scalar,Options,IndexType> it an alias of the third template parameter \c IndexType. */
+    typedef typename internal::traits<Derived>::StorageIndex StorageIndex;
+
     typedef typename internal::add_const_on_value_type_if_arithmetic<
                          typename internal::packet_traits<Scalar>::type
                      >::type PacketReturnType;
 
     typedef SparseMatrixBase StorageBaseType;
-    typedef EigenBase<Derived> Base;
+
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector;
+    typedef Matrix<Scalar,Dynamic,1> ScalarVector;
     
     template<typename OtherDerived>
-    Derived& operator=(const EigenBase<OtherDerived> &other)
-    {
-      other.derived().evalTo(derived());
-      return derived();
-    }
+    Derived& operator=(const EigenBase<OtherDerived> &other);
 
     enum {
 
@@ -83,11 +92,6 @@
           * constructed from this one. See the \ref flags "list of flags".
           */
 
-      CoeffReadCost = internal::traits<Derived>::CoeffReadCost,
-        /**< This is a rough measure of how expensive it is to read one coefficient from
-          * this expression.
-          */
-
       IsRowMajor = Flags&RowMajorBit ? 1 : 0,
       
       InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime)
@@ -103,10 +107,11 @@
                         CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, Eigen::Transpose<const Derived> >,
                         Transpose<const Derived>
                      >::type AdjointReturnType;
+    typedef Transpose<Derived> TransposeReturnType;
+    typedef typename internal::add_const<Transpose<const Derived> >::type ConstTransposeReturnType;
 
-
-    typedef SparseMatrix<Scalar, Flags&RowMajorBit ? RowMajor : ColMajor, Index> PlainObject;
-
+    // FIXME storage order do not match evaluator storage order
+    typedef SparseMatrix<Scalar, Flags&RowMajorBit ? RowMajor : ColMajor, StorageIndex> PlainObject;
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** This is the "real scalar" type; if the \a Scalar type is already real numbers
@@ -124,6 +129,8 @@
     /** \internal Represents a matrix with all coefficients equal to one another*/
     typedef CwiseNullaryOp<internal::scalar_constant_op<Scalar>,Matrix<Scalar,Dynamic,Dynamic> > ConstantReturnType;
 
+    /** type of the equivalent dense matrix */
+    typedef Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> DenseMatrixType;
     /** type of the equivalent square matrix */
     typedef Matrix<Scalar,EIGEN_SIZE_MAX(RowsAtCompileTime,ColsAtCompileTime),
                           EIGEN_SIZE_MAX(RowsAtCompileTime,ColsAtCompileTime)> SquareMatrixType;
@@ -132,9 +139,21 @@
     inline Derived& derived() { return *static_cast<Derived*>(this); }
     inline Derived& const_cast_derived() const
     { return *static_cast<Derived*>(const_cast<SparseMatrixBase*>(this)); }
+
+    typedef EigenBase<Derived> Base;
+
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
 #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::SparseMatrixBase
+#ifdef EIGEN_PARSED_BY_DOXYGEN
+#define EIGEN_DOC_UNARY_ADDONS(METHOD,OP)           /** <p>This method does not change the sparsity of \c *this: the OP is applied to explicitly stored coefficients only. \sa SparseCompressedBase::coeffs() </p> */
+#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL      /** <p> \warning This method returns a read-only expression for any sparse matrices. \sa \ref TutorialSparse_SubMatrices "Sparse block operations" </p> */
+#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND) /** <p> \warning This method returns a read-write expression for COND sparse matrices only. Otherwise, the returned expression is read-only. \sa \ref TutorialSparse_SubMatrices "Sparse block operations" </p> */
+#else
+#define EIGEN_DOC_UNARY_ADDONS(X,Y)
+#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND)
+#endif
 #   include "../plugins/CommonCwiseUnaryOps.h"
 #   include "../plugins/CommonCwiseBinaryOps.h"
 #   include "../plugins/MatrixCwiseUnaryOps.h"
@@ -143,8 +162,10 @@
 #   ifdef EIGEN_SPARSEMATRIXBASE_PLUGIN
 #     include EIGEN_SPARSEMATRIXBASE_PLUGIN
 #   endif
-#   undef EIGEN_CURRENT_STORAGE_BASE_CLASS
 #undef EIGEN_CURRENT_STORAGE_BASE_CLASS
+#undef EIGEN_DOC_UNARY_ADDONS
+#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF
 
     /** \returns the number of rows. \sa cols() */
     inline Index rows() const { return derived().rows(); }
@@ -153,9 +174,6 @@
     /** \returns the number of coefficients, which is \a rows()*cols().
       * \sa rows(), cols(). */
     inline Index size() const { return rows() * cols(); }
-    /** \returns the number of nonzero coefficients which is in practice the number
-      * of stored coefficients. */
-    inline Index nonZeros() const { return derived().nonZeros(); }
     /** \returns true if either the number of rows or the number of columns is equal to 1.
       * In other words, this function returns
       * \code rows()==1 || cols()==1 \endcode
@@ -175,93 +193,23 @@
 
     
     template<typename OtherDerived>
-    Derived& operator=(const ReturnByValue<OtherDerived>& other)
-    {
-      other.evalTo(derived());
-      return derived();
-    }
-
+    Derived& operator=(const ReturnByValue<OtherDerived>& other);
 
     template<typename OtherDerived>
-    inline Derived& operator=(const SparseMatrixBase<OtherDerived>& other)
-    {
-      return assign(other.derived());
-    }
+    inline Derived& operator=(const SparseMatrixBase<OtherDerived>& other);
 
-    inline Derived& operator=(const Derived& other)
-    {
-//       if (other.isRValue())
-//         derived().swap(other.const_cast_derived());
-//       else
-      return assign(other.derived());
-    }
+    inline Derived& operator=(const Derived& other);
 
   protected:
 
     template<typename OtherDerived>
-    inline Derived& assign(const OtherDerived& other)
-    {
-      const bool transpose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
-      const Index outerSize = (int(OtherDerived::Flags) & RowMajorBit) ? other.rows() : other.cols();
-      if ((!transpose) && other.isRValue())
-      {
-        // eval without temporary
-        derived().resize(other.rows(), other.cols());
-        derived().setZero();
-        derived().reserve((std::max)(this->rows(),this->cols())*2);
-        for (Index j=0; j<outerSize; ++j)
-        {
-          derived().startVec(j);
-          for (typename OtherDerived::InnerIterator it(other, j); it; ++it)
-          {
-            Scalar v = it.value();
-            derived().insertBackByOuterInner(j,it.index()) = v;
-          }
-        }
-        derived().finalize();
-      }
-      else
-      {
-        assignGeneric(other);
-      }
-      return derived();
-    }
+    inline Derived& assign(const OtherDerived& other);
 
     template<typename OtherDerived>
-    inline void assignGeneric(const OtherDerived& other)
-    {
-      //const bool transpose = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit);
-      eigen_assert(( ((internal::traits<Derived>::SupportedAccessPatterns&OuterRandomAccessPattern)==OuterRandomAccessPattern) ||
-                  (!((Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit)))) &&
-                  "the transpose operation is supposed to be handled in SparseMatrix::operator=");
-
-      enum { Flip = (Flags & RowMajorBit) != (OtherDerived::Flags & RowMajorBit) };
-
-      const Index outerSize = other.outerSize();
-      //typedef typename internal::conditional<transpose, LinkedVectorMatrix<Scalar,Flags&RowMajorBit>, Derived>::type TempType;
-      // thanks to shallow copies, we always eval to a tempary
-      Derived temp(other.rows(), other.cols());
-
-      temp.reserve((std::max)(this->rows(),this->cols())*2);
-      for (Index j=0; j<outerSize; ++j)
-      {
-        temp.startVec(j);
-        for (typename OtherDerived::InnerIterator it(other.derived(), j); it; ++it)
-        {
-          Scalar v = it.value();
-          temp.insertBackByOuterInner(Flip?it.index():j,Flip?j:it.index()) = v;
-        }
-      }
-      temp.finalize();
-
-      derived() = temp.markAsRValue();
-    }
+    inline void assignGeneric(const OtherDerived& other);
 
   public:
 
-    template<typename Lhs, typename Rhs>
-    inline Derived& operator=(const SparseSparseProduct<Lhs,Rhs>& product);
-
     friend std::ostream & operator << (std::ostream & s, const SparseMatrixBase& m)
     {
       typedef typename Derived::Nested Nested;
@@ -269,11 +217,12 @@
 
       if (Flags&RowMajorBit)
       {
-        const Nested nm(m.derived());
+        Nested nm(m.derived());
+        internal::evaluator<NestedCleaned> thisEval(nm);
         for (Index row=0; row<nm.outerSize(); ++row)
         {
           Index col = 0;
-          for (typename NestedCleaned::InnerIterator it(nm.derived(), row); it; ++it)
+          for (typename internal::evaluator<NestedCleaned>::InnerIterator it(thisEval, row); it; ++it)
           {
             for ( ; col<it.index(); ++col)
               s << "0 ";
@@ -287,10 +236,11 @@
       }
       else
       {
-        const Nested nm(m.derived());
+        Nested nm(m.derived());
+        internal::evaluator<NestedCleaned> thisEval(nm);
         if (m.cols() == 1) {
           Index row = 0;
-          for (typename NestedCleaned::InnerIterator it(nm.derived(), 0); it; ++it)
+          for (typename internal::evaluator<NestedCleaned>::InnerIterator it(thisEval, 0); it; ++it)
           {
             for ( ; row<it.index(); ++row)
               s << "0" << std::endl;
@@ -302,8 +252,8 @@
         }
         else
         {
-          SparseMatrix<Scalar, RowMajorBit, Index> trans = m;
-          s << static_cast<const SparseMatrixBase<SparseMatrix<Scalar, RowMajorBit, Index> >&>(trans);
+          SparseMatrix<Scalar, RowMajorBit, StorageIndex> trans = m;
+          s << static_cast<const SparseMatrixBase<SparseMatrix<Scalar, RowMajorBit, StorageIndex> >&>(trans);
         }
       }
       return s;
@@ -313,56 +263,65 @@
     Derived& operator+=(const SparseMatrixBase<OtherDerived>& other);
     template<typename OtherDerived>
     Derived& operator-=(const SparseMatrixBase<OtherDerived>& other);
+    
+    template<typename OtherDerived>
+    Derived& operator+=(const DiagonalBase<OtherDerived>& other);
+    template<typename OtherDerived>
+    Derived& operator-=(const DiagonalBase<OtherDerived>& other);
+
+    template<typename OtherDerived>
+    Derived& operator+=(const EigenBase<OtherDerived> &other);
+    template<typename OtherDerived>
+    Derived& operator-=(const EigenBase<OtherDerived> &other);
 
     Derived& operator*=(const Scalar& other);
     Derived& operator/=(const Scalar& other);
 
-    #define EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE \
-      CwiseBinaryOp< \
-        internal::scalar_product_op< \
-          typename internal::scalar_product_traits< \
-            typename internal::traits<Derived>::Scalar, \
-            typename internal::traits<OtherDerived>::Scalar \
-          >::ReturnType \
-        >, \
-        const Derived, \
-        const OtherDerived \
-      >
+    template<typename OtherDerived> struct CwiseProductDenseReturnType {
+      typedef CwiseBinaryOp<internal::scalar_product_op<typename ScalarBinaryOpTraits<
+                                                          typename internal::traits<Derived>::Scalar,
+                                                          typename internal::traits<OtherDerived>::Scalar
+                                                        >::ReturnType>,
+                            const Derived,
+                            const OtherDerived
+                          > Type;
+    };
 
     template<typename OtherDerived>
-    EIGEN_STRONG_INLINE const EIGEN_SPARSE_CWISE_PRODUCT_RETURN_TYPE
+    EIGEN_STRONG_INLINE const typename CwiseProductDenseReturnType<OtherDerived>::Type
     cwiseProduct(const MatrixBase<OtherDerived> &other) const;
 
-    // sparse * sparse
-    template<typename OtherDerived>
-    const typename SparseSparseProductReturnType<Derived,OtherDerived>::Type
-    operator*(const SparseMatrixBase<OtherDerived> &other) const;
-
     // sparse * diagonal
     template<typename OtherDerived>
-    const SparseDiagonalProduct<Derived,OtherDerived>
-    operator*(const DiagonalBase<OtherDerived> &other) const;
+    const Product<Derived,OtherDerived>
+    operator*(const DiagonalBase<OtherDerived> &other) const
+    { return Product<Derived,OtherDerived>(derived(), other.derived()); }
 
     // diagonal * sparse
     template<typename OtherDerived> friend
-    const SparseDiagonalProduct<OtherDerived,Derived>
+    const Product<OtherDerived,Derived>
     operator*(const DiagonalBase<OtherDerived> &lhs, const SparseMatrixBase& rhs)
-    { return SparseDiagonalProduct<OtherDerived,Derived>(lhs.derived(), rhs.derived()); }
-
-    /** dense * sparse (return a dense object unless it is an outer product) */
-    template<typename OtherDerived> friend
-    const typename DenseSparseProductReturnType<OtherDerived,Derived>::Type
-    operator*(const MatrixBase<OtherDerived>& lhs, const Derived& rhs)
-    { return typename DenseSparseProductReturnType<OtherDerived,Derived>::Type(lhs.derived(),rhs); }
-
-    /** sparse * dense (returns a dense object unless it is an outer product) */
+    { return Product<OtherDerived,Derived>(lhs.derived(), rhs.derived()); }
+    
+    // sparse * sparse
     template<typename OtherDerived>
-    const typename SparseDenseProductReturnType<Derived,OtherDerived>::Type
+    const Product<Derived,OtherDerived,AliasFreeProduct>
+    operator*(const SparseMatrixBase<OtherDerived> &other) const;
+    
+    // sparse * dense
+    template<typename OtherDerived>
+    const Product<Derived,OtherDerived>
     operator*(const MatrixBase<OtherDerived> &other) const
-    { return typename SparseDenseProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived()); }
+    { return Product<Derived,OtherDerived>(derived(), other.derived()); }
+    
+    // dense * sparse
+    template<typename OtherDerived> friend
+    const Product<OtherDerived,Derived>
+    operator*(const MatrixBase<OtherDerived> &lhs, const SparseMatrixBase& rhs)
+    { return Product<OtherDerived,Derived>(lhs.derived(), rhs.derived()); }
     
      /** \returns an expression of P H P^-1 where H is the matrix represented by \c *this */
-    SparseSymmetricPermutationProduct<Derived,Upper|Lower> twistedBy(const PermutationMatrix<Dynamic,Dynamic,Index>& perm) const
+    SparseSymmetricPermutationProduct<Derived,Upper|Lower> twistedBy(const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& perm) const
     {
       return SparseSymmetricPermutationProduct<Derived,Upper|Lower>(derived(), perm);
     }
@@ -370,22 +329,16 @@
     template<typename OtherDerived>
     Derived& operator*=(const SparseMatrixBase<OtherDerived>& other);
 
-    #ifdef EIGEN2_SUPPORT
-    // deprecated
-    template<typename OtherDerived>
-    typename internal::plain_matrix_type_column_major<OtherDerived>::type
-    solveTriangular(const MatrixBase<OtherDerived>& other) const;
-
-    // deprecated
-    template<typename OtherDerived>
-    void solveTriangularInPlace(MatrixBase<OtherDerived>& other) const;
-    #endif // EIGEN2_SUPPORT
-
     template<int Mode>
-    inline const SparseTriangularView<Derived, Mode> triangularView() const;
+    inline const TriangularView<const Derived, Mode> triangularView() const;
+    
+    template<unsigned int UpLo> struct SelfAdjointViewReturnType { typedef SparseSelfAdjointView<Derived, UpLo> Type; };
+    template<unsigned int UpLo> struct ConstSelfAdjointViewReturnType { typedef const SparseSelfAdjointView<const Derived, UpLo> Type; };
 
-    template<unsigned int UpLo> inline const SparseSelfAdjointView<Derived, UpLo> selfadjointView() const;
-    template<unsigned int UpLo> inline SparseSelfAdjointView<Derived, UpLo> selfadjointView();
+    template<unsigned int UpLo> inline 
+    typename ConstSelfAdjointViewReturnType<UpLo>::Type selfadjointView() const;
+    template<unsigned int UpLo> inline
+    typename SelfAdjointViewReturnType<UpLo>::Type selfadjointView();
 
     template<typename OtherDerived> Scalar dot(const MatrixBase<OtherDerived>& other) const;
     template<typename OtherDerived> Scalar dot(const SparseMatrixBase<OtherDerived>& other) const;
@@ -393,9 +346,9 @@
     RealScalar norm()  const;
     RealScalar blueNorm() const;
 
-    Transpose<Derived> transpose() { return derived(); }
-    const Transpose<const Derived> transpose() const { return derived(); }
-    const AdjointReturnType adjoint() const { return transpose(); }
+    TransposeReturnType transpose() { return TransposeReturnType(derived()); }
+    const ConstTransposeReturnType transpose() const { return ConstTransposeReturnType(derived()); }
+    const AdjointReturnType adjoint() const { return AdjointReturnType(transpose()); }
 
     // inner-vector
     typedef Block<Derived,IsRowMajor?1:Dynamic,IsRowMajor?Dynamic:1,true>       InnerVectorReturnType;
@@ -409,25 +362,14 @@
     InnerVectorsReturnType innerVectors(Index outerStart, Index outerSize);
     const ConstInnerVectorsReturnType innerVectors(Index outerStart, Index outerSize) const;
 
-    /** \internal use operator= */
-    template<typename DenseDerived>
-    void evalTo(MatrixBase<DenseDerived>& dst) const
+    DenseMatrixType toDense() const
     {
-      dst.setZero();
-      for (Index j=0; j<outerSize(); ++j)
-        for (typename Derived::InnerIterator i(derived(),j); i; ++i)
-          dst.coeffRef(i.row(),i.col()) = i.value();
-    }
-
-    Matrix<Scalar,RowsAtCompileTime,ColsAtCompileTime> toDense() const
-    {
-      return derived();
+      return DenseMatrixType(derived());
     }
 
     template<typename OtherDerived>
     bool isApprox(const SparseMatrixBase<OtherDerived>& other,
-                  const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const
-    { return toDense().isApprox(other.toDense(),prec); }
+                  const RealScalar& prec = NumTraits<Scalar>::dummy_precision()) const;
 
     template<typename OtherDerived>
     bool isApprox(const MatrixBase<OtherDerived>& other,
@@ -443,10 +385,19 @@
     { return typename internal::eval<Derived>::type(derived()); }
 
     Scalar sum() const;
+    
+    inline const SparseView<Derived>
+    pruned(const Scalar& reference = Scalar(0), const RealScalar& epsilon = NumTraits<Scalar>::dummy_precision()) const;
 
   protected:
 
     bool m_isRValue;
+
+    static inline StorageIndex convert_index(const Index idx) {
+      return internal::convert_index<StorageIndex>(idx);
+    }
+  private:
+    template<typename Dest> void evalTo(Dest &) const;
 };
 
 } // end namespace Eigen
diff --git a/Eigen/src/SparseCore/SparsePermutation.h b/Eigen/src/SparseCore/SparsePermutation.h
index 75e2100..ef38357 100644
--- a/Eigen/src/SparseCore/SparsePermutation.h
+++ b/Eigen/src/SparseCore/SparsePermutation.h
@@ -16,131 +16,161 @@
 
 namespace internal {
 
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct traits<permut_sparsematrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
+template<typename ExpressionType, int Side, bool Transposed>
+struct permutation_matrix_product<ExpressionType, Side, Transposed, SparseShape>
 {
-  typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-  typedef typename MatrixTypeNestedCleaned::Scalar Scalar;
-  typedef typename MatrixTypeNestedCleaned::Index Index;
-  enum {
-    SrcStorageOrder = MatrixTypeNestedCleaned::Flags&RowMajorBit ? RowMajor : ColMajor,
-    MoveOuter = SrcStorageOrder==RowMajor ? Side==OnTheLeft : Side==OnTheRight
-  };
+    typedef typename nested_eval<ExpressionType, 1>::type MatrixType;
+    typedef typename remove_all<MatrixType>::type MatrixTypeCleaned;
 
-  typedef typename internal::conditional<MoveOuter,
-        SparseMatrix<Scalar,SrcStorageOrder,Index>,
-        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,Index> >::type ReturnType;
-};
-
-template<typename PermutationType, typename MatrixType, int Side, bool Transposed>
-struct permut_sparsematrix_product_retval
- : public ReturnByValue<permut_sparsematrix_product_retval<PermutationType, MatrixType, Side, Transposed> >
-{
-    typedef typename remove_all<typename MatrixType::Nested>::type MatrixTypeNestedCleaned;
-    typedef typename MatrixTypeNestedCleaned::Scalar Scalar;
-    typedef typename MatrixTypeNestedCleaned::Index Index;
+    typedef typename MatrixTypeCleaned::Scalar Scalar;
+    typedef typename MatrixTypeCleaned::StorageIndex StorageIndex;
 
     enum {
-      SrcStorageOrder = MatrixTypeNestedCleaned::Flags&RowMajorBit ? RowMajor : ColMajor,
+      SrcStorageOrder = MatrixTypeCleaned::Flags&RowMajorBit ? RowMajor : ColMajor,
       MoveOuter = SrcStorageOrder==RowMajor ? Side==OnTheLeft : Side==OnTheRight
     };
+    
+    typedef typename internal::conditional<MoveOuter,
+        SparseMatrix<Scalar,SrcStorageOrder,StorageIndex>,
+        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,StorageIndex> >::type ReturnType;
 
-    permut_sparsematrix_product_retval(const PermutationType& perm, const MatrixType& matrix)
-      : m_permutation(perm), m_matrix(matrix)
-    {}
-
-    inline int rows() const { return m_matrix.rows(); }
-    inline int cols() const { return m_matrix.cols(); }
-
-    template<typename Dest> inline void evalTo(Dest& dst) const
+    template<typename Dest,typename PermutationType>
+    static inline void run(Dest& dst, const PermutationType& perm, const ExpressionType& xpr)
     {
+      MatrixType mat(xpr);
       if(MoveOuter)
       {
-        SparseMatrix<Scalar,SrcStorageOrder,Index> tmp(m_matrix.rows(), m_matrix.cols());
-        Matrix<Index,Dynamic,1> sizes(m_matrix.outerSize());
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
+        SparseMatrix<Scalar,SrcStorageOrder,StorageIndex> tmp(mat.rows(), mat.cols());
+        Matrix<StorageIndex,Dynamic,1> sizes(mat.outerSize());
+        for(Index j=0; j<mat.outerSize(); ++j)
         {
-          Index jp = m_permutation.indices().coeff(j);
-          sizes[((Side==OnTheLeft) ^ Transposed) ? jp : j] = m_matrix.innerVector(((Side==OnTheRight) ^ Transposed) ? jp : j).nonZeros();
+          Index jp = perm.indices().coeff(j);
+          sizes[((Side==OnTheLeft) ^ Transposed) ? jp : j] = StorageIndex(mat.innerVector(((Side==OnTheRight) ^ Transposed) ? jp : j).nonZeros());
         }
         tmp.reserve(sizes);
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
+        for(Index j=0; j<mat.outerSize(); ++j)
         {
-          Index jp = m_permutation.indices().coeff(j);
+          Index jp = perm.indices().coeff(j);
           Index jsrc = ((Side==OnTheRight) ^ Transposed) ? jp : j;
           Index jdst = ((Side==OnTheLeft) ^ Transposed) ? jp : j;
-          for(typename MatrixTypeNestedCleaned::InnerIterator it(m_matrix,jsrc); it; ++it)
+          for(typename MatrixTypeCleaned::InnerIterator it(mat,jsrc); it; ++it)
             tmp.insertByOuterInner(jdst,it.index()) = it.value();
         }
         dst = tmp;
       }
       else
       {
-        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,Index> tmp(m_matrix.rows(), m_matrix.cols());
-        Matrix<Index,Dynamic,1> sizes(tmp.outerSize());
+        SparseMatrix<Scalar,int(SrcStorageOrder)==RowMajor?ColMajor:RowMajor,StorageIndex> tmp(mat.rows(), mat.cols());
+        Matrix<StorageIndex,Dynamic,1> sizes(tmp.outerSize());
         sizes.setZero();
-        PermutationMatrix<Dynamic,Dynamic,Index> perm;
+        PermutationMatrix<Dynamic,Dynamic,StorageIndex> perm_cpy;
         if((Side==OnTheLeft) ^ Transposed)
-          perm = m_permutation;
+          perm_cpy = perm;
         else
-          perm = m_permutation.transpose();
+          perm_cpy = perm.transpose();
 
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
-          for(typename MatrixTypeNestedCleaned::InnerIterator it(m_matrix,j); it; ++it)
-            sizes[perm.indices().coeff(it.index())]++;
+        for(Index j=0; j<mat.outerSize(); ++j)
+          for(typename MatrixTypeCleaned::InnerIterator it(mat,j); it; ++it)
+            sizes[perm_cpy.indices().coeff(it.index())]++;
         tmp.reserve(sizes);
-        for(Index j=0; j<m_matrix.outerSize(); ++j)
-          for(typename MatrixTypeNestedCleaned::InnerIterator it(m_matrix,j); it; ++it)
-            tmp.insertByOuterInner(perm.indices().coeff(it.index()),j) = it.value();
+        for(Index j=0; j<mat.outerSize(); ++j)
+          for(typename MatrixTypeCleaned::InnerIterator it(mat,j); it; ++it)
+            tmp.insertByOuterInner(perm_cpy.indices().coeff(it.index()),j) = it.value();
         dst = tmp;
       }
     }
-
-  protected:
-    const PermutationType& m_permutation;
-    typename MatrixType::Nested m_matrix;
 };
 
 }
 
+namespace internal {
 
+template <int ProductTag> struct product_promote_storage_type<Sparse,             PermutationStorage, ProductTag> { typedef Sparse ret; };
+template <int ProductTag> struct product_promote_storage_type<PermutationStorage, Sparse,             ProductTag> { typedef Sparse ret; };
+
+// TODO, the following two overloads are only needed to define the right temporary type through 
+// typename traits<permutation_sparse_matrix_product<Rhs,Lhs,OnTheRight,false> >::ReturnType
+// whereas it should be correctly handled by traits<Product<> >::PlainObject
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, AliasFreeProduct>, ProductTag, PermutationShape, SparseShape>
+  : public evaluator<typename permutation_matrix_product<Rhs,OnTheLeft,false,SparseShape>::ReturnType>
+{
+  typedef Product<Lhs, Rhs, AliasFreeProduct> XprType;
+  typedef typename permutation_matrix_product<Rhs,OnTheLeft,false,SparseShape>::ReturnType PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+
+  explicit product_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<Lhs, Rhs, PermutationShape, SparseShape, ProductTag>::evalTo(m_result, xpr.lhs(), xpr.rhs());
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+template<typename Lhs, typename Rhs, int ProductTag>
+struct product_evaluator<Product<Lhs, Rhs, AliasFreeProduct>, ProductTag, SparseShape, PermutationShape >
+  : public evaluator<typename permutation_matrix_product<Lhs,OnTheRight,false,SparseShape>::ReturnType>
+{
+  typedef Product<Lhs, Rhs, AliasFreeProduct> XprType;
+  typedef typename permutation_matrix_product<Lhs,OnTheRight,false,SparseShape>::ReturnType PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  enum {
+    Flags = Base::Flags | EvalBeforeNestingBit
+  };
+
+  explicit product_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<Lhs, Rhs, SparseShape, PermutationShape, ProductTag>::evalTo(m_result, xpr.lhs(), xpr.rhs());
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+} // end namespace internal
 
 /** \returns the matrix with the permutation applied to the columns
   */
 template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, false>
+inline const Product<SparseDerived, PermDerived, AliasFreeProduct>
 operator*(const SparseMatrixBase<SparseDerived>& matrix, const PermutationBase<PermDerived>& perm)
-{
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, false>(perm, matrix.derived());
-}
+{ return Product<SparseDerived, PermDerived, AliasFreeProduct>(matrix.derived(), perm.derived()); }
 
 /** \returns the matrix with the permutation applied to the rows
   */
 template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, false>
+inline const Product<PermDerived, SparseDerived, AliasFreeProduct>
 operator*( const PermutationBase<PermDerived>& perm, const SparseMatrixBase<SparseDerived>& matrix)
-{
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, false>(perm, matrix.derived());
-}
-
+{ return  Product<PermDerived, SparseDerived, AliasFreeProduct>(perm.derived(), matrix.derived()); }
 
 
 /** \returns the matrix with the inverse permutation applied to the columns.
   */
-template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, true>
-operator*(const SparseMatrixBase<SparseDerived>& matrix, const Transpose<PermutationBase<PermDerived> >& tperm)
+template<typename SparseDerived, typename PermutationType>
+inline const Product<SparseDerived, Inverse<PermutationType>, AliasFreeProduct>
+operator*(const SparseMatrixBase<SparseDerived>& matrix, const InverseImpl<PermutationType, PermutationStorage>& tperm)
 {
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheRight, true>(tperm.nestedPermutation(), matrix.derived());
+  return Product<SparseDerived, Inverse<PermutationType>, AliasFreeProduct>(matrix.derived(), tperm.derived());
 }
 
 /** \returns the matrix with the inverse permutation applied to the rows.
   */
-template<typename SparseDerived, typename PermDerived>
-inline const internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, true>
-operator*(const Transpose<PermutationBase<PermDerived> >& tperm, const SparseMatrixBase<SparseDerived>& matrix)
+template<typename SparseDerived, typename PermutationType>
+inline const Product<Inverse<PermutationType>, SparseDerived, AliasFreeProduct>
+operator*(const InverseImpl<PermutationType,PermutationStorage>& tperm, const SparseMatrixBase<SparseDerived>& matrix)
 {
-  return internal::permut_sparsematrix_product_retval<PermutationBase<PermDerived>, SparseDerived, OnTheLeft, true>(tperm.nestedPermutation(), matrix.derived());
+  return Product<Inverse<PermutationType>, SparseDerived, AliasFreeProduct>(tperm.derived(), matrix.derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/SparseCore/SparseProduct.h b/Eigen/src/SparseCore/SparseProduct.h
index cf76630..4cbf687 100644
--- a/Eigen/src/SparseCore/SparseProduct.h
+++ b/Eigen/src/SparseCore/SparseProduct.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2010 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,158 +12,6 @@
 
 namespace Eigen { 
 
-template<typename Lhs, typename Rhs>
-struct SparseSparseProductReturnType
-{
-  typedef typename internal::traits<Lhs>::Scalar Scalar;
-  typedef typename internal::traits<Lhs>::Index Index;
-  enum {
-    LhsRowMajor = internal::traits<Lhs>::Flags & RowMajorBit,
-    RhsRowMajor = internal::traits<Rhs>::Flags & RowMajorBit,
-    TransposeRhs = (!LhsRowMajor) && RhsRowMajor,
-    TransposeLhs = LhsRowMajor && (!RhsRowMajor)
-  };
-
-  typedef typename internal::conditional<TransposeLhs,
-    SparseMatrix<Scalar,0,Index>,
-    typename internal::nested<Lhs,Rhs::RowsAtCompileTime>::type>::type LhsNested;
-
-  typedef typename internal::conditional<TransposeRhs,
-    SparseMatrix<Scalar,0,Index>,
-    typename internal::nested<Rhs,Lhs::RowsAtCompileTime>::type>::type RhsNested;
-
-  typedef SparseSparseProduct<LhsNested, RhsNested> Type;
-};
-
-namespace internal {
-template<typename LhsNested, typename RhsNested>
-struct traits<SparseSparseProduct<LhsNested, RhsNested> >
-{
-  typedef MatrixXpr XprKind;
-  // clean the nested types:
-  typedef typename remove_all<LhsNested>::type _LhsNested;
-  typedef typename remove_all<RhsNested>::type _RhsNested;
-  typedef typename _LhsNested::Scalar Scalar;
-  typedef typename promote_index_type<typename traits<_LhsNested>::Index,
-                                         typename traits<_RhsNested>::Index>::type Index;
-
-  enum {
-    LhsCoeffReadCost = _LhsNested::CoeffReadCost,
-    RhsCoeffReadCost = _RhsNested::CoeffReadCost,
-    LhsFlags = _LhsNested::Flags,
-    RhsFlags = _RhsNested::Flags,
-
-    RowsAtCompileTime    = _LhsNested::RowsAtCompileTime,
-    ColsAtCompileTime    = _RhsNested::ColsAtCompileTime,
-    MaxRowsAtCompileTime = _LhsNested::MaxRowsAtCompileTime,
-    MaxColsAtCompileTime = _RhsNested::MaxColsAtCompileTime,
-
-    InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(_LhsNested::ColsAtCompileTime, _RhsNested::RowsAtCompileTime),
-
-    EvalToRowMajor = (RhsFlags & LhsFlags & RowMajorBit),
-
-    RemovedBits = ~(EvalToRowMajor ? 0 : RowMajorBit),
-
-    Flags = (int(LhsFlags | RhsFlags) & HereditaryBits & RemovedBits)
-          | EvalBeforeAssigningBit
-          | EvalBeforeNestingBit,
-
-    CoeffReadCost = Dynamic
-  };
-
-  typedef Sparse StorageKind;
-};
-
-} // end namespace internal
-
-template<typename LhsNested, typename RhsNested>
-class SparseSparseProduct : internal::no_assignment_operator,
-  public SparseMatrixBase<SparseSparseProduct<LhsNested, RhsNested> >
-{
-  public:
-
-    typedef SparseMatrixBase<SparseSparseProduct> Base;
-    EIGEN_DENSE_PUBLIC_INTERFACE(SparseSparseProduct)
-
-  private:
-
-    typedef typename internal::traits<SparseSparseProduct>::_LhsNested _LhsNested;
-    typedef typename internal::traits<SparseSparseProduct>::_RhsNested _RhsNested;
-
-  public:
-
-    template<typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE SparseSparseProduct(const Lhs& lhs, const Rhs& rhs)
-      : m_lhs(lhs), m_rhs(rhs), m_tolerance(0), m_conservative(true)
-    {
-      init();
-    }
-
-    template<typename Lhs, typename Rhs>
-    EIGEN_STRONG_INLINE SparseSparseProduct(const Lhs& lhs, const Rhs& rhs, const RealScalar& tolerance)
-      : m_lhs(lhs), m_rhs(rhs), m_tolerance(tolerance), m_conservative(false)
-    {
-      init();
-    }
-
-    SparseSparseProduct pruned(const Scalar& reference = 0, const RealScalar& epsilon = NumTraits<RealScalar>::dummy_precision()) const
-    {
-      using std::abs;
-      return SparseSparseProduct(m_lhs,m_rhs,abs(reference)*epsilon);
-    }
-
-    template<typename Dest>
-    void evalTo(Dest& result) const
-    {
-      if(m_conservative)
-        internal::conservative_sparse_sparse_product_selector<_LhsNested, _RhsNested, Dest>::run(lhs(),rhs(),result);
-      else
-        internal::sparse_sparse_product_with_pruning_selector<_LhsNested, _RhsNested, Dest>::run(lhs(),rhs(),result,m_tolerance);
-    }
-
-    EIGEN_STRONG_INLINE Index rows() const { return m_lhs.rows(); }
-    EIGEN_STRONG_INLINE Index cols() const { return m_rhs.cols(); }
-
-    EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; }
-    EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; }
-
-  protected:
-    void init()
-    {
-      eigen_assert(m_lhs.cols() == m_rhs.rows());
-
-      enum {
-        ProductIsValid = _LhsNested::ColsAtCompileTime==Dynamic
-                      || _RhsNested::RowsAtCompileTime==Dynamic
-                      || int(_LhsNested::ColsAtCompileTime)==int(_RhsNested::RowsAtCompileTime),
-        AreVectors = _LhsNested::IsVectorAtCompileTime && _RhsNested::IsVectorAtCompileTime,
-        SameSizes = EIGEN_PREDICATE_SAME_MATRIX_SIZE(_LhsNested,_RhsNested)
-      };
-      // note to the lost user:
-      //    * for a dot product use: v1.dot(v2)
-      //    * for a coeff-wise product use: v1.cwise()*v2
-      EIGEN_STATIC_ASSERT(ProductIsValid || !(AreVectors && SameSizes),
-        INVALID_VECTOR_VECTOR_PRODUCT__IF_YOU_WANTED_A_DOT_OR_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTIONS)
-      EIGEN_STATIC_ASSERT(ProductIsValid || !(SameSizes && !AreVectors),
-        INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION)
-      EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT)
-    }
-
-    LhsNested m_lhs;
-    RhsNested m_rhs;
-    RealScalar m_tolerance;
-    bool m_conservative;
-};
-
-// sparse = sparse * sparse
-template<typename Derived>
-template<typename Lhs, typename Rhs>
-inline Derived& SparseMatrixBase<Derived>::operator=(const SparseSparseProduct<Lhs,Rhs>& product)
-{
-  product.evalTo(derived());
-  return derived();
-}
-
 /** \returns an expression of the product of two sparse matrices.
   * By default a conservative product preserving the symbolic non zeros is performed.
   * The automatic pruning of the small values can be achieved by calling the pruned() function
@@ -177,12 +25,145 @@
   * */
 template<typename Derived>
 template<typename OtherDerived>
-inline const typename SparseSparseProductReturnType<Derived,OtherDerived>::Type
+inline const Product<Derived,OtherDerived,AliasFreeProduct>
 SparseMatrixBase<Derived>::operator*(const SparseMatrixBase<OtherDerived> &other) const
 {
-  return typename SparseSparseProductReturnType<Derived,OtherDerived>::Type(derived(), other.derived());
+  return Product<Derived,OtherDerived,AliasFreeProduct>(derived(), other.derived());
 }
 
+namespace internal {
+
+// sparse * sparse
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseShape, SparseShape, ProductType>
+{
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs)
+  {
+    evalTo(dst, lhs, rhs, typename evaluator_traits<Dest>::Shape());
+  }
+
+  // dense += sparse * sparse
+  template<typename Dest,typename ActualLhs>
+  static void addTo(Dest& dst, const ActualLhs& lhs, const Rhs& rhs, typename enable_if<is_same<typename evaluator_traits<Dest>::Shape,DenseShape>::value,int*>::type* = 0)
+  {
+    typedef typename nested_eval<ActualLhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    internal::sparse_sparse_to_dense_product_selector<typename remove_all<LhsNested>::type,
+                                                      typename remove_all<RhsNested>::type, Dest>::run(lhsNested,rhsNested,dst);
+  }
+
+  // dense -= sparse * sparse
+  template<typename Dest>
+  static void subTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, typename enable_if<is_same<typename evaluator_traits<Dest>::Shape,DenseShape>::value,int*>::type* = 0)
+  {
+    addTo(dst, -lhs, rhs);
+  }
+
+protected:
+
+  // sparse = sparse * sparse
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, SparseShape)
+  {
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhs);
+    internal::conservative_sparse_sparse_product_selector<typename remove_all<LhsNested>::type,
+                                                          typename remove_all<RhsNested>::type, Dest>::run(lhsNested,rhsNested,dst);
+  }
+
+  // dense = sparse * sparse
+  template<typename Dest>
+  static void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, DenseShape)
+  {
+    dst.setZero();
+    addTo(dst, lhs, rhs);
+  }
+};
+
+// sparse * sparse-triangular
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseShape, SparseTriangularShape, ProductType>
+ : public generic_product_impl<Lhs, Rhs, SparseShape, SparseShape, ProductType>
+{};
+
+// sparse-triangular * sparse
+template<typename Lhs, typename Rhs, int ProductType>
+struct generic_product_impl<Lhs, Rhs, SparseTriangularShape, SparseShape, ProductType>
+ : public generic_product_impl<Lhs, Rhs, SparseShape, SparseShape, ProductType>
+{};
+
+// dense = sparse-product (can be sparse*sparse, sparse*perm, etc.)
+template< typename DstXprType, typename Lhs, typename Rhs>
+struct Assignment<DstXprType, Product<Lhs,Rhs,AliasFreeProduct>, internal::assign_op<typename DstXprType::Scalar,typename Product<Lhs,Rhs,AliasFreeProduct>::Scalar>, Sparse2Dense>
+{
+  typedef Product<Lhs,Rhs,AliasFreeProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &)
+  {
+    Index dstRows = src.rows();
+    Index dstCols = src.cols();
+    if((dst.rows()!=dstRows) || (dst.cols()!=dstCols))
+      dst.resize(dstRows, dstCols);
+    
+    generic_product_impl<Lhs, Rhs>::evalTo(dst,src.lhs(),src.rhs());
+  }
+};
+
+// dense += sparse-product (can be sparse*sparse, sparse*perm, etc.)
+template< typename DstXprType, typename Lhs, typename Rhs>
+struct Assignment<DstXprType, Product<Lhs,Rhs,AliasFreeProduct>, internal::add_assign_op<typename DstXprType::Scalar,typename Product<Lhs,Rhs,AliasFreeProduct>::Scalar>, Sparse2Dense>
+{
+  typedef Product<Lhs,Rhs,AliasFreeProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &)
+  {
+    generic_product_impl<Lhs, Rhs>::addTo(dst,src.lhs(),src.rhs());
+  }
+};
+
+// dense -= sparse-product (can be sparse*sparse, sparse*perm, etc.)
+template< typename DstXprType, typename Lhs, typename Rhs>
+struct Assignment<DstXprType, Product<Lhs,Rhs,AliasFreeProduct>, internal::sub_assign_op<typename DstXprType::Scalar,typename Product<Lhs,Rhs,AliasFreeProduct>::Scalar>, Sparse2Dense>
+{
+  typedef Product<Lhs,Rhs,AliasFreeProduct> SrcXprType;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> &)
+  {
+    generic_product_impl<Lhs, Rhs>::subTo(dst,src.lhs(),src.rhs());
+  }
+};
+
+template<typename Lhs, typename Rhs, int Options>
+struct unary_evaluator<SparseView<Product<Lhs, Rhs, Options> >, IteratorBased>
+ : public evaluator<typename Product<Lhs, Rhs, DefaultProduct>::PlainObject>
+{
+  typedef SparseView<Product<Lhs, Rhs, Options> > XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  explicit unary_evaluator(const XprType& xpr)
+    : m_result(xpr.rows(), xpr.cols())
+  {
+    using std::abs;
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(xpr.nestedExpression().lhs());
+    RhsNested rhsNested(xpr.nestedExpression().rhs());
+
+    internal::sparse_sparse_product_with_pruning_selector<typename remove_all<LhsNested>::type,
+                                                          typename remove_all<RhsNested>::type, PlainObject>::run(lhsNested,rhsNested,m_result,
+                                                                                                                  abs(xpr.reference())*xpr.epsilon());
+  }
+
+protected:
+  PlainObject m_result;
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSEPRODUCT_H
diff --git a/Eigen/src/SparseCore/SparseRedux.h b/Eigen/src/SparseCore/SparseRedux.h
index f3da93a..4587749 100644
--- a/Eigen/src/SparseCore/SparseRedux.h
+++ b/Eigen/src/SparseCore/SparseRedux.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -18,8 +18,9 @@
 {
   eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix");
   Scalar res(0);
+  internal::evaluator<Derived> thisEval(derived());
   for (Index j=0; j<outerSize(); ++j)
-    for (typename Derived::InnerIterator iter(derived(),j); iter; ++iter)
+    for (typename internal::evaluator<Derived>::InnerIterator iter(thisEval,j); iter; ++iter)
       res += iter.value();
   return res;
 }
@@ -29,7 +30,10 @@
 SparseMatrix<_Scalar,_Options,_Index>::sum() const
 {
   eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix");
-  return Matrix<Scalar,1,Dynamic>::Map(&m_data.value(0), m_data.size()).sum();
+  if(this->isCompressed())
+    return Matrix<Scalar,1,Dynamic>::Map(m_data.valuePtr(), m_data.size()).sum();
+  else
+    return Base::sum();
 }
 
 template<typename _Scalar, int _Options, typename _Index>
@@ -37,7 +41,7 @@
 SparseVector<_Scalar,_Options,_Index>::sum() const
 {
   eigen_assert(rows()>0 && cols()>0 && "you are using a non initialized matrix");
-  return Matrix<Scalar,1,Dynamic>::Map(&m_data.value(0), m_data.size()).sum();
+  return Matrix<Scalar,1,Dynamic>::Map(m_data.valuePtr(), m_data.size()).sum();
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/SparseCore/SparseRef.h b/Eigen/src/SparseCore/SparseRef.h
new file mode 100644
index 0000000..d91f38f
--- /dev/null
+++ b/Eigen/src/SparseCore/SparseRef.h
@@ -0,0 +1,397 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2015 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSE_REF_H
+#define EIGEN_SPARSE_REF_H
+
+namespace Eigen {
+
+enum {
+  StandardCompressedFormat = 2 /**< used by Ref<SparseMatrix> to specify whether the input storage must be in standard compressed form */
+};
+  
+namespace internal {
+
+template<typename Derived> class SparseRefBase;
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  enum {
+    Options = _Options,
+    Flags = traits<PlainObjectType>::Flags | CompressedAccessBit | NestByRefBit
+  };
+
+  template<typename Derived> struct match {
+    enum {
+      StorageOrderMatch = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)),
+      MatchAtCompileTime = (Derived::Flags&CompressedAccessBit) && StorageOrderMatch
+    };
+    typedef typename internal::conditional<MatchAtCompileTime,internal::true_type,internal::false_type>::type type;
+  };
+  
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+{
+  enum {
+    Flags = (traits<SparseMatrix<MatScalar,MatOptions,MatIndex> >::Flags | CompressedAccessBit | NestByRefBit) & ~LvalueBit
+  };
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<SparseVector<MatScalar,MatOptions,MatIndex> >
+{
+  typedef SparseVector<MatScalar,MatOptions,MatIndex> PlainObjectType;
+  enum {
+    Options = _Options,
+    Flags = traits<PlainObjectType>::Flags | CompressedAccessBit | NestByRefBit
+  };
+
+  template<typename Derived> struct match {
+    enum {
+      MatchAtCompileTime = (Derived::Flags&CompressedAccessBit) && Derived::IsVectorAtCompileTime
+    };
+    typedef typename internal::conditional<MatchAtCompileTime,internal::true_type,internal::false_type>::type type;
+  };
+
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int _Options, typename _StrideType>
+struct traits<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+  : public traits<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, _Options, _StrideType> >
+{
+  enum {
+    Flags = (traits<SparseVector<MatScalar,MatOptions,MatIndex> >::Flags | CompressedAccessBit | NestByRefBit) & ~LvalueBit
+  };
+};
+
+template<typename Derived>
+struct traits<SparseRefBase<Derived> > : public traits<Derived> {};
+
+template<typename Derived> class SparseRefBase
+  : public SparseMapBase<Derived>
+{
+public:
+
+  typedef SparseMapBase<Derived> Base;
+  EIGEN_SPARSE_PUBLIC_INTERFACE(SparseRefBase)
+
+  SparseRefBase()
+    : Base(RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime, 0, 0, 0, 0, 0)
+  {}
+  
+protected:
+
+  template<typename Expression>
+  void construct(Expression& expr)
+  {
+    if(expr.outerIndexPtr()==0)
+      ::new (static_cast<Base*>(this)) Base(expr.size(), expr.nonZeros(), expr.innerIndexPtr(), expr.valuePtr());
+    else
+      ::new (static_cast<Base*>(this)) Base(expr.rows(), expr.cols(), expr.nonZeros(), expr.outerIndexPtr(), expr.innerIndexPtr(), expr.valuePtr(), expr.innerNonZeroPtr());
+  }
+};
+
+} // namespace internal
+
+
+/** 
+  * \ingroup SparseCore_Module
+  *
+  * \brief A sparse matrix expression referencing an existing sparse expression
+  *
+  * \tparam SparseMatrixType the equivalent sparse matrix type of the referenced data, it must be a template instance of class SparseMatrix.
+  * \tparam Options specifies whether the a standard compressed format is required \c Options is  \c #StandardCompressedFormat, or \c 0.
+  *                The default is \c 0.
+  *
+  * \sa class Ref
+  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType >
+  : public internal::SparseRefBase<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType > >
+#else
+template<typename SparseMatrixType, int Options>
+class Ref<SparseMatrixType, Options>
+  : public SparseMapBase<Derived,WriteAccessors> // yes, that's weird to use Derived here, but that works!
+#endif
+{
+    typedef SparseMatrix<MatScalar,MatOptions,MatIndex> PlainObjectType;
+    typedef internal::traits<Ref> Traits;
+    template<int OtherOptions>
+    inline Ref(const SparseMatrix<MatScalar,OtherOptions,MatIndex>& expr);
+    template<int OtherOptions>
+    inline Ref(const MappedSparseMatrix<MatScalar,OtherOptions,MatIndex>& expr);
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<int OtherOptions>
+    inline Ref(SparseMatrix<MatScalar,OtherOptions,MatIndex>& expr)
+    {
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<SparseMatrix<MatScalar,OtherOptions,MatIndex> >::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      eigen_assert( ((Options & int(StandardCompressedFormat))==0) || (expr.isCompressed()) );
+      Base::construct(expr.derived());
+    }
+    
+    template<int OtherOptions>
+    inline Ref(MappedSparseMatrix<MatScalar,OtherOptions,MatIndex>& expr)
+    {
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<SparseMatrix<MatScalar,OtherOptions,MatIndex> >::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      eigen_assert( ((Options & int(StandardCompressedFormat))==0) || (expr.isCompressed()) );
+      Base::construct(expr.derived());
+    }
+    
+    template<typename Derived>
+    inline Ref(const SparseCompressedBase<Derived>& expr)
+    #else
+    /** Implicit constructor from any sparse expression (2D matrix or 1D vector) */
+    template<typename Derived>
+    inline Ref(SparseCompressedBase<Derived>& expr)
+    #endif
+    {
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      eigen_assert( ((Options & int(StandardCompressedFormat))==0) || (expr.isCompressed()) );
+      Base::construct(expr.const_cast_derived());
+    }
+};
+
+// this is the const ref version
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public internal::SparseRefBase<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+{
+    typedef SparseMatrix<MatScalar,MatOptions,MatIndex> TPlainObjectType;
+    typedef internal::traits<Ref> Traits;
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+    template<typename Derived>
+    inline Ref(const SparseMatrixBase<Derived>& expr) : m_hasCopy(false)
+    {
+      construct(expr.derived(), typename Traits::template match<Derived>::type());
+    }
+
+    inline Ref(const Ref& other) : Base(other), m_hasCopy(false) {
+      // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
+    }
+
+    template<typename OtherRef>
+    inline Ref(const RefBase<OtherRef>& other) : m_hasCopy(false) {
+      construct(other.derived(), typename Traits::template match<OtherRef>::type());
+    }
+
+    ~Ref() {
+      if(m_hasCopy) {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        obj->~TPlainObjectType();
+      }
+    }
+
+  protected:
+
+    template<typename Expression>
+    void construct(const Expression& expr,internal::true_type)
+    {
+      if((Options & int(StandardCompressedFormat)) && (!expr.isCompressed()))
+      {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        ::new (obj) TPlainObjectType(expr);
+        m_hasCopy = true;
+        Base::construct(*obj);
+      }
+      else
+      {
+        Base::construct(expr);
+      }
+    }
+
+    template<typename Expression>
+    void construct(const Expression& expr, internal::false_type)
+    {
+      TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+      ::new (obj) TPlainObjectType(expr);
+      m_hasCopy = true;
+      Base::construct(*obj);
+    }
+
+  protected:
+    char m_object_bytes[sizeof(TPlainObjectType)];
+    bool m_hasCopy;
+};
+
+
+
+/**
+  * \ingroup SparseCore_Module
+  *
+  * \brief A sparse vector expression referencing an existing sparse vector expression
+  *
+  * \tparam SparseVectorType the equivalent sparse vector type of the referenced data, it must be a template instance of class SparseVector.
+  *
+  * \sa class Ref
+  */
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType >
+  : public internal::SparseRefBase<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType > >
+#else
+template<typename SparseVectorType>
+class Ref<SparseVectorType>
+  : public SparseMapBase<Derived,WriteAccessors>
+#endif
+{
+    typedef SparseVector<MatScalar,MatOptions,MatIndex> PlainObjectType;
+    typedef internal::traits<Ref> Traits;
+    template<int OtherOptions>
+    inline Ref(const SparseVector<MatScalar,OtherOptions,MatIndex>& expr);
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    template<int OtherOptions>
+    inline Ref(SparseVector<MatScalar,OtherOptions,MatIndex>& expr)
+    {
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<SparseVector<MatScalar,OtherOptions,MatIndex> >::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      Base::construct(expr.derived());
+    }
+
+    template<typename Derived>
+    inline Ref(const SparseCompressedBase<Derived>& expr)
+    #else
+    /** Implicit constructor from any 1D sparse vector expression */
+    template<typename Derived>
+    inline Ref(SparseCompressedBase<Derived>& expr)
+    #endif
+    {
+      EIGEN_STATIC_ASSERT(bool(internal::is_lvalue<Derived>::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY);
+      EIGEN_STATIC_ASSERT(bool(Traits::template match<Derived>::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH);
+      Base::construct(expr.const_cast_derived());
+    }
+};
+
+// this is the const ref version
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+class Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType>
+  : public internal::SparseRefBase<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+{
+    typedef SparseVector<MatScalar,MatOptions,MatIndex> TPlainObjectType;
+    typedef internal::traits<Ref> Traits;
+  public:
+
+    typedef internal::SparseRefBase<Ref> Base;
+    EIGEN_SPARSE_PUBLIC_INTERFACE(Ref)
+
+    template<typename Derived>
+    inline Ref(const SparseMatrixBase<Derived>& expr) : m_hasCopy(false)
+    {
+      construct(expr.derived(), typename Traits::template match<Derived>::type());
+    }
+
+    inline Ref(const Ref& other) : Base(other), m_hasCopy(false) {
+      // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy
+    }
+
+    template<typename OtherRef>
+    inline Ref(const RefBase<OtherRef>& other) : m_hasCopy(false) {
+      construct(other.derived(), typename Traits::template match<OtherRef>::type());
+    }
+
+    ~Ref() {
+      if(m_hasCopy) {
+        TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+        obj->~TPlainObjectType();
+      }
+    }
+
+  protected:
+
+    template<typename Expression>
+    void construct(const Expression& expr,internal::true_type)
+    {
+      Base::construct(expr);
+    }
+
+    template<typename Expression>
+    void construct(const Expression& expr, internal::false_type)
+    {
+      TPlainObjectType* obj = reinterpret_cast<TPlainObjectType*>(m_object_bytes);
+      ::new (obj) TPlainObjectType(expr);
+      m_hasCopy = true;
+      Base::construct(*obj);
+    }
+
+  protected:
+    char m_object_bytes[sizeof(TPlainObjectType)];
+    bool m_hasCopy;
+};
+
+namespace internal {
+
+// FIXME shall we introduce a general evaluatior_ref that we can specialize for any sparse object once, and thus remove this copy-pasta thing...
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<const SparseMatrix<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;  
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+template<typename MatScalar, int MatOptions, typename MatIndex, int Options, typename StrideType>
+struct evaluator<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> >
+  : evaluator<SparseCompressedBase<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > >
+{
+  typedef evaluator<SparseCompressedBase<Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> > > Base;
+  typedef Ref<const SparseVector<MatScalar,MatOptions,MatIndex>, Options, StrideType> XprType;
+  evaluator() : Base() {}
+  explicit evaluator(const XprType &mat) : Base(mat) {}
+};
+
+}
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSE_REF_H
diff --git a/Eigen/src/SparseCore/SparseSelfAdjointView.h b/Eigen/src/SparseCore/SparseSelfAdjointView.h
index 0eda96b..65611b3 100644
--- a/Eigen/src/SparseCore/SparseSelfAdjointView.h
+++ b/Eigen/src/SparseCore/SparseSelfAdjointView.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -11,14 +11,14 @@
 #define EIGEN_SPARSE_SELFADJOINTVIEW_H
 
 namespace Eigen { 
-
+  
 /** \ingroup SparseCore_Module
   * \class SparseSelfAdjointView
   *
   * \brief Pseudo expression to manipulate a triangular sparse matrix as a selfadjoint matrix.
   *
   * \param MatrixType the type of the dense matrix storing the coefficients
-  * \param UpLo can be either \c #Lower or \c #Upper
+  * \param Mode can be either \c #Lower or \c #Upper
   *
   * This class is an expression of a sefladjoint matrix from a triangular part of a matrix
   * with given dense storage of the coefficients. It is the return type of MatrixBase::selfadjointView()
@@ -26,38 +26,40 @@
   *
   * \sa SparseMatrixBase::selfadjointView()
   */
-template<typename Lhs, typename Rhs, int UpLo>
-class SparseSelfAdjointTimeDenseProduct;
-
-template<typename Lhs, typename Rhs, int UpLo>
-class DenseTimeSparseSelfAdjointProduct;
-
 namespace internal {
   
-template<typename MatrixType, unsigned int UpLo>
-struct traits<SparseSelfAdjointView<MatrixType,UpLo> > : traits<MatrixType> {
+template<typename MatrixType, unsigned int Mode>
+struct traits<SparseSelfAdjointView<MatrixType,Mode> > : traits<MatrixType> {
 };
 
-template<int SrcUpLo,int DstUpLo,typename MatrixType,int DestOrder>
-void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm = 0);
+template<int SrcMode,int DstMode,typename MatrixType,int DestOrder>
+void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm = 0);
 
-template<int UpLo,typename MatrixType,int DestOrder>
-void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm = 0);
+template<int Mode,typename MatrixType,int DestOrder>
+void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm = 0);
 
 }
 
-template<typename MatrixType, unsigned int UpLo> class SparseSelfAdjointView
-  : public EigenBase<SparseSelfAdjointView<MatrixType,UpLo> >
+template<typename MatrixType, unsigned int _Mode> class SparseSelfAdjointView
+  : public EigenBase<SparseSelfAdjointView<MatrixType,_Mode> >
 {
   public:
+    
+    enum {
+      Mode = _Mode,
+      TransposeMode = ((Mode & Upper) ? Lower : 0) | ((Mode & Lower) ? Upper : 0),
+      RowsAtCompileTime = internal::traits<SparseSelfAdjointView>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<SparseSelfAdjointView>::ColsAtCompileTime
+    };
 
+    typedef EigenBase<SparseSelfAdjointView> Base;
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
-    typedef Matrix<Index,Dynamic,1> VectorI;
-    typedef typename MatrixType::Nested MatrixTypeNested;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+    typedef typename internal::ref_selector<MatrixType>::non_const_type MatrixTypeNested;
     typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
-
-    inline SparseSelfAdjointView(const MatrixType& matrix) : m_matrix(matrix)
+    
+    explicit inline SparseSelfAdjointView(MatrixType& matrix) : m_matrix(matrix)
     {
       eigen_assert(rows()==cols() && "SelfAdjointView is only for squared matrices");
     }
@@ -67,7 +69,7 @@
 
     /** \internal \returns a reference to the nested matrix */
     const _MatrixTypeNested& matrix() const { return m_matrix; }
-    _MatrixTypeNested& matrix() { return m_matrix.const_cast_derived(); }
+    typename internal::remove_reference<MatrixTypeNested>::type& matrix() { return m_matrix; }
 
     /** \returns an expression of the matrix product between a sparse self-adjoint matrix \c *this and a sparse matrix \a rhs.
       *
@@ -75,10 +77,10 @@
       * Indeed, the SparseSelfadjointView operand is first copied into a temporary SparseMatrix before computing the product.
       */
     template<typename OtherDerived>
-    SparseSparseProduct<typename OtherDerived::PlainObject, OtherDerived>
+    Product<SparseSelfAdjointView, OtherDerived>
     operator*(const SparseMatrixBase<OtherDerived>& rhs) const
     {
-      return SparseSparseProduct<typename OtherDerived::PlainObject, OtherDerived>(*this, rhs.derived());
+      return Product<SparseSelfAdjointView, OtherDerived>(*this, rhs.derived());
     }
 
     /** \returns an expression of the matrix product between a sparse matrix \a lhs and a sparse self-adjoint matrix \a rhs.
@@ -87,26 +89,26 @@
       * Indeed, the SparseSelfadjointView operand is first copied into a temporary SparseMatrix before computing the product.
       */
     template<typename OtherDerived> friend
-    SparseSparseProduct<OtherDerived, typename OtherDerived::PlainObject >
+    Product<OtherDerived, SparseSelfAdjointView>
     operator*(const SparseMatrixBase<OtherDerived>& lhs, const SparseSelfAdjointView& rhs)
     {
-      return SparseSparseProduct<OtherDerived, typename OtherDerived::PlainObject>(lhs.derived(), rhs);
+      return Product<OtherDerived, SparseSelfAdjointView>(lhs.derived(), rhs);
     }
     
     /** Efficient sparse self-adjoint matrix times dense vector/matrix product */
     template<typename OtherDerived>
-    SparseSelfAdjointTimeDenseProduct<MatrixType,OtherDerived,UpLo>
+    Product<SparseSelfAdjointView,OtherDerived>
     operator*(const MatrixBase<OtherDerived>& rhs) const
     {
-      return SparseSelfAdjointTimeDenseProduct<MatrixType,OtherDerived,UpLo>(m_matrix, rhs.derived());
+      return Product<SparseSelfAdjointView,OtherDerived>(*this, rhs.derived());
     }
 
     /** Efficient dense vector/matrix times sparse self-adjoint matrix product */
     template<typename OtherDerived> friend
-    DenseTimeSparseSelfAdjointProduct<OtherDerived,MatrixType,UpLo>
+    Product<OtherDerived,SparseSelfAdjointView>
     operator*(const MatrixBase<OtherDerived>& lhs, const SparseSelfAdjointView& rhs)
     {
-      return DenseTimeSparseSelfAdjointProduct<OtherDerived,_MatrixTypeNested,UpLo>(lhs.derived(), rhs.m_matrix);
+      return Product<OtherDerived,SparseSelfAdjointView>(lhs.derived(), rhs);
     }
 
     /** Perform a symmetric rank K update of the selfadjoint matrix \c *this:
@@ -120,56 +122,48 @@
     template<typename DerivedU>
     SparseSelfAdjointView& rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha = Scalar(1));
     
-    /** \internal triggered by sparse_matrix = SparseSelfadjointView; */
-    template<typename DestScalar,int StorageOrder> void evalTo(SparseMatrix<DestScalar,StorageOrder,Index>& _dest) const
-    {
-      internal::permute_symm_to_fullsymm<UpLo>(m_matrix, _dest);
-    }
-    
-    template<typename DestScalar> void evalTo(DynamicSparseMatrix<DestScalar,ColMajor,Index>& _dest) const
-    {
-      // TODO directly evaluate into _dest;
-      SparseMatrix<DestScalar,ColMajor,Index> tmp(_dest.rows(),_dest.cols());
-      internal::permute_symm_to_fullsymm<UpLo>(m_matrix, tmp);
-      _dest = tmp;
-    }
-    
     /** \returns an expression of P H P^-1 */
-    SparseSymmetricPermutationProduct<_MatrixTypeNested,UpLo> twistedBy(const PermutationMatrix<Dynamic,Dynamic,Index>& perm) const
+    // TODO implement twists in a more evaluator friendly fashion
+    SparseSymmetricPermutationProduct<_MatrixTypeNested,Mode> twistedBy(const PermutationMatrix<Dynamic,Dynamic,StorageIndex>& perm) const
     {
-      return SparseSymmetricPermutationProduct<_MatrixTypeNested,UpLo>(m_matrix, perm);
+      return SparseSymmetricPermutationProduct<_MatrixTypeNested,Mode>(m_matrix, perm);
     }
-    
-    template<typename SrcMatrixType,int SrcUpLo>
-    SparseSelfAdjointView& operator=(const SparseSymmetricPermutationProduct<SrcMatrixType,SrcUpLo>& permutedMatrix)
+
+    template<typename SrcMatrixType,int SrcMode>
+    SparseSelfAdjointView& operator=(const SparseSymmetricPermutationProduct<SrcMatrixType,SrcMode>& permutedMatrix)
     {
-      permutedMatrix.evalTo(*this);
+      internal::call_assignment_no_alias_no_transpose(*this, permutedMatrix);
       return *this;
     }
 
-
     SparseSelfAdjointView& operator=(const SparseSelfAdjointView& src)
     {
-      PermutationMatrix<Dynamic> pnull;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> pnull;
       return *this = src.twistedBy(pnull);
     }
 
-    template<typename SrcMatrixType,unsigned int SrcUpLo>
-    SparseSelfAdjointView& operator=(const SparseSelfAdjointView<SrcMatrixType,SrcUpLo>& src)
+    template<typename SrcMatrixType,unsigned int SrcMode>
+    SparseSelfAdjointView& operator=(const SparseSelfAdjointView<SrcMatrixType,SrcMode>& src)
     {
-      PermutationMatrix<Dynamic> pnull;
+      PermutationMatrix<Dynamic,Dynamic,StorageIndex> pnull;
       return *this = src.twistedBy(pnull);
     }
     
-
-    // const SparseLLT<PlainObject, UpLo> llt() const;
-    // const SparseLDLT<PlainObject, UpLo> ldlt() const;
-
+    void resize(Index rows, Index cols)
+    {
+      EIGEN_ONLY_USED_FOR_DEBUG(rows);
+      EIGEN_ONLY_USED_FOR_DEBUG(cols);
+      eigen_assert(rows == this->rows() && cols == this->cols()
+                && "SparseSelfadjointView::resize() does not actually allow to resize.");
+    }
+    
   protected:
 
-    typename MatrixType::Nested m_matrix;
-    mutable VectorI m_countPerRow;
-    mutable VectorI m_countPerCol;
+    MatrixTypeNested m_matrix;
+    //mutable VectorI m_countPerRow;
+    //mutable VectorI m_countPerCol;
+  private:
+    template<typename Dest> void evalTo(Dest &) const;
 };
 
 /***************************************************************************
@@ -178,146 +172,268 @@
 
 template<typename Derived>
 template<unsigned int UpLo>
-const SparseSelfAdjointView<Derived, UpLo> SparseMatrixBase<Derived>::selfadjointView() const
+typename SparseMatrixBase<Derived>::template ConstSelfAdjointViewReturnType<UpLo>::Type SparseMatrixBase<Derived>::selfadjointView() const
 {
-  return derived();
+  return SparseSelfAdjointView<const Derived, UpLo>(derived());
 }
 
 template<typename Derived>
 template<unsigned int UpLo>
-SparseSelfAdjointView<Derived, UpLo> SparseMatrixBase<Derived>::selfadjointView()
+typename SparseMatrixBase<Derived>::template SelfAdjointViewReturnType<UpLo>::Type SparseMatrixBase<Derived>::selfadjointView()
 {
-  return derived();
+  return SparseSelfAdjointView<Derived, UpLo>(derived());
 }
 
 /***************************************************************************
 * Implementation of SparseSelfAdjointView methods
 ***************************************************************************/
 
-template<typename MatrixType, unsigned int UpLo>
+template<typename MatrixType, unsigned int Mode>
 template<typename DerivedU>
-SparseSelfAdjointView<MatrixType,UpLo>&
-SparseSelfAdjointView<MatrixType,UpLo>::rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha)
+SparseSelfAdjointView<MatrixType,Mode>&
+SparseSelfAdjointView<MatrixType,Mode>::rankUpdate(const SparseMatrixBase<DerivedU>& u, const Scalar& alpha)
 {
-  SparseMatrix<Scalar,MatrixType::Flags&RowMajorBit?RowMajor:ColMajor> tmp = u * u.adjoint();
+  SparseMatrix<Scalar,(MatrixType::Flags&RowMajorBit)?RowMajor:ColMajor> tmp = u * u.adjoint();
   if(alpha==Scalar(0))
-    m_matrix.const_cast_derived() = tmp.template triangularView<UpLo>();
+    m_matrix = tmp.template triangularView<Mode>();
   else
-    m_matrix.const_cast_derived() += alpha * tmp.template triangularView<UpLo>();
+    m_matrix += alpha * tmp.template triangularView<Mode>();
 
   return *this;
 }
 
+namespace internal {
+  
+// TODO currently a selfadjoint expression has the form SelfAdjointView<.,.>
+//      in the future selfadjoint-ness should be defined by the expression traits
+//      such that Transpose<SelfAdjointView<.,.> > is valid. (currently TriangularBase::transpose() is overloaded to make it work)
+template<typename MatrixType, unsigned int Mode>
+struct evaluator_traits<SparseSelfAdjointView<MatrixType,Mode> >
+{
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef SparseSelfAdjointShape Shape;
+};
+
+struct SparseSelfAdjoint2Sparse {};
+
+template<> struct AssignmentKind<SparseShape,SparseSelfAdjointShape> { typedef SparseSelfAdjoint2Sparse Kind; };
+template<> struct AssignmentKind<SparseSelfAdjointShape,SparseShape> { typedef Sparse2Sparse Kind; };
+
+template< typename DstXprType, typename SrcXprType, typename Functor>
+struct Assignment<DstXprType, SrcXprType, Functor, SparseSelfAdjoint2Sparse>
+{
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  typedef internal::assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar> AssignOpType;
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignOpType&/*func*/)
+  {
+    internal::permute_symm_to_fullsymm<SrcXprType::Mode>(src.matrix(), dst);
+  }
+
+  // FIXME: the handling of += and -= in sparse matrices should be cleanup so that next two overloads could be reduced to:
+  template<typename DestScalar,int StorageOrder,typename AssignFunc>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src, const AssignFunc& func)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    call_assignment_no_alias_no_transpose(dst, tmp, func);
+  }
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
+                  const internal::add_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    dst += tmp;
+  }
+
+  template<typename DestScalar,int StorageOrder>
+  static void run(SparseMatrix<DestScalar,StorageOrder,StorageIndex> &dst, const SrcXprType &src,
+                  const internal::sub_assign_op<typename DstXprType::Scalar,typename SrcXprType::Scalar>& /* func */)
+  {
+    SparseMatrix<DestScalar,StorageOrder,StorageIndex> tmp(src.rows(),src.cols());
+    run(tmp, src, AssignOpType());
+    dst -= tmp;
+  }
+  
+  template<typename DestScalar>
+  static void run(DynamicSparseMatrix<DestScalar,ColMajor,StorageIndex>& dst, const SrcXprType &src, const AssignOpType&/*func*/)
+  {
+    // TODO directly evaluate into dst;
+    SparseMatrix<DestScalar,ColMajor,StorageIndex> tmp(dst.rows(),dst.cols());
+    internal::permute_symm_to_fullsymm<SrcXprType::Mode>(src.matrix(), tmp);
+    dst = tmp;
+  }
+};
+
+} // end namespace internal
+
 /***************************************************************************
 * Implementation of sparse self-adjoint time dense matrix
 ***************************************************************************/
 
 namespace internal {
-template<typename Lhs, typename Rhs, int UpLo>
-struct traits<SparseSelfAdjointTimeDenseProduct<Lhs,Rhs,UpLo> >
- : traits<ProductBase<SparseSelfAdjointTimeDenseProduct<Lhs,Rhs,UpLo>, Lhs, Rhs> >
+
+template<int Mode, typename SparseLhsType, typename DenseRhsType, typename DenseResType, typename AlphaType>
+inline void sparse_selfadjoint_time_dense_product(const SparseLhsType& lhs, const DenseRhsType& rhs, DenseResType& res, const AlphaType& alpha)
 {
-  typedef Dense StorageKind;
-};
-}
+  EIGEN_ONLY_USED_FOR_DEBUG(alpha);
+  
+  typedef typename internal::nested_eval<SparseLhsType,DenseRhsType::MaxColsAtCompileTime>::type SparseLhsTypeNested;
+  typedef typename internal::remove_all<SparseLhsTypeNested>::type SparseLhsTypeNestedCleaned;
+  typedef evaluator<SparseLhsTypeNestedCleaned> LhsEval;
+  typedef typename LhsEval::InnerIterator LhsIterator;
+  typedef typename SparseLhsType::Scalar LhsScalar;
+  
+  enum {
+    LhsIsRowMajor = (LhsEval::Flags&RowMajorBit)==RowMajorBit,
+    ProcessFirstHalf =
+              ((Mode&(Upper|Lower))==(Upper|Lower))
+          || ( (Mode&Upper) && !LhsIsRowMajor)
+          || ( (Mode&Lower) && LhsIsRowMajor),
+    ProcessSecondHalf = !ProcessFirstHalf
+  };
+  
+  SparseLhsTypeNested lhs_nested(lhs);
+  LhsEval lhsEval(lhs_nested);
 
-template<typename Lhs, typename Rhs, int UpLo>
-class SparseSelfAdjointTimeDenseProduct
-  : public ProductBase<SparseSelfAdjointTimeDenseProduct<Lhs,Rhs,UpLo>, Lhs, Rhs>
-{
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(SparseSelfAdjointTimeDenseProduct)
-
-    SparseSelfAdjointTimeDenseProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {}
-
-    template<typename Dest> void scaleAndAddTo(Dest& dest, const Scalar& alpha) const
+  // work on one column at once
+  for (Index k=0; k<rhs.cols(); ++k)
+  {
+    for (Index j=0; j<lhs.outerSize(); ++j)
     {
-      EIGEN_ONLY_USED_FOR_DEBUG(alpha);
-      // TODO use alpha
-      eigen_assert(alpha==Scalar(1) && "alpha != 1 is not implemented yet, sorry");
-      typedef typename internal::remove_all<Lhs>::type _Lhs;
-      typedef typename _Lhs::InnerIterator LhsInnerIterator;
-      enum {
-        LhsIsRowMajor = (_Lhs::Flags&RowMajorBit)==RowMajorBit,
-        ProcessFirstHalf =
-                 ((UpLo&(Upper|Lower))==(Upper|Lower))
-              || ( (UpLo&Upper) && !LhsIsRowMajor)
-              || ( (UpLo&Lower) && LhsIsRowMajor),
-        ProcessSecondHalf = !ProcessFirstHalf
-      };
-      for (Index j=0; j<m_lhs.outerSize(); ++j)
+      LhsIterator i(lhsEval,j);
+      // handle diagonal coeff
+      if (ProcessSecondHalf)
       {
-        LhsInnerIterator i(m_lhs,j);
-        if (ProcessSecondHalf)
+        while (i && i.index()<j) ++i;
+        if(i && i.index()==j)
         {
-          while (i && i.index()<j) ++i;
-          if(i && i.index()==j)
-          {
-            dest.row(j) += i.value() * m_rhs.row(j);
-            ++i;
-          }
+          res.coeffRef(j,k) += alpha * i.value() * rhs.coeff(j,k);
+          ++i;
         }
-        for(; (ProcessFirstHalf ? i && i.index() < j : i) ; ++i)
-        {
-          Index a = LhsIsRowMajor ? j : i.index();
-          Index b = LhsIsRowMajor ? i.index() : j;
-          typename Lhs::Scalar v = i.value();
-          dest.row(a) += (v) * m_rhs.row(b);
-          dest.row(b) += numext::conj(v) * m_rhs.row(a);
-        }
-        if (ProcessFirstHalf && i && (i.index()==j))
-          dest.row(j) += i.value() * m_rhs.row(j);
       }
+
+      // premultiplied rhs for scatters
+      typename ScalarBinaryOpTraits<AlphaType, typename DenseRhsType::Scalar>::ReturnType rhs_j(alpha*rhs(j,k));
+      // accumulator for partial scalar product
+      typename DenseResType::Scalar res_j(0);
+      for(; (ProcessFirstHalf ? i && i.index() < j : i) ; ++i)
+      {
+        LhsScalar lhs_ij = i.value();
+        if(!LhsIsRowMajor) lhs_ij = numext::conj(lhs_ij);
+        res_j += lhs_ij * rhs.coeff(i.index(),k);
+        res(i.index(),k) += numext::conj(lhs_ij) * rhs_j;
+      }
+      res.coeffRef(j,k) += alpha * res_j;
+
+      // handle diagonal coeff
+      if (ProcessFirstHalf && i && (i.index()==j))
+        res.coeffRef(j,k) += alpha * i.value() * rhs.coeff(j,k);
     }
-
-  private:
-    SparseSelfAdjointTimeDenseProduct& operator=(const SparseSelfAdjointTimeDenseProduct&);
-};
-
-namespace internal {
-template<typename Lhs, typename Rhs, int UpLo>
-struct traits<DenseTimeSparseSelfAdjointProduct<Lhs,Rhs,UpLo> >
- : traits<ProductBase<DenseTimeSparseSelfAdjointProduct<Lhs,Rhs,UpLo>, Lhs, Rhs> >
-{};
+  }
 }
 
-template<typename Lhs, typename Rhs, int UpLo>
-class DenseTimeSparseSelfAdjointProduct
-  : public ProductBase<DenseTimeSparseSelfAdjointProduct<Lhs,Rhs,UpLo>, Lhs, Rhs>
+
+template<typename LhsView, typename Rhs, int ProductType>
+struct generic_product_impl<LhsView, Rhs, SparseSelfAdjointShape, DenseShape, ProductType>
+: generic_product_impl_base<LhsView, Rhs, generic_product_impl<LhsView, Rhs, SparseSelfAdjointShape, DenseShape, ProductType> >
 {
-  public:
-    EIGEN_PRODUCT_PUBLIC_INTERFACE(DenseTimeSparseSelfAdjointProduct)
-
-    DenseTimeSparseSelfAdjointProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs)
-    {}
-
-    template<typename Dest> void scaleAndAddTo(Dest& /*dest*/, const Scalar& /*alpha*/) const
-    {
-      // TODO
-    }
-
-  private:
-    DenseTimeSparseSelfAdjointProduct& operator=(const DenseTimeSparseSelfAdjointProduct&);
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const LhsView& lhsView, const Rhs& rhs, const typename Dest::Scalar& alpha)
+  {
+    typedef typename LhsView::_MatrixTypeNested Lhs;
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhsView.matrix());
+    RhsNested rhsNested(rhs);
+    
+    internal::sparse_selfadjoint_time_dense_product<LhsView::Mode>(lhsNested, rhsNested, dst, alpha);
+  }
 };
 
+template<typename Lhs, typename RhsView, int ProductType>
+struct generic_product_impl<Lhs, RhsView, DenseShape, SparseSelfAdjointShape, ProductType>
+: generic_product_impl_base<Lhs, RhsView, generic_product_impl<Lhs, RhsView, DenseShape, SparseSelfAdjointShape, ProductType> >
+{
+  template<typename Dest>
+  static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const RhsView& rhsView, const typename Dest::Scalar& alpha)
+  {
+    typedef typename RhsView::_MatrixTypeNested Rhs;
+    typedef typename nested_eval<Lhs,Dynamic>::type LhsNested;
+    typedef typename nested_eval<Rhs,Dynamic>::type RhsNested;
+    LhsNested lhsNested(lhs);
+    RhsNested rhsNested(rhsView.matrix());
+    
+    // transpose everything
+    Transpose<Dest> dstT(dst);
+    internal::sparse_selfadjoint_time_dense_product<RhsView::TransposeMode>(rhsNested.transpose(), lhsNested.transpose(), dstT, alpha);
+  }
+};
+
+// NOTE: these two overloads are needed to evaluate the sparse selfadjoint view into a full sparse matrix
+// TODO: maybe the copy could be handled by generic_product_impl so that these overloads would not be needed anymore
+
+template<typename LhsView, typename Rhs, int ProductTag>
+struct product_evaluator<Product<LhsView, Rhs, DefaultProduct>, ProductTag, SparseSelfAdjointShape, SparseShape>
+  : public evaluator<typename Product<typename Rhs::PlainObject, Rhs, DefaultProduct>::PlainObject>
+{
+  typedef Product<LhsView, Rhs, DefaultProduct> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  product_evaluator(const XprType& xpr)
+    : m_lhs(xpr.lhs()), m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<typename Rhs::PlainObject, Rhs, SparseShape, SparseShape, ProductTag>::evalTo(m_result, m_lhs, xpr.rhs());
+  }
+  
+protected:
+  typename Rhs::PlainObject m_lhs;
+  PlainObject m_result;
+};
+
+template<typename Lhs, typename RhsView, int ProductTag>
+struct product_evaluator<Product<Lhs, RhsView, DefaultProduct>, ProductTag, SparseShape, SparseSelfAdjointShape>
+  : public evaluator<typename Product<Lhs, typename Lhs::PlainObject, DefaultProduct>::PlainObject>
+{
+  typedef Product<Lhs, RhsView, DefaultProduct> XprType;
+  typedef typename XprType::PlainObject PlainObject;
+  typedef evaluator<PlainObject> Base;
+
+  product_evaluator(const XprType& xpr)
+    : m_rhs(xpr.rhs()), m_result(xpr.rows(), xpr.cols())
+  {
+    ::new (static_cast<Base*>(this)) Base(m_result);
+    generic_product_impl<Lhs, typename Lhs::PlainObject, SparseShape, SparseShape, ProductTag>::evalTo(m_result, xpr.lhs(), m_rhs);
+  }
+  
+protected:
+  typename Lhs::PlainObject m_rhs;
+  PlainObject m_result;
+};
+
+} // namespace internal
+
 /***************************************************************************
 * Implementation of symmetric copies and permutations
 ***************************************************************************/
 namespace internal {
-  
-template<typename MatrixType, int UpLo>
-struct traits<SparseSymmetricPermutationProduct<MatrixType,UpLo> > : traits<MatrixType> {
-};
 
-template<int UpLo,typename MatrixType,int DestOrder>
-void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm)
+template<int Mode,typename MatrixType,int DestOrder>
+void permute_symm_to_fullsymm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DestOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm)
 {
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef typename MatrixType::Scalar Scalar;
-  typedef SparseMatrix<Scalar,DestOrder,Index> Dest;
-  typedef Matrix<Index,Dynamic,1> VectorI;
+  typedef SparseMatrix<Scalar,DestOrder,StorageIndex> Dest;
+  typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+  typedef evaluator<MatrixType> MatEval;
+  typedef typename evaluator<MatrixType>::InnerIterator MatIterator;
   
+  MatEval matEval(mat);
   Dest& dest(_dest.derived());
   enum {
     StorageOrderMatch = int(Dest::IsRowMajor) == int(MatrixType::IsRowMajor)
@@ -331,17 +447,17 @@
   for(Index j = 0; j<size; ++j)
   {
     Index jp = perm ? perm[j] : j;
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    for(MatIterator it(matEval,j); it; ++it)
     {
       Index i = it.index();
       Index r = it.row();
       Index c = it.col();
       Index ip = perm ? perm[i] : i;
-      if(UpLo==(Upper|Lower))
+      if(Mode==(Upper|Lower))
         count[StorageOrderMatch ? jp : ip]++;
       else if(r==c)
         count[ip]++;
-      else if(( UpLo==Lower && r>c) || ( UpLo==Upper && r<c))
+      else if(( Mode==Lower && r>c) || ( Mode==Upper && r<c))
       {
         count[ip]++;
         count[jp]++;
@@ -359,18 +475,18 @@
     count[j] = dest.outerIndexPtr()[j];
   
   // copy data
-  for(Index j = 0; j<size; ++j)
+  for(StorageIndex j = 0; j<size; ++j)
   {
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    for(MatIterator it(matEval,j); it; ++it)
     {
-      Index i = it.index();
+      StorageIndex i = internal::convert_index<StorageIndex>(it.index());
       Index r = it.row();
       Index c = it.col();
       
-      Index jp = perm ? perm[j] : j;
-      Index ip = perm ? perm[i] : i;
+      StorageIndex jp = perm ? perm[j] : j;
+      StorageIndex ip = perm ? perm[i] : i;
       
-      if(UpLo==(Upper|Lower))
+      if(Mode==(Upper|Lower))
       {
         Index k = count[StorageOrderMatch ? jp : ip]++;
         dest.innerIndexPtr()[k] = StorageOrderMatch ? ip : jp;
@@ -382,7 +498,7 @@
         dest.innerIndexPtr()[k] = ip;
         dest.valuePtr()[k] = it.value();
       }
-      else if(( (UpLo&Lower)==Lower && r>c) || ( (UpLo&Upper)==Upper && r<c))
+      else if(( (Mode&Lower)==Lower && r>c) || ( (Mode&Upper)==Upper && r<c))
       {
         if(!StorageOrderMatch)
           std::swap(ip,jp);
@@ -397,35 +513,40 @@
   }
 }
 
-template<int _SrcUpLo,int _DstUpLo,typename MatrixType,int DstOrder>
-void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DstOrder,typename MatrixType::Index>& _dest, const typename MatrixType::Index* perm)
+template<int _SrcMode,int _DstMode,typename MatrixType,int DstOrder>
+void permute_symm_to_symm(const MatrixType& mat, SparseMatrix<typename MatrixType::Scalar,DstOrder,typename MatrixType::StorageIndex>& _dest, const typename MatrixType::StorageIndex* perm)
 {
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef typename MatrixType::Scalar Scalar;
-  SparseMatrix<Scalar,DstOrder,Index>& dest(_dest.derived());
-  typedef Matrix<Index,Dynamic,1> VectorI;
+  SparseMatrix<Scalar,DstOrder,StorageIndex>& dest(_dest.derived());
+  typedef Matrix<StorageIndex,Dynamic,1> VectorI;
+  typedef evaluator<MatrixType> MatEval;
+  typedef typename evaluator<MatrixType>::InnerIterator MatIterator;
+
   enum {
     SrcOrder = MatrixType::IsRowMajor ? RowMajor : ColMajor,
     StorageOrderMatch = int(SrcOrder) == int(DstOrder),
-    DstUpLo = DstOrder==RowMajor ? (_DstUpLo==Upper ? Lower : Upper) : _DstUpLo,
-    SrcUpLo = SrcOrder==RowMajor ? (_SrcUpLo==Upper ? Lower : Upper) : _SrcUpLo
+    DstMode = DstOrder==RowMajor ? (_DstMode==Upper ? Lower : Upper) : _DstMode,
+    SrcMode = SrcOrder==RowMajor ? (_SrcMode==Upper ? Lower : Upper) : _SrcMode
   };
+
+  MatEval matEval(mat);
   
   Index size = mat.rows();
   VectorI count(size);
   count.setZero();
   dest.resize(size,size);
-  for(Index j = 0; j<size; ++j)
+  for(StorageIndex j = 0; j<size; ++j)
   {
-    Index jp = perm ? perm[j] : j;
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    StorageIndex jp = perm ? perm[j] : j;
+    for(MatIterator it(matEval,j); it; ++it)
     {
-      Index i = it.index();
-      if((int(SrcUpLo)==int(Lower) && i<j) || (int(SrcUpLo)==int(Upper) && i>j))
+      StorageIndex i = it.index();
+      if((int(SrcMode)==int(Lower) && i<j) || (int(SrcMode)==int(Upper) && i>j))
         continue;
                   
-      Index ip = perm ? perm[i] : i;
-      count[int(DstUpLo)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
+      StorageIndex ip = perm ? perm[i] : i;
+      count[int(DstMode)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
     }
   }
   dest.outerIndexPtr()[0] = 0;
@@ -435,23 +556,23 @@
   for(Index j=0; j<size; ++j)
     count[j] = dest.outerIndexPtr()[j];
   
-  for(Index j = 0; j<size; ++j)
+  for(StorageIndex j = 0; j<size; ++j)
   {
     
-    for(typename MatrixType::InnerIterator it(mat,j); it; ++it)
+    for(MatIterator it(matEval,j); it; ++it)
     {
-      Index i = it.index();
-      if((int(SrcUpLo)==int(Lower) && i<j) || (int(SrcUpLo)==int(Upper) && i>j))
+      StorageIndex i = it.index();
+      if((int(SrcMode)==int(Lower) && i<j) || (int(SrcMode)==int(Upper) && i>j))
         continue;
                   
-      Index jp = perm ? perm[j] : j;
-      Index ip = perm? perm[i] : i;
+      StorageIndex jp = perm ? perm[j] : j;
+      StorageIndex ip = perm? perm[i] : i;
       
-      Index k = count[int(DstUpLo)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
-      dest.innerIndexPtr()[k] = int(DstUpLo)==int(Lower) ? (std::max)(ip,jp) : (std::min)(ip,jp);
+      Index k = count[int(DstMode)==int(Lower) ? (std::min)(ip,jp) : (std::max)(ip,jp)]++;
+      dest.innerIndexPtr()[k] = int(DstMode)==int(Lower) ? (std::max)(ip,jp) : (std::min)(ip,jp);
       
       if(!StorageOrderMatch) std::swap(ip,jp);
-      if( ((int(DstUpLo)==int(Lower) && ip<jp) || (int(DstUpLo)==int(Upper) && ip>jp)))
+      if( ((int(DstMode)==int(Lower) && ip<jp) || (int(DstMode)==int(Upper) && ip>jp)))
         dest.valuePtr()[k] = numext::conj(it.value());
       else
         dest.valuePtr()[k] = it.value();
@@ -461,19 +582,33 @@
 
 }
 
-template<typename MatrixType,int UpLo>
+// TODO implement twists in a more evaluator friendly fashion
+
+namespace internal {
+
+template<typename MatrixType, int Mode>
+struct traits<SparseSymmetricPermutationProduct<MatrixType,Mode> > : traits<MatrixType> {
+};
+
+}
+
+template<typename MatrixType,int Mode>
 class SparseSymmetricPermutationProduct
-  : public EigenBase<SparseSymmetricPermutationProduct<MatrixType,UpLo> >
+  : public EigenBase<SparseSymmetricPermutationProduct<MatrixType,Mode> >
 {
   public:
     typedef typename MatrixType::Scalar Scalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    enum {
+      RowsAtCompileTime = internal::traits<SparseSymmetricPermutationProduct>::RowsAtCompileTime,
+      ColsAtCompileTime = internal::traits<SparseSymmetricPermutationProduct>::ColsAtCompileTime
+    };
   protected:
-    typedef PermutationMatrix<Dynamic,Dynamic,Index> Perm;
+    typedef PermutationMatrix<Dynamic,Dynamic,StorageIndex> Perm;
   public:
-    typedef Matrix<Index,Dynamic,1> VectorI;
+    typedef Matrix<StorageIndex,Dynamic,1> VectorI;
     typedef typename MatrixType::Nested MatrixTypeNested;
-    typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
+    typedef typename internal::remove_all<MatrixTypeNested>::type NestedExpression;
     
     SparseSymmetricPermutationProduct(const MatrixType& mat, const Perm& perm)
       : m_matrix(mat), m_perm(perm)
@@ -481,20 +616,9 @@
     
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
-    
-    template<typename DestScalar, int Options, typename DstIndex>
-    void evalTo(SparseMatrix<DestScalar,Options,DstIndex>& _dest) const
-    {
-//       internal::permute_symm_to_fullsymm<UpLo>(m_matrix,_dest,m_perm.indices().data());
-      SparseMatrix<DestScalar,(Options&RowMajor)==RowMajor ? ColMajor : RowMajor, DstIndex> tmp;
-      internal::permute_symm_to_fullsymm<UpLo>(m_matrix,tmp,m_perm.indices().data());
-      _dest = tmp;
-    }
-    
-    template<typename DestType,unsigned int DestUpLo> void evalTo(SparseSelfAdjointView<DestType,DestUpLo>& dest) const
-    {
-      internal::permute_symm_to_symm<UpLo,DestUpLo>(m_matrix,dest.matrix(),m_perm.indices().data());
-    }
+        
+    const NestedExpression& matrix() const { return m_matrix; }
+    const Perm& perm() const { return m_perm; }
     
   protected:
     MatrixTypeNested m_matrix;
@@ -502,6 +626,31 @@
 
 };
 
+namespace internal {
+  
+template<typename DstXprType, typename MatrixType, int Mode, typename Scalar>
+struct Assignment<DstXprType, SparseSymmetricPermutationProduct<MatrixType,Mode>, internal::assign_op<Scalar,typename MatrixType::Scalar>, Sparse2Sparse>
+{
+  typedef SparseSymmetricPermutationProduct<MatrixType,Mode> SrcXprType;
+  typedef typename DstXprType::StorageIndex DstIndex;
+  template<int Options>
+  static void run(SparseMatrix<Scalar,Options,DstIndex> &dst, const SrcXprType &src, const internal::assign_op<Scalar,typename MatrixType::Scalar> &)
+  {
+    // internal::permute_symm_to_fullsymm<Mode>(m_matrix,_dest,m_perm.indices().data());
+    SparseMatrix<Scalar,(Options&RowMajor)==RowMajor ? ColMajor : RowMajor, DstIndex> tmp;
+    internal::permute_symm_to_fullsymm<Mode>(src.matrix(),tmp,src.perm().indices().data());
+    dst = tmp;
+  }
+  
+  template<typename DestType,unsigned int DestMode>
+  static void run(SparseSelfAdjointView<DestType,DestMode>& dst, const SrcXprType &src, const internal::assign_op<Scalar,typename MatrixType::Scalar> &)
+  {
+    internal::permute_symm_to_symm<Mode,DestMode>(src.matrix(),dst.matrix(),src.perm().indices().data());
+  }
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSE_SELFADJOINTVIEW_H
diff --git a/Eigen/src/SparseCore/SparseSolverBase.h b/Eigen/src/SparseCore/SparseSolverBase.h
new file mode 100644
index 0000000..b4c9a42
--- /dev/null
+++ b/Eigen/src/SparseCore/SparseSolverBase.h
@@ -0,0 +1,124 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2014 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_SPARSESOLVERBASE_H
+#define EIGEN_SPARSESOLVERBASE_H
+
+namespace Eigen { 
+
+namespace internal {
+
+  /** \internal
+  * Helper functions to solve with a sparse right-hand-side and result.
+  * The rhs is decomposed into small vertical panels which are solved through dense temporaries.
+  */
+template<typename Decomposition, typename Rhs, typename Dest>
+typename enable_if<Rhs::ColsAtCompileTime!=1 && Dest::ColsAtCompileTime!=1>::type
+solve_sparse_through_dense_panels(const Decomposition &dec, const Rhs& rhs, Dest &dest)
+{
+  EIGEN_STATIC_ASSERT((Dest::Flags&RowMajorBit)==0,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES);
+  typedef typename Dest::Scalar DestScalar;
+  // we process the sparse rhs per block of NbColsAtOnce columns temporarily stored into a dense matrix.
+  static const Index NbColsAtOnce = 4;
+  Index rhsCols = rhs.cols();
+  Index size = rhs.rows();
+  // the temporary matrices do not need more columns than NbColsAtOnce:
+  Index tmpCols = (std::min)(rhsCols, NbColsAtOnce); 
+  Eigen::Matrix<DestScalar,Dynamic,Dynamic> tmp(size,tmpCols);
+  Eigen::Matrix<DestScalar,Dynamic,Dynamic> tmpX(size,tmpCols);
+  for(Index k=0; k<rhsCols; k+=NbColsAtOnce)
+  {
+    Index actualCols = std::min<Index>(rhsCols-k, NbColsAtOnce);
+    tmp.leftCols(actualCols) = rhs.middleCols(k,actualCols);
+    tmpX.leftCols(actualCols) = dec.solve(tmp.leftCols(actualCols));
+    dest.middleCols(k,actualCols) = tmpX.leftCols(actualCols).sparseView();
+  }
+}
+
+// Overload for vector as rhs
+template<typename Decomposition, typename Rhs, typename Dest>
+typename enable_if<Rhs::ColsAtCompileTime==1 || Dest::ColsAtCompileTime==1>::type
+solve_sparse_through_dense_panels(const Decomposition &dec, const Rhs& rhs, Dest &dest)
+{
+  typedef typename Dest::Scalar DestScalar;
+  Index size = rhs.rows();
+  Eigen::Matrix<DestScalar,Dynamic,1> rhs_dense(rhs);
+  Eigen::Matrix<DestScalar,Dynamic,1> dest_dense(size);
+  dest_dense = dec.solve(rhs_dense);
+  dest = dest_dense.sparseView();
+}
+
+} // end namespace internal
+
+/** \class SparseSolverBase
+  * \ingroup SparseCore_Module
+  * \brief A base class for sparse solvers
+  *
+  * \tparam Derived the actual type of the solver.
+  *
+  */
+template<typename Derived>
+class SparseSolverBase : internal::noncopyable
+{
+  public:
+
+    /** Default constructor */
+    SparseSolverBase()
+      : m_isInitialized(false)
+    {}
+
+    ~SparseSolverBase()
+    {}
+
+    Derived& derived() { return *static_cast<Derived*>(this); }
+    const Derived& derived() const { return *static_cast<const Derived*>(this); }
+    
+    /** \returns an expression of the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      *
+      * \sa compute()
+      */
+    template<typename Rhs>
+    inline const Solve<Derived, Rhs>
+    solve(const MatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "Solver is not initialized.");
+      eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<Derived, Rhs>(derived(), b.derived());
+    }
+    
+    /** \returns an expression of the solution x of \f$ A x = b \f$ using the current decomposition of A.
+      *
+      * \sa compute()
+      */
+    template<typename Rhs>
+    inline const Solve<Derived, Rhs>
+    solve(const SparseMatrixBase<Rhs>& b) const
+    {
+      eigen_assert(m_isInitialized && "Solver is not initialized.");
+      eigen_assert(derived().rows()==b.rows() && "solve(): invalid number of rows of the right hand side matrix b");
+      return Solve<Derived, Rhs>(derived(), b.derived());
+    }
+    
+    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** \internal default implementation of solving with a sparse rhs */
+    template<typename Rhs,typename Dest>
+    void _solve_impl(const SparseMatrixBase<Rhs> &b, SparseMatrixBase<Dest> &dest) const
+    {
+      internal::solve_sparse_through_dense_panels(derived(), b.derived(), dest.derived());
+    }
+    #endif // EIGEN_PARSED_BY_DOXYGEN
+
+  protected:
+    
+    mutable bool m_isInitialized;
+};
+
+} // end namespace Eigen
+
+#endif // EIGEN_SPARSESOLVERBASE_H
diff --git a/Eigen/src/SparseCore/SparseSparseProductWithPruning.h b/Eigen/src/SparseCore/SparseSparseProductWithPruning.h
index fcc18f5..88820a4 100644
--- a/Eigen/src/SparseCore/SparseSparseProductWithPruning.h
+++ b/Eigen/src/SparseCore/SparseSparseProductWithPruning.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -21,8 +21,9 @@
 {
   // return sparse_sparse_product_with_pruning_impl2(lhs,rhs,res);
 
-  typedef typename remove_all<Lhs>::type::Scalar Scalar;
-  typedef typename remove_all<Lhs>::type::Index Index;
+  typedef typename remove_all<Rhs>::type::Scalar RhsScalar;
+  typedef typename remove_all<ResultType>::type::Scalar ResScalar;
+  typedef typename remove_all<Lhs>::type::StorageIndex StorageIndex;
 
   // make sure to call innerSize/outerSize since we fake the storage order.
   Index rows = lhs.innerSize();
@@ -31,24 +32,27 @@
   eigen_assert(lhs.outerSize() == rhs.innerSize());
 
   // allocate a temporary buffer
-  AmbiVector<Scalar,Index> tempVector(rows);
-
-  // estimate the number of non zero entries
-  // given a rhs column containing Y non zeros, we assume that the respective Y columns
-  // of the lhs differs in average of one non zeros, thus the number of non zeros for
-  // the product of a rhs column with the lhs is X+Y where X is the average number of non zero
-  // per column of the lhs.
-  // Therefore, we have nnz(lhs*rhs) = nnz(lhs) + nnz(rhs)
-  Index estimated_nnz_prod = lhs.nonZeros() + rhs.nonZeros();
+  AmbiVector<ResScalar,StorageIndex> tempVector(rows);
 
   // mimics a resizeByInnerOuter:
   if(ResultType::IsRowMajor)
     res.resize(cols, rows);
   else
     res.resize(rows, cols);
+  
+  evaluator<Lhs> lhsEval(lhs);
+  evaluator<Rhs> rhsEval(rhs);
+  
+  // estimate the number of non zero entries
+  // given a rhs column containing Y non zeros, we assume that the respective Y columns
+  // of the lhs differs in average of one non zeros, thus the number of non zeros for
+  // the product of a rhs column with the lhs is X+Y where X is the average number of non zero
+  // per column of the lhs.
+  // Therefore, we have nnz(lhs*rhs) = nnz(lhs) + nnz(rhs)
+  Index estimated_nnz_prod = lhsEval.nonZerosEstimate() + rhsEval.nonZerosEstimate();
 
   res.reserve(estimated_nnz_prod);
-  double ratioColRes = double(estimated_nnz_prod)/double(lhs.rows()*rhs.cols());
+  double ratioColRes = double(estimated_nnz_prod)/(double(lhs.rows())*double(rhs.cols()));
   for (Index j=0; j<cols; ++j)
   {
     // FIXME:
@@ -56,18 +60,18 @@
     // let's do a more accurate determination of the nnz ratio for the current column j of res
     tempVector.init(ratioColRes);
     tempVector.setZero();
-    for (typename Rhs::InnerIterator rhsIt(rhs, j); rhsIt; ++rhsIt)
+    for (typename evaluator<Rhs>::InnerIterator rhsIt(rhsEval, j); rhsIt; ++rhsIt)
     {
       // FIXME should be written like this: tmp += rhsIt.value() * lhs.col(rhsIt.index())
       tempVector.restart();
-      Scalar x = rhsIt.value();
-      for (typename Lhs::InnerIterator lhsIt(lhs, rhsIt.index()); lhsIt; ++lhsIt)
+      RhsScalar x = rhsIt.value();
+      for (typename evaluator<Lhs>::InnerIterator lhsIt(lhsEval, rhsIt.index()); lhsIt; ++lhsIt)
       {
         tempVector.coeffRef(lhsIt.index()) += lhsIt.value() * x;
       }
     }
     res.startVec(j);
-    for (typename AmbiVector<Scalar,Index>::Iterator it(tempVector,tolerance); it; ++it)
+    for (typename AmbiVector<ResScalar,StorageIndex>::Iterator it(tempVector,tolerance); it; ++it)
       res.insertBackByOuterInner(j,it.index()) = it.value();
   }
   res.finalize();
@@ -82,7 +86,6 @@
 template<typename Lhs, typename Rhs, typename ResultType>
 struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,ColMajor,ColMajor>
 {
-  typedef typename traits<typename remove_all<Lhs>::type>::Scalar Scalar;
   typedef typename ResultType::RealScalar RealScalar;
 
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
@@ -100,7 +103,7 @@
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
   {
     // we need a col-major matrix to hold the result
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::Index> SparseTemporaryType;
+    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename ResultType::StorageIndex> SparseTemporaryType;
     SparseTemporaryType _res(res.rows(), res.cols());
     internal::sparse_sparse_product_with_pruning_impl<Lhs,Rhs,SparseTemporaryType>(lhs, rhs, _res, tolerance);
     res = _res;
@@ -126,8 +129,8 @@
   typedef typename ResultType::RealScalar RealScalar;
   static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
   {
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename Lhs::Index> ColMajorMatrixLhs;
-    typedef SparseMatrix<typename ResultType::Scalar,ColMajor,typename Lhs::Index> ColMajorMatrixRhs;
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixLhs;
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixRhs;
     ColMajorMatrixLhs colLhs(lhs);
     ColMajorMatrixRhs colRhs(rhs);
     internal::sparse_sparse_product_with_pruning_impl<ColMajorMatrixLhs,ColMajorMatrixRhs,ResultType>(colLhs, colRhs, res, tolerance);
@@ -140,8 +143,53 @@
   }
 };
 
-// NOTE the 2 others cases (col row *) must never occur since they are caught
-// by ProductReturnType which transforms it to (col col *) by evaluating rhs.
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,RowMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Lhs::Scalar,RowMajor,typename Lhs::StorageIndex> RowMajorMatrixLhs;
+    RowMajorMatrixLhs rowLhs(lhs);
+    sparse_sparse_product_with_pruning_selector<RowMajorMatrixLhs,Rhs,ResultType,RowMajor,RowMajor>(rowLhs,rhs,res,tolerance);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,RowMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Rhs::Scalar,RowMajor,typename Lhs::StorageIndex> RowMajorMatrixRhs;
+    RowMajorMatrixRhs rowRhs(rhs);
+    sparse_sparse_product_with_pruning_selector<Lhs,RowMajorMatrixRhs,ResultType,RowMajor,RowMajor,RowMajor>(lhs,rowRhs,res,tolerance);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,ColMajor,RowMajor,ColMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Rhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixRhs;
+    ColMajorMatrixRhs colRhs(rhs);
+    internal::sparse_sparse_product_with_pruning_impl<Lhs,ColMajorMatrixRhs,ResultType>(lhs, colRhs, res, tolerance);
+  }
+};
+
+template<typename Lhs, typename Rhs, typename ResultType>
+struct sparse_sparse_product_with_pruning_selector<Lhs,Rhs,ResultType,RowMajor,ColMajor,ColMajor>
+{
+  typedef typename ResultType::RealScalar RealScalar;
+  static void run(const Lhs& lhs, const Rhs& rhs, ResultType& res, const RealScalar& tolerance)
+  {
+    typedef SparseMatrix<typename Lhs::Scalar,ColMajor,typename Lhs::StorageIndex> ColMajorMatrixLhs;
+    ColMajorMatrixLhs colLhs(lhs);
+    internal::sparse_sparse_product_with_pruning_impl<ColMajorMatrixLhs,Rhs,ResultType>(colLhs, rhs, res, tolerance);
+  }
+};
 
 } // end namespace internal
 
diff --git a/Eigen/src/SparseCore/SparseTranspose.h b/Eigen/src/SparseCore/SparseTranspose.h
index 76d031d..3757d4c 100644
--- a/Eigen/src/SparseCore/SparseTranspose.h
+++ b/Eigen/src/SparseCore/SparseTranspose.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -12,52 +12,81 @@
 
 namespace Eigen { 
 
-template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>
-  : public SparseMatrixBase<Transpose<MatrixType> >
-{
-    typedef typename internal::remove_all<typename MatrixType::Nested>::type _MatrixTypeNested;
+namespace internal {
+  template<typename MatrixType,int CompressedAccess=int(MatrixType::Flags&CompressedAccessBit)>
+  class SparseTransposeImpl
+    : public SparseMatrixBase<Transpose<MatrixType> >
+  {};
+  
+  template<typename MatrixType>
+  class SparseTransposeImpl<MatrixType,CompressedAccessBit>
+    : public SparseCompressedBase<Transpose<MatrixType> >
+  {
+    typedef SparseCompressedBase<Transpose<MatrixType> > Base;
   public:
-
-    EIGEN_SPARSE_PUBLIC_INTERFACE(Transpose<MatrixType> )
-
-    class InnerIterator;
-    class ReverseInnerIterator;
+    using Base::derived;
+    typedef typename Base::Scalar Scalar;
+    typedef typename Base::StorageIndex StorageIndex;
 
     inline Index nonZeros() const { return derived().nestedExpression().nonZeros(); }
-};
+    
+    inline const Scalar* valuePtr() const { return derived().nestedExpression().valuePtr(); }
+    inline const StorageIndex* innerIndexPtr() const { return derived().nestedExpression().innerIndexPtr(); }
+    inline const StorageIndex* outerIndexPtr() const { return derived().nestedExpression().outerIndexPtr(); }
+    inline const StorageIndex* innerNonZeroPtr() const { return derived().nestedExpression().innerNonZeroPtr(); }
 
-// NOTE: VC10 and VC11 trigger an ICE if don't put typename TransposeImpl<MatrixType,Sparse>:: in front of Index,
-// a typedef typename TransposeImpl<MatrixType,Sparse>::Index Index;
-// does not fix the issue.
-// An alternative is to define the nested class in the parent class itself.
-template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>::InnerIterator
-  : public _MatrixTypeNested::InnerIterator
+    inline Scalar* valuePtr() { return derived().nestedExpression().valuePtr(); }
+    inline StorageIndex* innerIndexPtr() { return derived().nestedExpression().innerIndexPtr(); }
+    inline StorageIndex* outerIndexPtr() { return derived().nestedExpression().outerIndexPtr(); }
+    inline StorageIndex* innerNonZeroPtr() { return derived().nestedExpression().innerNonZeroPtr(); }
+  };
+}
+  
+template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>
+  : public internal::SparseTransposeImpl<MatrixType>
 {
-    typedef typename _MatrixTypeNested::InnerIterator Base;
-    typedef typename TransposeImpl::Index Index;
-  public:
-
-    EIGEN_STRONG_INLINE InnerIterator(const TransposeImpl& trans, typename TransposeImpl<MatrixType,Sparse>::Index outer)
-      : Base(trans.derived().nestedExpression(), outer)
-    {}
-    typename TransposeImpl<MatrixType,Sparse>::Index row() const { return Base::col(); }
-    typename TransposeImpl<MatrixType,Sparse>::Index col() const { return Base::row(); }
+  protected:
+    typedef internal::SparseTransposeImpl<MatrixType> Base;
 };
 
-template<typename MatrixType> class TransposeImpl<MatrixType,Sparse>::ReverseInnerIterator
-  : public _MatrixTypeNested::ReverseInnerIterator
+namespace internal {
+  
+template<typename ArgType>
+struct unary_evaluator<Transpose<ArgType>, IteratorBased>
+  : public evaluator_base<Transpose<ArgType> >
 {
-    typedef typename _MatrixTypeNested::ReverseInnerIterator Base;
-    typedef typename TransposeImpl::Index Index;
+    typedef typename evaluator<ArgType>::InnerIterator        EvalIterator;
   public:
+    typedef Transpose<ArgType> XprType;
+    
+    inline Index nonZerosEstimate() const {
+      return m_argImpl.nonZerosEstimate();
+    }
 
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const TransposeImpl& xpr, typename TransposeImpl<MatrixType,Sparse>::Index outer)
-      : Base(xpr.derived().nestedExpression(), outer)
-    {}
-    typename TransposeImpl<MatrixType,Sparse>::Index row() const { return Base::col(); }
-    typename TransposeImpl<MatrixType,Sparse>::Index col() const { return Base::row(); }
+    class InnerIterator : public EvalIterator
+    {
+    public:
+      EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& unaryOp, Index outer)
+        : EvalIterator(unaryOp.m_argImpl,outer)
+      {}
+      
+      Index row() const { return EvalIterator::col(); }
+      Index col() const { return EvalIterator::row(); }
+    };
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& op) :m_argImpl(op.nestedExpression()) {}
+
+  protected:
+    evaluator<ArgType> m_argImpl;
 };
 
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif // EIGEN_SPARSETRANSPOSE_H
diff --git a/Eigen/src/SparseCore/SparseTriangularView.h b/Eigen/src/SparseCore/SparseTriangularView.h
index 333127b..9ac1202 100644
--- a/Eigen/src/SparseCore/SparseTriangularView.h
+++ b/Eigen/src/SparseCore/SparseTriangularView.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2009-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -11,19 +11,19 @@
 #ifndef EIGEN_SPARSE_TRIANGULARVIEW_H
 #define EIGEN_SPARSE_TRIANGULARVIEW_H
 
-namespace Eigen { 
+namespace Eigen {
 
-namespace internal {
-  
-template<typename MatrixType, int Mode>
-struct traits<SparseTriangularView<MatrixType,Mode> >
-: public traits<MatrixType>
-{};
-
-} // namespace internal
-
-template<typename MatrixType, int Mode> class SparseTriangularView
-  : public SparseMatrixBase<SparseTriangularView<MatrixType,Mode> >
+/** \ingroup SparseCore_Module
+  *
+  * \brief Base class for a triangular part in a \b sparse matrix
+  *
+  * This class is an abstract base class of class TriangularView, and objects of type TriangularViewImpl cannot be instantiated.
+  * It extends class TriangularView with additional methods which are available for sparse expressions only.
+  *
+  * \sa class TriangularView, SparseMatrixBase::triangularView()
+  */
+template<typename MatrixType, unsigned int Mode> class TriangularViewImpl<MatrixType,Mode,Sparse>
+  : public SparseMatrixBase<TriangularView<MatrixType,Mode> >
 {
     enum { SkipFirst = ((Mode&Lower) && !(MatrixType::Flags&RowMajorBit))
                     || ((Mode&Upper) &&  (MatrixType::Flags&RowMajorBit)),
@@ -31,147 +31,157 @@
            SkipDiag = (Mode&ZeroDiag) ? 1 : 0,
            HasUnitDiag = (Mode&UnitDiag) ? 1 : 0
     };
+    
+    typedef TriangularView<MatrixType,Mode> TriangularViewType;
+    
+  protected:
+    // dummy solve function to make TriangularView happy.
+    void solve() const;
 
+    typedef SparseMatrixBase<TriangularViewType> Base;
   public:
     
-    EIGEN_SPARSE_PUBLIC_INTERFACE(SparseTriangularView)
-
-    class InnerIterator;
-    class ReverseInnerIterator;
-
-    inline Index rows() const { return m_matrix.rows(); }
-    inline Index cols() const { return m_matrix.cols(); }
-
+    EIGEN_SPARSE_PUBLIC_INTERFACE(TriangularViewType)
+    
     typedef typename MatrixType::Nested MatrixTypeNested;
     typedef typename internal::remove_reference<MatrixTypeNested>::type MatrixTypeNestedNonRef;
     typedef typename internal::remove_all<MatrixTypeNested>::type MatrixTypeNestedCleaned;
 
-    inline SparseTriangularView(const MatrixType& matrix) : m_matrix(matrix) {}
-
-    /** \internal */
-    inline const MatrixTypeNestedCleaned& nestedExpression() const { return m_matrix; }
-
-    template<typename OtherDerived>
-    typename internal::plain_matrix_type_column_major<OtherDerived>::type
-    solve(const MatrixBase<OtherDerived>& other) const;
-
-    template<typename OtherDerived> void solveInPlace(MatrixBase<OtherDerived>& other) const;
-    template<typename OtherDerived> void solveInPlace(SparseMatrixBase<OtherDerived>& other) const;
-
-  protected:
-    MatrixTypeNested m_matrix;
-};
-
-template<typename MatrixType, int Mode>
-class SparseTriangularView<MatrixType,Mode>::InnerIterator : public MatrixTypeNestedCleaned::InnerIterator
-{
-    typedef typename MatrixTypeNestedCleaned::InnerIterator Base;
-    typedef typename SparseTriangularView::Index Index;
-  public:
-
-    EIGEN_STRONG_INLINE InnerIterator(const SparseTriangularView& view, Index outer)
-      : Base(view.nestedExpression(), outer), m_returnOne(false)
-    {
-      if(SkipFirst)
-      {
-        while((*this) && ((HasUnitDiag||SkipDiag)  ? this->index()<=outer : this->index()<outer))
-          Base::operator++();
-        if(HasUnitDiag)
-          m_returnOne = true;
-      }
-      else if(HasUnitDiag && ((!Base::operator bool()) || Base::index()>=Base::outer()))
-      {
-        if((!SkipFirst) && Base::operator bool())
-          Base::operator++();
-        m_returnOne = true;
-      }
+    template<typename RhsType, typename DstType>
+    EIGEN_DEVICE_FUNC
+    EIGEN_STRONG_INLINE void _solve_impl(const RhsType &rhs, DstType &dst) const {
+      if(!(internal::is_same<RhsType,DstType>::value && internal::extract_data(dst) == internal::extract_data(rhs)))
+        dst = rhs;
+      this->solveInPlace(dst);
     }
 
-    EIGEN_STRONG_INLINE InnerIterator& operator++()
-    {
-      if(HasUnitDiag && m_returnOne)
-        m_returnOne = false;
-      else
+    /** Applies the inverse of \c *this to the dense vector or matrix \a other, "in-place" */
+    template<typename OtherDerived> void solveInPlace(MatrixBase<OtherDerived>& other) const;
+
+    /** Applies the inverse of \c *this to the sparse vector or matrix \a other, "in-place" */
+    template<typename OtherDerived> void solveInPlace(SparseMatrixBase<OtherDerived>& other) const;
+  
+};
+
+namespace internal {
+
+template<typename ArgType, unsigned int Mode>
+struct unary_evaluator<TriangularView<ArgType,Mode>, IteratorBased>
+ : evaluator_base<TriangularView<ArgType,Mode> >
+{
+  typedef TriangularView<ArgType,Mode> XprType;
+  
+protected:
+  
+  typedef typename XprType::Scalar Scalar;
+  typedef typename XprType::StorageIndex StorageIndex;
+  typedef typename evaluator<ArgType>::InnerIterator EvalIterator;
+  
+  enum { SkipFirst = ((Mode&Lower) && !(ArgType::Flags&RowMajorBit))
+                    || ((Mode&Upper) &&  (ArgType::Flags&RowMajorBit)),
+         SkipLast = !SkipFirst,
+         SkipDiag = (Mode&ZeroDiag) ? 1 : 0,
+         HasUnitDiag = (Mode&UnitDiag) ? 1 : 0
+  };
+  
+public:
+  
+  enum {
+    CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+    Flags = XprType::Flags
+  };
+    
+  explicit unary_evaluator(const XprType &xpr) : m_argImpl(xpr.nestedExpression()), m_arg(xpr.nestedExpression()) {}
+  
+  inline Index nonZerosEstimate() const {
+    return m_argImpl.nonZerosEstimate();
+  }
+  
+  class InnerIterator : public EvalIterator
+  {
+      typedef EvalIterator Base;
+    public:
+
+      EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& xprEval, Index outer)
+        : Base(xprEval.m_argImpl,outer), m_returnOne(false), m_containsDiag(Base::outer()<xprEval.m_arg.innerSize())
       {
-        Base::operator++();
-        if(HasUnitDiag && (!SkipFirst) && ((!Base::operator bool()) || Base::index()>=Base::outer()))
+        if(SkipFirst)
+        {
+          while((*this) && ((HasUnitDiag||SkipDiag)  ? this->index()<=outer : this->index()<outer))
+            Base::operator++();
+          if(HasUnitDiag)
+            m_returnOne = m_containsDiag;
+        }
+        else if(HasUnitDiag && ((!Base::operator bool()) || Base::index()>=Base::outer()))
         {
           if((!SkipFirst) && Base::operator bool())
             Base::operator++();
-          m_returnOne = true;
+          m_returnOne = m_containsDiag;
         }
       }
-      return *this;
-    }
 
-    inline Index row() const { return (MatrixType::Flags&RowMajorBit ? Base::outer() : this->index()); }
-    inline Index col() const { return (MatrixType::Flags&RowMajorBit ? this->index() : Base::outer()); }
-    inline Index index() const
-    {
-      if(HasUnitDiag && m_returnOne)  return Base::outer();
-      else                            return Base::index();
-    }
-    inline Scalar value() const
-    {
-      if(HasUnitDiag && m_returnOne)  return Scalar(1);
-      else                            return Base::value();
-    }
-
-    EIGEN_STRONG_INLINE operator bool() const
-    {
-      if(HasUnitDiag && m_returnOne)
-        return true;
-      if(SkipFirst) return  Base::operator bool();
-      else
+      EIGEN_STRONG_INLINE InnerIterator& operator++()
       {
-        if (SkipDiag) return (Base::operator bool() && this->index() < this->outer());
-        else return (Base::operator bool() && this->index() <= this->outer());
+        if(HasUnitDiag && m_returnOne)
+          m_returnOne = false;
+        else
+        {
+          Base::operator++();
+          if(HasUnitDiag && (!SkipFirst) && ((!Base::operator bool()) || Base::index()>=Base::outer()))
+          {
+            if((!SkipFirst) && Base::operator bool())
+              Base::operator++();
+            m_returnOne = m_containsDiag;
+          }
+        }
+        return *this;
       }
-    }
-  protected:
-    bool m_returnOne;
+      
+      EIGEN_STRONG_INLINE operator bool() const
+      {
+        if(HasUnitDiag && m_returnOne)
+          return true;
+        if(SkipFirst) return  Base::operator bool();
+        else
+        {
+          if (SkipDiag) return (Base::operator bool() && this->index() < this->outer());
+          else return (Base::operator bool() && this->index() <= this->outer());
+        }
+      }
+
+//       inline Index row() const { return (ArgType::Flags&RowMajorBit ? Base::outer() : this->index()); }
+//       inline Index col() const { return (ArgType::Flags&RowMajorBit ? this->index() : Base::outer()); }
+      inline StorageIndex index() const
+      {
+        if(HasUnitDiag && m_returnOne)  return internal::convert_index<StorageIndex>(Base::outer());
+        else                            return Base::index();
+      }
+      inline Scalar value() const
+      {
+        if(HasUnitDiag && m_returnOne)  return Scalar(1);
+        else                            return Base::value();
+      }
+
+    protected:
+      bool m_returnOne;
+      bool m_containsDiag;
+    private:
+      Scalar& valueRef();
+  };
+  
+protected:
+  evaluator<ArgType> m_argImpl;
+  const ArgType& m_arg;
 };
 
-template<typename MatrixType, int Mode>
-class SparseTriangularView<MatrixType,Mode>::ReverseInnerIterator : public MatrixTypeNestedCleaned::ReverseInnerIterator
-{
-    typedef typename MatrixTypeNestedCleaned::ReverseInnerIterator Base;
-    typedef typename SparseTriangularView::Index Index;
-  public:
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator(const SparseTriangularView& view, Index outer)
-      : Base(view.nestedExpression(), outer)
-    {
-      eigen_assert((!HasUnitDiag) && "ReverseInnerIterator does not support yet triangular views with a unit diagonal");
-      if(SkipLast) {
-        while((*this) && (SkipDiag ? this->index()>=outer : this->index()>outer))
-          --(*this);
-      }
-    }
-
-    EIGEN_STRONG_INLINE ReverseInnerIterator& operator--()
-    { Base::operator--(); return *this; }
-
-    inline Index row() const { return Base::row(); }
-    inline Index col() const { return Base::col(); }
-
-    EIGEN_STRONG_INLINE operator bool() const
-    {
-      if (SkipLast) return Base::operator bool() ;
-      else
-      {
-        if(SkipDiag) return (Base::operator bool() && this->index() > this->outer());
-        else return (Base::operator bool() && this->index() >= this->outer());
-      }
-    }
-};
+} // end namespace internal
 
 template<typename Derived>
 template<int Mode>
-inline const SparseTriangularView<Derived, Mode>
+inline const TriangularView<const Derived, Mode>
 SparseMatrixBase<Derived>::triangularView() const
 {
-  return derived();
+  return TriangularView<const Derived, Mode>(derived());
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/SparseCore/SparseUtil.h b/Eigen/src/SparseCore/SparseUtil.h
index 0ba4713..74df0d4 100644
--- a/Eigen/src/SparseCore/SparseUtil.h
+++ b/Eigen/src/SparseCore/SparseUtil.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -37,43 +37,23 @@
 }
 
 #define EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATORS(Derived) \
-EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, =) \
-EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, +=) \
-EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, -=) \
-EIGEN_SPARSE_INHERIT_SCALAR_ASSIGNMENT_OPERATOR(Derived, *=) \
-EIGEN_SPARSE_INHERIT_SCALAR_ASSIGNMENT_OPERATOR(Derived, /=)
+EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(Derived, =)
 
-#define _EIGEN_SPARSE_PUBLIC_INTERFACE(Derived, BaseClass) \
-  typedef BaseClass Base; \
-  typedef typename Eigen::internal::traits<Derived >::Scalar Scalar; \
-  typedef typename Eigen::NumTraits<Scalar>::Real RealScalar; \
-  typedef typename Eigen::internal::nested<Derived >::type Nested; \
-  typedef typename Eigen::internal::traits<Derived >::StorageKind StorageKind; \
-  typedef typename Eigen::internal::traits<Derived >::Index Index; \
-  enum { RowsAtCompileTime = Eigen::internal::traits<Derived >::RowsAtCompileTime, \
-        ColsAtCompileTime = Eigen::internal::traits<Derived >::ColsAtCompileTime, \
-        Flags = Eigen::internal::traits<Derived >::Flags, \
-        CoeffReadCost = Eigen::internal::traits<Derived >::CoeffReadCost, \
-        SizeAtCompileTime = Base::SizeAtCompileTime, \
-        IsVectorAtCompileTime = Base::IsVectorAtCompileTime }; \
-  using Base::derived; \
-  using Base::const_cast_derived;
 
 #define EIGEN_SPARSE_PUBLIC_INTERFACE(Derived) \
-  _EIGEN_SPARSE_PUBLIC_INTERFACE(Derived, Eigen::SparseMatrixBase<Derived >)
+  EIGEN_GENERIC_PUBLIC_INTERFACE(Derived)
 
+  
 const int CoherentAccessPattern     = 0x1;
 const int InnerRandomAccessPattern  = 0x2 | CoherentAccessPattern;
 const int OuterRandomAccessPattern  = 0x4 | CoherentAccessPattern;
 const int RandomAccessPattern       = 0x8 | OuterRandomAccessPattern | InnerRandomAccessPattern;
 
-template<typename Derived> class SparseMatrixBase;
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class SparseMatrix;
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class DynamicSparseMatrix;
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class SparseVector;
-template<typename _Scalar, int _Flags = 0, typename _Index = int>  class MappedSparseMatrix;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class SparseMatrix;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class DynamicSparseMatrix;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class SparseVector;
+template<typename _Scalar, int _Flags = 0, typename _StorageIndex = int>  class MappedSparseMatrix;
 
-template<typename MatrixType, int Mode>           class SparseTriangularView;
 template<typename MatrixType, unsigned int UpLo>  class SparseSelfAdjointView;
 template<typename Lhs, typename Rhs>              class SparseDiagonalProduct;
 template<typename MatrixType> class SparseView;
@@ -85,42 +65,44 @@
 
 template<typename Lhs, typename Rhs> struct SparseSparseProductReturnType;
 template<typename Lhs, typename Rhs,
-         int InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(internal::traits<Lhs>::ColsAtCompileTime,internal::traits<Rhs>::RowsAtCompileTime)> struct DenseSparseProductReturnType;         
+         int InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(internal::traits<Lhs>::ColsAtCompileTime,internal::traits<Rhs>::RowsAtCompileTime)> struct DenseSparseProductReturnType;
+         
 template<typename Lhs, typename Rhs,
          int InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(internal::traits<Lhs>::ColsAtCompileTime,internal::traits<Rhs>::RowsAtCompileTime)> struct SparseDenseProductReturnType;
 template<typename MatrixType,int UpLo> class SparseSymmetricPermutationProduct;
 
 namespace internal {
 
-template<typename T,int Rows,int Cols> struct sparse_eval;
+template<typename T,int Rows,int Cols,int Flags> struct sparse_eval;
 
 template<typename T> struct eval<T,Sparse>
-  : public sparse_eval<T, traits<T>::RowsAtCompileTime,traits<T>::ColsAtCompileTime>
+  : sparse_eval<T, traits<T>::RowsAtCompileTime,traits<T>::ColsAtCompileTime,traits<T>::Flags>
 {};
 
-template<typename T,int Cols> struct sparse_eval<T,1,Cols> {
+template<typename T,int Cols,int Flags> struct sparse_eval<T,1,Cols,Flags> {
     typedef typename traits<T>::Scalar _Scalar;
-    typedef typename traits<T>::Index _Index;
+    typedef typename traits<T>::StorageIndex _StorageIndex;
   public:
-    typedef SparseVector<_Scalar, RowMajor, _Index> type;
+    typedef SparseVector<_Scalar, RowMajor, _StorageIndex> type;
 };
 
-template<typename T,int Rows> struct sparse_eval<T,Rows,1> {
+template<typename T,int Rows,int Flags> struct sparse_eval<T,Rows,1,Flags> {
     typedef typename traits<T>::Scalar _Scalar;
-    typedef typename traits<T>::Index _Index;
+    typedef typename traits<T>::StorageIndex _StorageIndex;
   public:
-    typedef SparseVector<_Scalar, ColMajor, _Index> type;
+    typedef SparseVector<_Scalar, ColMajor, _StorageIndex> type;
 };
 
-template<typename T,int Rows,int Cols> struct sparse_eval {
+// TODO this seems almost identical to plain_matrix_type<T, Sparse>
+template<typename T,int Rows,int Cols,int Flags> struct sparse_eval {
     typedef typename traits<T>::Scalar _Scalar;
-    typedef typename traits<T>::Index _Index;
-    enum { _Options = ((traits<T>::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
+    typedef typename traits<T>::StorageIndex _StorageIndex;
+    enum { _Options = ((Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
   public:
-    typedef SparseMatrix<_Scalar, _Options, _Index> type;
+    typedef SparseMatrix<_Scalar, _Options, _StorageIndex> type;
 };
 
-template<typename T> struct sparse_eval<T,1,1> {
+template<typename T,int Flags> struct sparse_eval<T,1,1,Flags> {
     typedef typename traits<T>::Scalar _Scalar;
   public:
     typedef Matrix<_Scalar, 1, 1> type;
@@ -129,12 +111,35 @@
 template<typename T> struct plain_matrix_type<T,Sparse>
 {
   typedef typename traits<T>::Scalar _Scalar;
-  typedef typename traits<T>::Index _Index;
-  enum { _Options = ((traits<T>::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
+  typedef typename traits<T>::StorageIndex _StorageIndex;
+  enum { _Options = ((evaluator<T>::Flags&RowMajorBit)==RowMajorBit) ? RowMajor : ColMajor };
   public:
-    typedef SparseMatrix<_Scalar, _Options, _Index> type;
+    typedef SparseMatrix<_Scalar, _Options, _StorageIndex> type;
 };
 
+template<typename T>
+struct plain_object_eval<T,Sparse>
+  : sparse_eval<T, traits<T>::RowsAtCompileTime,traits<T>::ColsAtCompileTime, evaluator<T>::Flags>
+{};
+
+template<typename Decomposition, typename RhsType>
+struct solve_traits<Decomposition,RhsType,Sparse>
+{
+  typedef typename sparse_eval<RhsType, RhsType::RowsAtCompileTime, RhsType::ColsAtCompileTime,traits<RhsType>::Flags>::type PlainObject;
+};
+
+template<typename Derived>
+struct generic_xpr_base<Derived, MatrixXpr, Sparse>
+{
+  typedef SparseMatrixBase<Derived> type;
+};
+
+struct SparseTriangularShape  { static std::string debugName() { return "SparseTriangularShape"; } };
+struct SparseSelfAdjointShape { static std::string debugName() { return "SparseSelfAdjointShape"; } };
+
+template<> struct glue_shapes<SparseShape,SelfAdjointShape> { typedef SparseSelfAdjointShape type;  };
+template<> struct glue_shapes<SparseShape,TriangularShape > { typedef SparseTriangularShape  type;  };
+
 } // end namespace internal
 
 /** \ingroup SparseCore_Module
@@ -145,26 +150,26 @@
   *
   * \sa SparseMatrix::setFromTriplets()
   */
-template<typename Scalar, typename Index=typename SparseMatrix<Scalar>::Index >
+template<typename Scalar, typename StorageIndex=typename SparseMatrix<Scalar>::StorageIndex >
 class Triplet
 {
 public:
   Triplet() : m_row(0), m_col(0), m_value(0) {}
 
-  Triplet(const Index& i, const Index& j, const Scalar& v = Scalar(0))
+  Triplet(const StorageIndex& i, const StorageIndex& j, const Scalar& v = Scalar(0))
     : m_row(i), m_col(j), m_value(v)
   {}
 
   /** \returns the row index of the element */
-  const Index& row() const { return m_row; }
+  const StorageIndex& row() const { return m_row; }
 
   /** \returns the column index of the element */
-  const Index& col() const { return m_col; }
+  const StorageIndex& col() const { return m_col; }
 
   /** \returns the value of the element */
   const Scalar& value() const { return m_value; }
 protected:
-  Index m_row, m_col;
+  StorageIndex m_row, m_col;
   Scalar m_value;
 };
 
diff --git a/Eigen/src/SparseCore/SparseVector.h b/Eigen/src/SparseCore/SparseVector.h
index 7e15c81..19b0fbc 100644
--- a/Eigen/src/SparseCore/SparseVector.h
+++ b/Eigen/src/SparseCore/SparseVector.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -22,15 +22,15 @@
   * See http://www.netlib.org/linalg/html_templates/node91.html for details on the storage scheme.
   *
   * This class can be extended with the help of the plugin mechanism described on the page
-  * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_SPARSEVECTOR_PLUGIN.
+  * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_SPARSEVECTOR_PLUGIN.
   */
 
 namespace internal {
-template<typename _Scalar, int _Options, typename _Index>
-struct traits<SparseVector<_Scalar, _Options, _Index> >
+template<typename _Scalar, int _Options, typename _StorageIndex>
+struct traits<SparseVector<_Scalar, _Options, _StorageIndex> >
 {
   typedef _Scalar Scalar;
-  typedef _Index Index;
+  typedef _StorageIndex StorageIndex;
   typedef Sparse StorageKind;
   typedef MatrixXpr XprKind;
   enum {
@@ -40,8 +40,7 @@
     ColsAtCompileTime = IsColVector ? 1 : Dynamic,
     MaxRowsAtCompileTime = RowsAtCompileTime,
     MaxColsAtCompileTime = ColsAtCompileTime,
-    Flags = _Options | NestByRefBit | LvalueBit | (IsColVector ? 0 : RowMajorBit),
-    CoeffReadCost = NumTraits<Scalar>::ReadCost,
+    Flags = _Options | NestByRefBit | LvalueBit | (IsColVector ? 0 : RowMajorBit) | CompressedAccessBit,
     SupportedAccessPatterns = InnerRandomAccessPattern
   };
 };
@@ -61,18 +60,18 @@
 
 }
 
-template<typename _Scalar, int _Options, typename _Index>
+template<typename _Scalar, int _Options, typename _StorageIndex>
 class SparseVector
-  : public SparseMatrixBase<SparseVector<_Scalar, _Options, _Index> >
+  : public SparseCompressedBase<SparseVector<_Scalar, _Options, _StorageIndex> >
 {
-    typedef SparseMatrixBase<SparseVector> SparseBase;
-    
+    typedef SparseCompressedBase<SparseVector> Base;
+    using Base::convert_index;
   public:
     EIGEN_SPARSE_PUBLIC_INTERFACE(SparseVector)
     EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseVector, +=)
     EIGEN_SPARSE_INHERIT_ASSIGNMENT_OPERATOR(SparseVector, -=)
     
-    typedef internal::CompressedStorage<Scalar,Index> Storage;
+    typedef internal::CompressedStorage<Scalar,StorageIndex> Storage;
     enum { IsColVector = internal::traits<SparseVector>::IsColVector };
     
     enum {
@@ -84,11 +83,16 @@
     EIGEN_STRONG_INLINE Index innerSize() const { return m_size; }
     EIGEN_STRONG_INLINE Index outerSize() const { return 1; }
 
-    EIGEN_STRONG_INLINE const Scalar* valuePtr() const { return &m_data.value(0); }
-    EIGEN_STRONG_INLINE Scalar* valuePtr() { return &m_data.value(0); }
+    EIGEN_STRONG_INLINE const Scalar* valuePtr() const { return m_data.valuePtr(); }
+    EIGEN_STRONG_INLINE Scalar* valuePtr() { return m_data.valuePtr(); }
 
-    EIGEN_STRONG_INLINE const Index* innerIndexPtr() const { return &m_data.index(0); }
-    EIGEN_STRONG_INLINE Index* innerIndexPtr() { return &m_data.index(0); }
+    EIGEN_STRONG_INLINE const StorageIndex* innerIndexPtr() const { return m_data.indexPtr(); }
+    EIGEN_STRONG_INLINE StorageIndex* innerIndexPtr() { return m_data.indexPtr(); }
+
+    inline const StorageIndex* outerIndexPtr() const { return 0; }
+    inline StorageIndex* outerIndexPtr() { return 0; }
+    inline const StorageIndex* innerNonZeroPtr() const { return 0; }
+    inline StorageIndex* innerNonZeroPtr() { return 0; }
     
     /** \internal */
     inline Storage& data() { return m_data; }
@@ -103,13 +107,13 @@
     inline Scalar coeff(Index i) const
     {
       eigen_assert(i>=0 && i<m_size);
-      return m_data.at(i);
+      return m_data.at(StorageIndex(i));
     }
 
     inline Scalar& coeffRef(Index row, Index col)
     {
       eigen_assert(IsColVector ? (col==0 && row>=0 && row<m_size) : (row==0 && col>=0 && col<m_size));
-      return coeff(IsColVector ? row : col);
+      return coeffRef(IsColVector ? row : col);
     }
 
     /** \returns a reference to the coefficient value at given index \a i
@@ -121,18 +125,19 @@
     inline Scalar& coeffRef(Index i)
     {
       eigen_assert(i>=0 && i<m_size);
-      return m_data.atWithInsertion(i);
+
+      return m_data.atWithInsertion(StorageIndex(i));
     }
 
   public:
 
-    class InnerIterator;
-    class ReverseInnerIterator;
+    typedef typename Base::InnerIterator InnerIterator;
+    typedef typename Base::ReverseInnerIterator ReverseInnerIterator;
 
     inline void setZero() { m_data.clear(); }
 
     /** \returns the number of non zero coefficients */
-    inline Index nonZeros() const  { return static_cast<Index>(m_data.size()); }
+    inline Index nonZeros() const  { return m_data.size(); }
 
     inline void startVec(Index outer)
     {
@@ -151,6 +156,18 @@
       m_data.append(0, i);
       return m_data.value(m_data.size()-1);
     }
+    
+    Scalar& insertBackByOuterInnerUnordered(Index outer, Index inner)
+    {
+      EIGEN_UNUSED_VARIABLE(outer);
+      eigen_assert(outer==0);
+      return insertBackUnordered(inner);
+    }
+    inline Scalar& insertBackUnordered(Index i)
+    {
+      m_data.append(0, i);
+      return m_data.value(m_data.size()-1);
+    }
 
     inline Scalar& insert(Index row, Index col)
     {
@@ -158,6 +175,7 @@
       
       Index inner = IsColVector ? row : col;
       Index outer = IsColVector ? col : row;
+      EIGEN_ONLY_USED_FOR_DEBUG(outer);
       eigen_assert(outer==0);
       return insert(inner);
     }
@@ -176,7 +194,7 @@
         m_data.value(p+1) = m_data.value(p);
         --p;
       }
-      m_data.index(p+1) = i;
+      m_data.index(p+1) = convert_index(i);
       m_data.value(p+1) = 0;
       return m_data.value(p+1);
     }
@@ -188,28 +206,59 @@
 
     inline void finalize() {}
 
+    /** \copydoc SparseMatrix::prune(const Scalar&,const RealScalar&) */
     void prune(const Scalar& reference, const RealScalar& epsilon = NumTraits<RealScalar>::dummy_precision())
     {
       m_data.prune(reference,epsilon);
     }
 
+    /** Resizes the sparse vector to \a rows x \a cols
+      *
+      * This method is provided for compatibility with matrices.
+      * For a column vector, \a cols must be equal to 1.
+      * For a row vector, \a rows must be equal to 1.
+      *
+      * \sa resize(Index)
+      */
     void resize(Index rows, Index cols)
     {
-      eigen_assert(rows==1 || cols==1);
+      eigen_assert((IsColVector ? cols : rows)==1 && "Outer dimension must equal 1");
       resize(IsColVector ? rows : cols);
     }
 
+    /** Resizes the sparse vector to \a newSize
+      * This method deletes all entries, thus leaving an empty sparse vector
+      *
+      * \sa  conservativeResize(), setZero() */
     void resize(Index newSize)
     {
       m_size = newSize;
       m_data.clear();
     }
 
+    /** Resizes the sparse vector to \a newSize, while leaving old values untouched.
+      *
+      * If the size of the vector is decreased, then the storage of the out-of bounds coefficients is kept and reserved.
+      * Call .data().squeeze() to free extra memory.
+      *
+      * \sa reserve(), setZero()
+      */
+    void conservativeResize(Index newSize)
+    {
+      if (newSize < m_size)
+      {
+        Index i = 0;
+        while (i<m_data.size() && m_data.index(i)<newSize) ++i;
+        m_data.resize(i);
+      }
+      m_size = newSize;
+    }
+
     void resizeNonZeros(Index size) { m_data.resize(size); }
 
     inline SparseVector() : m_size(0) { check_template_parameters(); resize(0); }
 
-    inline SparseVector(Index size) : m_size(0) { check_template_parameters(); resize(size); }
+    explicit inline SparseVector(Index size) : m_size(0) { check_template_parameters(); resize(size); }
 
     inline SparseVector(Index rows, Index cols) : m_size(0) { check_template_parameters(); resize(rows,cols); }
 
@@ -217,12 +266,15 @@
     inline SparseVector(const SparseMatrixBase<OtherDerived>& other)
       : m_size(0)
     {
+      #ifdef EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+        EIGEN_SPARSE_CREATE_TEMPORARY_PLUGIN
+      #endif
       check_template_parameters();
       *this = other.derived();
     }
 
     inline SparseVector(const SparseVector& other)
-      : SparseBase(other), m_size(0)
+      : Base(other), m_size(0)
     {
       check_template_parameters();
       *this = other.derived();
@@ -238,6 +290,14 @@
       m_data.swap(other.m_data);
     }
 
+    template<int OtherOptions>
+    inline void swap(SparseMatrix<Scalar,OtherOptions,StorageIndex>& other)
+    {
+      eigen_assert(other.outerSize()==1);
+      std::swap(m_size, other.m_innerSize);
+      m_data.swap(other.m_data);
+    }
+
     inline SparseVector& operator=(const SparseVector& other)
     {
       if (other.isRValue())
@@ -336,7 +396,7 @@
   
     static void check_template_parameters()
     {
-      EIGEN_STATIC_ASSERT(NumTraits<Index>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
+      EIGEN_STATIC_ASSERT(NumTraits<StorageIndex>::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE);
       EIGEN_STATIC_ASSERT((_Options&(ColMajor|RowMajor))==Options,INVALID_MATRIX_TEMPLATE_PARAMETERS);
     }
     
@@ -344,77 +404,46 @@
     Index m_size;
 };
 
-template<typename Scalar, int _Options, typename _Index>
-class SparseVector<Scalar,_Options,_Index>::InnerIterator
-{
-  public:
-    InnerIterator(const SparseVector& vec, Index outer=0)
-      : m_data(vec.m_data), m_id(0), m_end(static_cast<Index>(m_data.size()))
-    {
-      EIGEN_UNUSED_VARIABLE(outer);
-      eigen_assert(outer==0);
-    }
-
-    InnerIterator(const internal::CompressedStorage<Scalar,Index>& data)
-      : m_data(data), m_id(0), m_end(static_cast<Index>(m_data.size()))
-    {}
-
-    inline InnerIterator& operator++() { m_id++; return *this; }
-
-    inline Scalar value() const { return m_data.value(m_id); }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_data.value(m_id)); }
-
-    inline Index index() const { return m_data.index(m_id); }
-    inline Index row() const { return IsColVector ? index() : 0; }
-    inline Index col() const { return IsColVector ? 0 : index(); }
-
-    inline operator bool() const { return (m_id < m_end); }
-
-  protected:
-    const internal::CompressedStorage<Scalar,Index>& m_data;
-    Index m_id;
-    const Index m_end;
-};
-
-template<typename Scalar, int _Options, typename _Index>
-class SparseVector<Scalar,_Options,_Index>::ReverseInnerIterator
-{
-  public:
-    ReverseInnerIterator(const SparseVector& vec, Index outer=0)
-      : m_data(vec.m_data), m_id(static_cast<Index>(m_data.size())), m_start(0)
-    {
-      EIGEN_UNUSED_VARIABLE(outer);
-      eigen_assert(outer==0);
-    }
-
-    ReverseInnerIterator(const internal::CompressedStorage<Scalar,Index>& data)
-      : m_data(data), m_id(static_cast<Index>(m_data.size())), m_start(0)
-    {}
-
-    inline ReverseInnerIterator& operator--() { m_id--; return *this; }
-
-    inline Scalar value() const { return m_data.value(m_id-1); }
-    inline Scalar& valueRef() { return const_cast<Scalar&>(m_data.value(m_id-1)); }
-
-    inline Index index() const { return m_data.index(m_id-1); }
-    inline Index row() const { return IsColVector ? index() : 0; }
-    inline Index col() const { return IsColVector ? 0 : index(); }
-
-    inline operator bool() const { return (m_id > m_start); }
-
-  protected:
-    const internal::CompressedStorage<Scalar,Index>& m_data;
-    Index m_id;
-    const Index m_start;
-};
-
 namespace internal {
 
+template<typename _Scalar, int _Options, typename _Index>
+struct evaluator<SparseVector<_Scalar,_Options,_Index> >
+  : evaluator_base<SparseVector<_Scalar,_Options,_Index> >
+{
+  typedef SparseVector<_Scalar,_Options,_Index> SparseVectorType;
+  typedef evaluator_base<SparseVectorType> Base;
+  typedef typename SparseVectorType::InnerIterator InnerIterator;
+  typedef typename SparseVectorType::ReverseInnerIterator ReverseInnerIterator;
+  
+  enum {
+    CoeffReadCost = NumTraits<_Scalar>::ReadCost,
+    Flags = SparseVectorType::Flags
+  };
+
+  evaluator() : Base() {}
+  
+  explicit evaluator(const SparseVectorType &mat) : m_matrix(&mat)
+  {
+    EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost);
+  }
+  
+  inline Index nonZerosEstimate() const {
+    return m_matrix->nonZeros();
+  }
+  
+  operator SparseVectorType&() { return m_matrix->const_cast_derived(); }
+  operator const SparseVectorType&() const { return *m_matrix; }
+  
+  const SparseVectorType *m_matrix;
+};
+
 template< typename Dest, typename Src>
 struct sparse_vector_assign_selector<Dest,Src,SVA_Inner> {
   static void run(Dest& dst, const Src& src) {
     eigen_internal_assert(src.innerSize()==src.size());
-    for(typename Src::InnerIterator it(src, 0); it; ++it)
+    typedef internal::evaluator<Src> SrcEvaluatorType;
+    SrcEvaluatorType srcEval(src);
+    for(typename SrcEvaluatorType::InnerIterator it(srcEval, 0); it; ++it)
       dst.insert(it.index()) = it.value();
   }
 };
@@ -423,9 +452,11 @@
 struct sparse_vector_assign_selector<Dest,Src,SVA_Outer> {
   static void run(Dest& dst, const Src& src) {
     eigen_internal_assert(src.outerSize()==src.size());
-    for(typename Dest::Index i=0; i<src.size(); ++i)
+    typedef internal::evaluator<Src> SrcEvaluatorType;
+    SrcEvaluatorType srcEval(src);
+    for(Index i=0; i<src.size(); ++i)
     {
-      typename Src::InnerIterator it(src, i);
+      typename SrcEvaluatorType::InnerIterator it(srcEval, i);
       if(it)
         dst.insert(i) = it.value();
     }
diff --git a/Eigen/src/SparseCore/SparseView.h b/Eigen/src/SparseCore/SparseView.h
index fd84504..7c4aea7 100644
--- a/Eigen/src/SparseCore/SparseView.h
+++ b/Eigen/src/SparseCore/SparseView.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2011-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2010 Daniel Lowengrub <lowdanie@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -18,7 +18,7 @@
 template<typename MatrixType>
 struct traits<SparseView<MatrixType> > : traits<MatrixType>
 {
-  typedef typename MatrixType::Index Index;
+  typedef typename MatrixType::StorageIndex StorageIndex;
   typedef Sparse StorageKind;
   enum {
     Flags = int(traits<MatrixType>::Flags) & (RowMajorBit)
@@ -27,71 +27,225 @@
 
 } // end namespace internal
 
+/** \ingroup SparseCore_Module
+  * \class SparseView
+  *
+  * \brief Expression of a dense or sparse matrix with zero or too small values removed
+  *
+  * \tparam MatrixType the type of the object of which we are removing the small entries
+  *
+  * This class represents an expression of a given dense or sparse matrix with
+  * entries smaller than \c reference * \c epsilon are removed.
+  * It is the return type of MatrixBase::sparseView() and SparseMatrixBase::pruned()
+  * and most of the time this is the only way it is used.
+  *
+  * \sa MatrixBase::sparseView(), SparseMatrixBase::pruned()
+  */
 template<typename MatrixType>
 class SparseView : public SparseMatrixBase<SparseView<MatrixType> >
 {
   typedef typename MatrixType::Nested MatrixTypeNested;
   typedef typename internal::remove_all<MatrixTypeNested>::type _MatrixTypeNested;
+  typedef SparseMatrixBase<SparseView > Base;
 public:
   EIGEN_SPARSE_PUBLIC_INTERFACE(SparseView)
+  typedef typename internal::remove_all<MatrixType>::type NestedExpression;
 
-  SparseView(const MatrixType& mat, const Scalar& m_reference = Scalar(0),
-             typename NumTraits<Scalar>::Real m_epsilon = NumTraits<Scalar>::dummy_precision()) : 
-    m_matrix(mat), m_reference(m_reference), m_epsilon(m_epsilon) {}
-
-  class InnerIterator;
+  explicit SparseView(const MatrixType& mat, const Scalar& reference = Scalar(0),
+                      const RealScalar &epsilon = NumTraits<Scalar>::dummy_precision())
+    : m_matrix(mat), m_reference(reference), m_epsilon(epsilon) {}
 
   inline Index rows() const { return m_matrix.rows(); }
   inline Index cols() const { return m_matrix.cols(); }
 
   inline Index innerSize() const { return m_matrix.innerSize(); }
   inline Index outerSize() const { return m_matrix.outerSize(); }
-
+  
+  /** \returns the nested expression */
+  const typename internal::remove_all<MatrixTypeNested>::type&
+  nestedExpression() const { return m_matrix; }
+  
+  Scalar reference() const { return m_reference; }
+  RealScalar epsilon() const { return m_epsilon; }
+  
 protected:
   MatrixTypeNested m_matrix;
   Scalar m_reference;
-  typename NumTraits<Scalar>::Real m_epsilon;
+  RealScalar m_epsilon;
 };
 
-template<typename MatrixType>
-class SparseView<MatrixType>::InnerIterator : public _MatrixTypeNested::InnerIterator
+namespace internal {
+
+// TODO find a way to unify the two following variants
+// This is tricky because implementing an inner iterator on top of an IndexBased evaluator is
+// not easy because the evaluators do not expose the sizes of the underlying expression.
+  
+template<typename ArgType>
+struct unary_evaluator<SparseView<ArgType>, IteratorBased>
+  : public evaluator_base<SparseView<ArgType> >
 {
-  typedef typename SparseView::Index Index;
-public:
-  typedef typename _MatrixTypeNested::InnerIterator IterBase;
-  InnerIterator(const SparseView& view, Index outer) :
-  IterBase(view.m_matrix, outer), m_view(view)
-  {
-    incrementToNonZero();
-  }
-
-  EIGEN_STRONG_INLINE InnerIterator& operator++()
-  {
-    IterBase::operator++();
-    incrementToNonZero();
-    return *this;
-  }
-
-  using IterBase::value;
-
-protected:
-  const SparseView& m_view;
-
-private:
-  void incrementToNonZero()
-  {
-    while((bool(*this)) && internal::isMuchSmallerThan(value(), m_view.m_reference, m_view.m_epsilon))
+    typedef typename evaluator<ArgType>::InnerIterator EvalIterator;
+  public:
+    typedef SparseView<ArgType> XprType;
+    
+    class InnerIterator : public EvalIterator
     {
-      IterBase::operator++();
-    }
-  }
+        typedef typename XprType::Scalar Scalar;
+      public:
+
+        EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& sve, Index outer)
+          : EvalIterator(sve.m_argImpl,outer), m_view(sve.m_view)
+        {
+          incrementToNonZero();
+        }
+
+        EIGEN_STRONG_INLINE InnerIterator& operator++()
+        {
+          EvalIterator::operator++();
+          incrementToNonZero();
+          return *this;
+        }
+
+        using EvalIterator::value;
+
+      protected:
+        const XprType &m_view;
+
+      private:
+        void incrementToNonZero()
+        {
+          while((bool(*this)) && internal::isMuchSmallerThan(value(), m_view.reference(), m_view.epsilon()))
+          {
+            EvalIterator::operator++();
+          }
+        }
+    };
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_view(xpr) {}
+
+  protected:
+    evaluator<ArgType> m_argImpl;
+    const XprType &m_view;
 };
 
-template<typename Derived>
-const SparseView<Derived> MatrixBase<Derived>::sparseView(const Scalar& m_reference,
-                                                          const typename NumTraits<Scalar>::Real& m_epsilon) const
+template<typename ArgType>
+struct unary_evaluator<SparseView<ArgType>, IndexBased>
+  : public evaluator_base<SparseView<ArgType> >
 {
-  return SparseView<Derived>(derived(), m_reference, m_epsilon);
+  public:
+    typedef SparseView<ArgType> XprType;
+  protected:
+    enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit };
+    typedef typename XprType::Scalar Scalar;
+    typedef typename XprType::StorageIndex StorageIndex;
+  public:
+    
+    class InnerIterator
+    {
+      public:
+
+        EIGEN_STRONG_INLINE InnerIterator(const unary_evaluator& sve, Index outer)
+          : m_sve(sve), m_inner(0), m_outer(outer), m_end(sve.m_view.innerSize())
+        {
+          incrementToNonZero();
+        }
+
+        EIGEN_STRONG_INLINE InnerIterator& operator++()
+        {
+          m_inner++;
+          incrementToNonZero();
+          return *this;
+        }
+
+        EIGEN_STRONG_INLINE Scalar value() const
+        {
+          return (IsRowMajor) ? m_sve.m_argImpl.coeff(m_outer, m_inner)
+                              : m_sve.m_argImpl.coeff(m_inner, m_outer);
+        }
+
+        EIGEN_STRONG_INLINE StorageIndex index() const { return m_inner; }
+        inline Index row() const { return IsRowMajor ? m_outer : index(); }
+        inline Index col() const { return IsRowMajor ? index() : m_outer; }
+
+        EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; }
+
+      protected:
+        const unary_evaluator &m_sve;
+        Index m_inner;
+        const Index m_outer;
+        const Index m_end;
+
+      private:
+        void incrementToNonZero()
+        {
+          while((bool(*this)) && internal::isMuchSmallerThan(value(), m_sve.m_view.reference(), m_sve.m_view.epsilon()))
+          {
+            m_inner++;
+          }
+        }
+    };
+    
+    enum {
+      CoeffReadCost = evaluator<ArgType>::CoeffReadCost,
+      Flags = XprType::Flags
+    };
+    
+    explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_view(xpr) {}
+
+  protected:
+    evaluator<ArgType> m_argImpl;
+    const XprType &m_view;
+};
+
+} // end namespace internal
+
+/** \ingroup SparseCore_Module
+  *
+  * \returns a sparse expression of the dense expression \c *this with values smaller than
+  * \a reference * \a epsilon removed.
+  *
+  * This method is typically used when prototyping to convert a quickly assembled dense Matrix \c D to a SparseMatrix \c S:
+  * \code
+  * MatrixXd D(n,m);
+  * SparseMatrix<double> S;
+  * S = D.sparseView();             // suppress numerical zeros (exact)
+  * S = D.sparseView(reference);
+  * S = D.sparseView(reference,epsilon);
+  * \endcode
+  * where \a reference is a meaningful non zero reference value,
+  * and \a epsilon is a tolerance factor defaulting to NumTraits<Scalar>::dummy_precision().
+  *
+  * \sa SparseMatrixBase::pruned(), class SparseView */
+template<typename Derived>
+const SparseView<Derived> MatrixBase<Derived>::sparseView(const Scalar& reference,
+                                                          const typename NumTraits<Scalar>::Real& epsilon) const
+{
+  return SparseView<Derived>(derived(), reference, epsilon);
+}
+
+/** \returns an expression of \c *this with values smaller than
+  * \a reference * \a epsilon removed.
+  *
+  * This method is typically used in conjunction with the product of two sparse matrices
+  * to automatically prune the smallest values as follows:
+  * \code
+  * C = (A*B).pruned();             // suppress numerical zeros (exact)
+  * C = (A*B).pruned(ref);
+  * C = (A*B).pruned(ref,epsilon);
+  * \endcode
+  * where \c ref is a meaningful non zero reference value.
+  * */
+template<typename Derived>
+const SparseView<Derived>
+SparseMatrixBase<Derived>::pruned(const Scalar& reference,
+                                  const RealScalar& epsilon) const
+{
+  return SparseView<Derived>(derived(), reference, epsilon);
 }
 
 } // end namespace Eigen
diff --git a/Eigen/src/SparseCore/TriangularSolver.h b/Eigen/src/SparseCore/TriangularSolver.h
index ccc12af..f9c56ba 100644
--- a/Eigen/src/SparseCore/TriangularSolver.h
+++ b/Eigen/src/SparseCore/TriangularSolver.h
@@ -28,16 +28,19 @@
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Lower,RowMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=0; i<lhs.rows(); ++i)
+      for(Index i=0; i<lhs.rows(); ++i)
       {
         Scalar tmp = other.coeff(i,col);
         Scalar lastVal(0);
-        int lastIndex = 0;
-        for(typename Lhs::InnerIterator it(lhs, i); it; ++it)
+        Index lastIndex = 0;
+        for(LhsIterator it(lhsEval, i); it; ++it)
         {
           lastVal = it.value();
           lastIndex = it.index();
@@ -62,15 +65,18 @@
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,RowMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=lhs.rows()-1 ; i>=0 ; --i)
+      for(Index i=lhs.rows()-1 ; i>=0 ; --i)
       {
         Scalar tmp = other.coeff(i,col);
         Scalar l_ii(0);
-        typename Lhs::InnerIterator it(lhs, i);
+        LhsIterator it(lhsEval, i);
         while(it && it.index()<i)
           ++it;
         if(!(Mode & UnitDiag))
@@ -86,10 +92,8 @@
           tmp -= it.value() * other.coeff(it.index(),col);
         }
 
-        if (Mode & UnitDiag)
-          other.coeffRef(i,col) = tmp;
-        else
-          other.coeffRef(i,col) = tmp/l_ii;
+        if (Mode & UnitDiag)  other.coeffRef(i,col) = tmp;
+        else                  other.coeffRef(i,col) = tmp/l_ii;
       }
     }
   }
@@ -100,16 +104,19 @@
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Lower,ColMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=0; i<lhs.cols(); ++i)
+      for(Index i=0; i<lhs.cols(); ++i)
       {
         Scalar& tmp = other.coeffRef(i,col);
         if (tmp!=Scalar(0)) // optimization when other is actually sparse
         {
-          typename Lhs::InnerIterator it(lhs, i);
+          LhsIterator it(lhsEval, i);
           while(it && it.index()<i)
             ++it;
           if(!(Mode & UnitDiag))
@@ -132,11 +139,14 @@
 struct sparse_solve_triangular_selector<Lhs,Rhs,Mode,Upper,ColMajor>
 {
   typedef typename Rhs::Scalar Scalar;
+  typedef evaluator<Lhs> LhsEval;
+  typedef typename evaluator<Lhs>::InnerIterator LhsIterator;
   static void run(const Lhs& lhs, Rhs& other)
   {
-    for(int col=0 ; col<other.cols() ; ++col)
+    LhsEval lhsEval(lhs);
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
-      for(int i=lhs.cols()-1; i>=0; --i)
+      for(Index i=lhs.cols()-1; i>=0; --i)
       {
         Scalar& tmp = other.coeffRef(i,col);
         if (tmp!=Scalar(0)) // optimization when other is actually sparse
@@ -144,13 +154,13 @@
           if(!(Mode & UnitDiag))
           {
             // TODO replace this by a binary search. make sure the binary search is safe for partially sorted elements
-            typename Lhs::ReverseInnerIterator it(lhs, i);
+            LhsIterator it(lhsEval, i);
             while(it && it.index()!=i)
-              --it;
+              ++it;
             eigen_assert(it && it.index()==i);
             other.coeffRef(i,col) /= it.value();
           }
-          typename Lhs::InnerIterator it(lhs, i);
+          LhsIterator it(lhsEval, i);
           for(; it && it.index()<i; ++it)
             other.coeffRef(it.index(), col) -= tmp * it.value();
         }
@@ -161,11 +171,13 @@
 
 } // end namespace internal
 
-template<typename ExpressionType,int Mode>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+
+template<typename ExpressionType,unsigned int Mode>
 template<typename OtherDerived>
-void SparseTriangularView<ExpressionType,Mode>::solveInPlace(MatrixBase<OtherDerived>& other) const
+void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(MatrixBase<OtherDerived>& other) const
 {
-  eigen_assert(m_matrix.cols() == m_matrix.rows() && m_matrix.cols() == other.rows());
+  eigen_assert(derived().cols() == derived().rows() && derived().cols() == other.rows());
   eigen_assert((!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
 
   enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit };
@@ -174,21 +186,12 @@
     typename internal::plain_matrix_type_column_major<OtherDerived>::type, OtherDerived&>::type OtherCopy;
   OtherCopy otherCopy(other.derived());
 
-  internal::sparse_solve_triangular_selector<ExpressionType, typename internal::remove_reference<OtherCopy>::type, Mode>::run(m_matrix, otherCopy);
+  internal::sparse_solve_triangular_selector<ExpressionType, typename internal::remove_reference<OtherCopy>::type, Mode>::run(derived().nestedExpression(), otherCopy);
 
   if (copy)
     other = otherCopy;
 }
-
-template<typename ExpressionType,int Mode>
-template<typename OtherDerived>
-typename internal::plain_matrix_type_column_major<OtherDerived>::type
-SparseTriangularView<ExpressionType,Mode>::solve(const MatrixBase<OtherDerived>& other) const
-{
-  typename internal::plain_matrix_type_column_major<OtherDerived>::type res(other);
-  solveInPlace(res);
-  return res;
-}
+#endif
 
 // pure sparse path
 
@@ -208,18 +211,18 @@
 struct sparse_solve_triangular_sparse_selector<Lhs,Rhs,Mode,UpLo,ColMajor>
 {
   typedef typename Rhs::Scalar Scalar;
-  typedef typename promote_index_type<typename traits<Lhs>::Index,
-                                         typename traits<Rhs>::Index>::type Index;
+  typedef typename promote_index_type<typename traits<Lhs>::StorageIndex,
+                                      typename traits<Rhs>::StorageIndex>::type StorageIndex;
   static void run(const Lhs& lhs, Rhs& other)
   {
     const bool IsLower = (UpLo==Lower);
-    AmbiVector<Scalar,Index> tempVector(other.rows()*2);
+    AmbiVector<Scalar,StorageIndex> tempVector(other.rows()*2);
     tempVector.setBounds(0,other.rows());
 
     Rhs res(other.rows(), other.cols());
     res.reserve(other.nonZeros());
 
-    for(int col=0 ; col<other.cols() ; ++col)
+    for(Index col=0 ; col<other.cols() ; ++col)
     {
       // FIXME estimate number of non zeros
       tempVector.init(.99/*float(other.col(col).nonZeros())/float(other.rows())*/);
@@ -230,7 +233,7 @@
         tempVector.coeffRef(rhsIt.index()) = rhsIt.value();
       }
 
-      for(int i=IsLower?0:lhs.cols()-1;
+      for(Index i=IsLower?0:lhs.cols()-1;
           IsLower?i<lhs.cols():i>=0;
           i+=IsLower?1:-1)
       {
@@ -267,9 +270,9 @@
       }
 
 
-      int count = 0;
+      Index count = 0;
       // FIXME compute a reference value to filter zeros
-      for (typename AmbiVector<Scalar,Index>::Iterator it(tempVector/*,1e-12*/); it; ++it)
+      for (typename AmbiVector<Scalar,StorageIndex>::Iterator it(tempVector/*,1e-12*/); it; ++it)
       {
         ++ count;
 //         std::cerr << "fill " << it.index() << ", " << col << "\n";
@@ -286,11 +289,12 @@
 
 } // end namespace internal
 
-template<typename ExpressionType,int Mode>
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+template<typename ExpressionType,unsigned int Mode>
 template<typename OtherDerived>
-void SparseTriangularView<ExpressionType,Mode>::solveInPlace(SparseMatrixBase<OtherDerived>& other) const
+void TriangularViewImpl<ExpressionType,Mode,Sparse>::solveInPlace(SparseMatrixBase<OtherDerived>& other) const
 {
-  eigen_assert(m_matrix.cols() == m_matrix.rows() && m_matrix.cols() == other.rows());
+  eigen_assert(derived().cols() == derived().rows() && derived().cols() == other.rows());
   eigen_assert( (!(Mode & ZeroDiag)) && bool(Mode & (Upper|Lower)));
 
 //   enum { copy = internal::traits<OtherDerived>::Flags & RowMajorBit };
@@ -299,35 +303,12 @@
 //     typename internal::plain_matrix_type_column_major<OtherDerived>::type, OtherDerived&>::type OtherCopy;
 //   OtherCopy otherCopy(other.derived());
 
-  internal::sparse_solve_triangular_sparse_selector<ExpressionType, OtherDerived, Mode>::run(m_matrix, other.derived());
+  internal::sparse_solve_triangular_sparse_selector<ExpressionType, OtherDerived, Mode>::run(derived().nestedExpression(), other.derived());
 
 //   if (copy)
 //     other = otherCopy;
 }
-
-#ifdef EIGEN2_SUPPORT
-
-// deprecated stuff:
-
-/** \deprecated */
-template<typename Derived>
-template<typename OtherDerived>
-void SparseMatrixBase<Derived>::solveTriangularInPlace(MatrixBase<OtherDerived>& other) const
-{
-  this->template triangular<Flags&(Upper|Lower)>().solveInPlace(other);
-}
-
-/** \deprecated */
-template<typename Derived>
-template<typename OtherDerived>
-typename internal::plain_matrix_type_column_major<OtherDerived>::type
-SparseMatrixBase<Derived>::solveTriangular(const MatrixBase<OtherDerived>& other) const
-{
-  typename internal::plain_matrix_type_column_major<OtherDerived>::type res(other);
-  derived().solveTriangularInPlace(res);
-  return res;
-}
-#endif // EIGEN2_SUPPORT
+#endif
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/SparseLU/CMakeLists.txt b/Eigen/src/SparseLU/CMakeLists.txt
deleted file mode 100644
index 69729ee..0000000
--- a/Eigen/src/SparseLU/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_SparseLU_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_SparseLU_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/SparseLU COMPONENT Devel
-  )
diff --git a/Eigen/src/SparseLU/SparseLU.h b/Eigen/src/SparseLU/SparseLU.h
index 4514cfd..7104831 100644
--- a/Eigen/src/SparseLU/SparseLU.h
+++ b/Eigen/src/SparseLU/SparseLU.h
@@ -2,7 +2,7 @@
 // for linear algebra.
 //
 // Copyright (C) 2012 Désiré Nuentsa-Wakam <desire.nuentsa_wakam@inria.fr>
-// Copyright (C) 2012 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2012-2014 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -14,7 +14,7 @@
 
 namespace Eigen {
 
-template <typename _MatrixType, typename _OrderingType = COLAMDOrdering<typename _MatrixType::Index> > class SparseLU;
+template <typename _MatrixType, typename _OrderingType = COLAMDOrdering<typename _MatrixType::StorageIndex> > class SparseLU;
 template <typename MappedSparseMatrixType> struct SparseLUMatrixLReturnType;
 template <typename MatrixLType, typename MatrixUType> struct SparseLUMatrixUReturnType;
 
@@ -64,33 +64,45 @@
   * 
   * \tparam _MatrixType The type of the sparse matrix. It must be a column-major SparseMatrix<>
   * \tparam _OrderingType The ordering method to use, either AMD, COLAMD or METIS. Default is COLMAD
+  *
+  * \implsparsesolverconcept
   * 
-  * 
-  * \sa \ref TutorialSparseDirectSolvers
+  * \sa \ref TutorialSparseSolverConcept
   * \sa \ref OrderingMethods_Module
   */
 template <typename _MatrixType, typename _OrderingType>
-class SparseLU : public internal::SparseLUImpl<typename _MatrixType::Scalar, typename _MatrixType::Index>
+class SparseLU : public SparseSolverBase<SparseLU<_MatrixType,_OrderingType> >, public internal::SparseLUImpl<typename _MatrixType::Scalar, typename _MatrixType::StorageIndex>
 {
+  protected:
+    typedef SparseSolverBase<SparseLU<_MatrixType,_OrderingType> > APIBase;
+    using APIBase::m_isInitialized;
   public:
+    using APIBase::_solve_impl;
+    
     typedef _MatrixType MatrixType; 
     typedef _OrderingType OrderingType;
     typedef typename MatrixType::Scalar Scalar; 
     typedef typename MatrixType::RealScalar RealScalar; 
-    typedef typename MatrixType::Index Index; 
-    typedef SparseMatrix<Scalar,ColMajor,Index> NCMatrix;
-    typedef internal::MappedSuperNodalMatrix<Scalar, Index> SCMatrix; 
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> NCMatrix;
+    typedef internal::MappedSuperNodalMatrix<Scalar, StorageIndex> SCMatrix;
     typedef Matrix<Scalar,Dynamic,1> ScalarVector;
-    typedef Matrix<Index,Dynamic,1> IndexVector;
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
-    typedef internal::SparseLUImpl<Scalar, Index> Base;
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
+    typedef internal::SparseLUImpl<Scalar, StorageIndex> Base;
+
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
     
   public:
-    SparseLU():m_isInitialized(true),m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
+    SparseLU():m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
     {
       initperfvalues(); 
     }
-    SparseLU(const MatrixType& matrix):m_isInitialized(true),m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
+    explicit SparseLU(const MatrixType& matrix)
+      : m_lastError(""),m_Ustore(0,0,0,0,0,0),m_symmetricmode(false),m_diagpivotthresh(1.0),m_detPermR(1)
     {
       initperfvalues(); 
       compute(matrix);
@@ -141,9 +153,9 @@
       * y = b; matrixU().solveInPlace(y);
       * \endcode
       */
-    SparseLUMatrixUReturnType<SCMatrix,MappedSparseMatrix<Scalar,ColMajor,Index> > matrixU() const
+    SparseLUMatrixUReturnType<SCMatrix,MappedSparseMatrix<Scalar,ColMajor,StorageIndex> > matrixU() const
     {
-      return SparseLUMatrixUReturnType<SCMatrix, MappedSparseMatrix<Scalar,ColMajor,Index> >(m_Lstore, m_Ustore);
+      return SparseLUMatrixUReturnType<SCMatrix, MappedSparseMatrix<Scalar,ColMajor,StorageIndex> >(m_Lstore, m_Ustore);
     }
 
     /**
@@ -168,6 +180,7 @@
       m_diagpivotthresh = thresh; 
     }
 
+#ifdef EIGEN_PARSED_BY_DOXYGEN
     /** \returns the solution X of \f$ A X = B \f$ using the current decomposition of A.
       *
       * \warning the destination matrix X in X = this->solve(B) must be colmun-major.
@@ -175,26 +188,8 @@
       * \sa compute()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<SparseLU, Rhs> solve(const MatrixBase<Rhs>& B) const 
-    {
-      eigen_assert(m_factorizationIsOk && "SparseLU is not initialized."); 
-      eigen_assert(rows()==B.rows()
-                    && "SparseLU::solve(): invalid number of rows of the right hand side matrix B");
-          return internal::solve_retval<SparseLU, Rhs>(*this, B.derived());
-    }
-
-    /** \returns the solution X of \f$ A X = B \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<SparseLU, Rhs> solve(const SparseMatrixBase<Rhs>& B) const 
-    {
-      eigen_assert(m_factorizationIsOk && "SparseLU is not initialized."); 
-      eigen_assert(rows()==B.rows()
-                    && "SparseLU::solve(): invalid number of rows of the right hand side matrix B");
-          return internal::sparse_solve_retval<SparseLU, Rhs>(*this, B.derived());
-    }
+    inline const Solve<SparseLU, Rhs> solve(const MatrixBase<Rhs>& B) const;
+#endif // EIGEN_PARSED_BY_DOXYGEN
     
     /** \brief Reports whether previous computation was successful.
       *
@@ -219,7 +214,7 @@
     }
 
     template<typename Rhs, typename Dest>
-    bool _solve(const MatrixBase<Rhs> &B, MatrixBase<Dest> &X_base) const
+    bool _solve_impl(const MatrixBase<Rhs> &B, MatrixBase<Dest> &X_base) const
     {
       Dest& X(X_base.derived());
       eigen_assert(m_factorizationIsOk && "The matrix should be factorized first");
@@ -255,8 +250,9 @@
       *
       * \sa logAbsDeterminant(), signDeterminant()
       */
-     Scalar absDeterminant()
+    Scalar absDeterminant()
     {
+      using std::abs;
       eigen_assert(m_factorizationIsOk && "The matrix should be factorized first.");
       // Initialize with the determinant of the row matrix
       Scalar det = Scalar(1.);
@@ -268,42 +264,43 @@
         {
           if(it.index() == j)
           {
-            using std::abs;
             det *= abs(it.value());
             break;
           }
         }
-       }
-       return det;
-     }
+      }
+      return det;
+    }
 
-     /** \returns the natural log of the absolute value of the determinant of the matrix
-       * of which **this is the QR decomposition
-       *
-       * \note This method is useful to work around the risk of overflow/underflow that's
-       * inherent to the determinant computation.
-       *
-       * \sa absDeterminant(), signDeterminant()
-       */
-     Scalar logAbsDeterminant() const
-     {
-       eigen_assert(m_factorizationIsOk && "The matrix should be factorized first.");
-       Scalar det = Scalar(0.);
-       for (Index j = 0; j < this->cols(); ++j)
-       {
-         for (typename SCMatrix::InnerIterator it(m_Lstore, j); it; ++it)
-         {
-           if(it.row() < j) continue;
-           if(it.row() == j)
-           {
-             using std::log; using std::abs;
-             det += log(abs(it.value()));
-             break;
-           }
-         }
-       }
-       return det;
-     }
+    /** \returns the natural log of the absolute value of the determinant of the matrix
+      * of which **this is the QR decomposition
+      *
+      * \note This method is useful to work around the risk of overflow/underflow that's
+      * inherent to the determinant computation.
+      *
+      * \sa absDeterminant(), signDeterminant()
+      */
+    Scalar logAbsDeterminant() const
+    {
+      using std::log;
+      using std::abs;
+
+      eigen_assert(m_factorizationIsOk && "The matrix should be factorized first.");
+      Scalar det = Scalar(0.);
+      for (Index j = 0; j < this->cols(); ++j)
+      {
+        for (typename SCMatrix::InnerIterator it(m_Lstore, j); it; ++it)
+        {
+          if(it.row() < j) continue;
+          if(it.row() == j)
+          {
+            det += log(abs(it.value()));
+            break;
+          }
+        }
+      }
+      return det;
+    }
 
     /** \returns A number representing the sign of the determinant
       *
@@ -355,7 +352,7 @@
           }
         }
       }
-      return det * Scalar(m_detPermR * m_detPermC);
+      return (m_detPermR * m_detPermC) > 0 ? det : -det;
     }
 
   protected:
@@ -372,13 +369,12 @@
       
     // Variables 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     bool m_factorizationIsOk;
     bool m_analysisIsOk;
     std::string m_lastError;
     NCMatrix m_mat; // The input (permuted ) matrix 
     SCMatrix m_Lstore; // The lower triangular matrix (supernodal)
-    MappedSparseMatrix<Scalar,ColMajor,Index> m_Ustore; // The upper triangular matrix
+    MappedSparseMatrix<Scalar,ColMajor,StorageIndex> m_Ustore; // The upper triangular matrix
     PermutationType m_perm_c; // Column permutation 
     PermutationType m_perm_r ; // Row permutation
     IndexVector m_etree; // Column elimination tree 
@@ -388,7 +384,7 @@
     // SparseLU options 
     bool m_symmetricmode;
     // values for performance 
-    internal::perfvalues<Index> m_perfv; 
+    internal::perfvalues m_perfv;
     RealScalar m_diagpivotthresh; // Specifies the threshold used for a diagonal entry to be an acceptable pivot
     Index m_nnzL, m_nnzU; // Nonzeros in L and U factors
     Index m_detPermR, m_detPermC; // Determinants of the permutation matrices
@@ -417,30 +413,32 @@
   
   //TODO  It is possible as in SuperLU to compute row and columns scaling vectors to equilibrate the matrix mat.
   
+  // Firstly, copy the whole input matrix. 
+  m_mat = mat;
+  
+  // Compute fill-in ordering
   OrderingType ord; 
-  ord(mat,m_perm_c);
+  ord(m_mat,m_perm_c);
   
   // Apply the permutation to the column of the input  matrix
-  //First copy the whole input matrix. 
-  m_mat = mat;
-  if (m_perm_c.size()) {
+  if (m_perm_c.size())
+  {
     m_mat.uncompress(); //NOTE: The effect of this command is only to create the InnerNonzeros pointers. FIXME : This vector is filled but not subsequently used.  
-    //Then, permute only the column pointers
-    const Index * outerIndexPtr;
-    if (mat.isCompressed()) outerIndexPtr = mat.outerIndexPtr();
-    else
-    {
-      Index *outerIndexPtr_t = new Index[mat.cols()+1];
-      for(Index i = 0; i <= mat.cols(); i++) outerIndexPtr_t[i] = m_mat.outerIndexPtr()[i];
-      outerIndexPtr = outerIndexPtr_t;
-    }
+    // Then, permute only the column pointers
+    ei_declare_aligned_stack_constructed_variable(StorageIndex,outerIndexPtr,mat.cols()+1,mat.isCompressed()?const_cast<StorageIndex*>(mat.outerIndexPtr()):0);
+    
+    // If the input matrix 'mat' is uncompressed, then the outer-indices do not match the ones of m_mat, and a copy is thus needed.
+    if(!mat.isCompressed()) 
+      IndexVector::Map(outerIndexPtr, mat.cols()+1) = IndexVector::Map(m_mat.outerIndexPtr(),mat.cols()+1);
+    
+    // Apply the permutation and compute the nnz per column.
     for (Index i = 0; i < mat.cols(); i++)
     {
       m_mat.outerIndexPtr()[m_perm_c.indices()(i)] = outerIndexPtr[i];
       m_mat.innerNonZeroPtr()[m_perm_c.indices()(i)] = outerIndexPtr[i+1] - outerIndexPtr[i];
     }
-    if(!mat.isCompressed()) delete[] outerIndexPtr;
   }
+  
   // Compute the column elimination tree of the permuted matrix 
   IndexVector firstRowElt;
   internal::coletree(m_mat, m_etree,firstRowElt); 
@@ -449,7 +447,7 @@
   if (!m_symmetricmode) {
     IndexVector post, iwork; 
     // Post order etree
-    internal::treePostorder(m_mat.cols(), m_etree, post); 
+    internal::treePostorder(StorageIndex(m_mat.cols()), m_etree, post); 
       
    
     // Renumber etree in postorder 
@@ -501,7 +499,7 @@
   eigen_assert(m_analysisIsOk && "analyzePattern() should be called first"); 
   eigen_assert((matrix.rows() == matrix.cols()) && "Only for squared matrices");
   
-  typedef typename IndexVector::Scalar Index; 
+  m_isInitialized = true;
   
   
   // Apply the column permutation computed in analyzepattern()
@@ -511,11 +509,11 @@
   {
     m_mat.uncompress(); //NOTE: The effect of this command is only to create the InnerNonzeros pointers.
     //Then, permute only the column pointers
-    const Index * outerIndexPtr;
+    const StorageIndex * outerIndexPtr;
     if (matrix.isCompressed()) outerIndexPtr = matrix.outerIndexPtr();
     else
     {
-      Index* outerIndexPtr_t = new Index[matrix.cols()+1];
+      StorageIndex* outerIndexPtr_t = new StorageIndex[matrix.cols()+1];
       for(Index i = 0; i <= matrix.cols(); i++) outerIndexPtr_t[i] = m_mat.outerIndexPtr()[i];
       outerIndexPtr = outerIndexPtr_t;
     }
@@ -529,7 +527,7 @@
   else 
   { //FIXME This should not be needed if the empty permutation is handled transparently
     m_perm_c.resize(matrix.cols());
-    for(Index i = 0; i < matrix.cols(); ++i) m_perm_c.indices()(i) = i;
+    for(StorageIndex i = 0; i < matrix.cols(); ++i) m_perm_c.indices()(i) = i;
   }
   
   Index m = m_mat.rows();
@@ -694,7 +692,7 @@
   // Create supernode matrix L 
   m_Lstore.setInfos(m, n, m_glu.lusup, m_glu.xlusup, m_glu.lsub, m_glu.xlsub, m_glu.supno, m_glu.xsup); 
   // Create the column major upper sparse matrix  U; 
-  new (&m_Ustore) MappedSparseMatrix<Scalar, ColMajor, Index> ( m, n, m_nnzU, m_glu.xusub.data(), m_glu.usub.data(), m_glu.ucol.data() ); 
+  new (&m_Ustore) MappedSparseMatrix<Scalar, ColMajor, StorageIndex> ( m, n, m_nnzU, m_glu.xusub.data(), m_glu.usub.data(), m_glu.ucol.data() );
   
   m_info = Success;
   m_factorizationIsOk = true;
@@ -703,9 +701,8 @@
 template<typename MappedSupernodalType>
 struct SparseLUMatrixLReturnType : internal::no_assignment_operator
 {
-  typedef typename MappedSupernodalType::Index Index;
   typedef typename MappedSupernodalType::Scalar Scalar;
-  SparseLUMatrixLReturnType(const MappedSupernodalType& mapL) : m_mapL(mapL)
+  explicit SparseLUMatrixLReturnType(const MappedSupernodalType& mapL) : m_mapL(mapL)
   { }
   Index rows() { return m_mapL.rows(); }
   Index cols() { return m_mapL.cols(); }
@@ -720,7 +717,6 @@
 template<typename MatrixLType, typename MatrixUType>
 struct SparseLUMatrixUReturnType : internal::no_assignment_operator
 {
-  typedef typename MatrixLType::Index Index;
   typedef typename MatrixLType::Scalar Scalar;
   SparseLUMatrixUReturnType(const MatrixLType& mapL, const MatrixUType& mapU)
   : m_mapL(mapL),m_mapU(mapU)
@@ -731,7 +727,7 @@
   template<typename Dest>   void solveInPlace(MatrixBase<Dest> &X) const
   {
     Index nrhs = X.cols();
-    Index n = X.rows();
+    Index n    = X.rows();
     // Backward solve with U
     for (Index k = m_mapL.nsuper(); k >= 0; k--)
     {
@@ -749,8 +745,8 @@
       }
       else
       {
-        Map<const Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > A( &(m_mapL.valuePtr()[luptr]), nsupc, nsupc, OuterStride<>(lda) );
-        Map< Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
+        Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > A( &(m_mapL.valuePtr()[luptr]), nsupc, nsupc, OuterStride<>(lda) );
+        Map< Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
         U = A.template triangularView<Upper>().solve(U);
       }
 
@@ -772,35 +768,6 @@
   const MatrixUType& m_mapU;
 };
 
-namespace internal {
-  
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct solve_retval<SparseLU<_MatrixType,Derived>, Rhs>
-  : solve_retval_base<SparseLU<_MatrixType,Derived>, Rhs>
-{
-  typedef SparseLU<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct sparse_solve_retval<SparseLU<_MatrixType,Derived>, Rhs>
-  : sparse_solve_retval_base<SparseLU<_MatrixType,Derived>, Rhs>
-{
-  typedef SparseLU<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-} // end namespace internal
-
 } // End namespace Eigen 
 
 #endif
diff --git a/Eigen/src/SparseLU/SparseLUImpl.h b/Eigen/src/SparseLU/SparseLUImpl.h
index 14d7089..fc0cfc4 100644
--- a/Eigen/src/SparseLU/SparseLUImpl.h
+++ b/Eigen/src/SparseLU/SparseLUImpl.h
@@ -16,17 +16,19 @@
   * \class SparseLUImpl
   * Base class for sparseLU
   */
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 class SparseLUImpl
 {
   public:
     typedef Matrix<Scalar,Dynamic,1> ScalarVector;
-    typedef Matrix<Index,Dynamic,1> IndexVector; 
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector; 
+    typedef Matrix<Scalar,Dynamic,Dynamic,ColMajor> ScalarMatrix;
+    typedef Map<ScalarMatrix, 0,  OuterStride<> > MappedMatrixBlock;
     typedef typename ScalarVector::RealScalar RealScalar; 
     typedef Ref<Matrix<Scalar,Dynamic,1> > BlockScalarVector;
-    typedef Ref<Matrix<Index,Dynamic,1> > BlockIndexVector;
+    typedef Ref<Matrix<StorageIndex,Dynamic,1> > BlockIndexVector;
     typedef LU_GlobalLU_t<IndexVector, ScalarVector> GlobalLU_t; 
-    typedef SparseMatrix<Scalar,ColMajor,Index> MatrixType; 
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> MatrixType; 
     
   protected:
      template <typename VectorType>
@@ -40,7 +42,7 @@
      Index snode_bmod (const Index jcol, const Index fsupc, ScalarVector& dense, GlobalLU_t& glu);
      Index pivotL(const Index jcol, const RealScalar& diagpivotthresh, IndexVector& perm_r, IndexVector& iperm_c, Index& pivrow, GlobalLU_t& glu);
      template <typename Traits>
-     void dfs_kernel(const Index jj, IndexVector& perm_r,
+     void dfs_kernel(const StorageIndex jj, IndexVector& perm_r,
                     Index& nseg, IndexVector& panel_lsub, IndexVector& segrep,
                     Ref<IndexVector> repfnz_col, IndexVector& xprune, Ref<IndexVector> marker, IndexVector& parent,
                     IndexVector& xplore, GlobalLU_t& glu, Index& nextl_col, Index krow, Traits& traits);
diff --git a/Eigen/src/SparseLU/SparseLU_Memory.h b/Eigen/src/SparseLU/SparseLU_Memory.h
index 1ffa7d5..4dc42e8 100644
--- a/Eigen/src/SparseLU/SparseLU_Memory.h
+++ b/Eigen/src/SparseLU/SparseLU_Memory.h
@@ -36,13 +36,12 @@
   
 enum { LUNoMarker = 3 };
 enum {emptyIdxLU = -1};
-template<typename Index>
 inline Index LUnumTempV(Index& m, Index& w, Index& t, Index& b)
 {
   return (std::max)(m, (t+b)*w);
 }
 
-template< typename Scalar, typename Index>
+template< typename Scalar>
 inline Index LUTempSpace(Index&m, Index& w)
 {
   return (2*w + 4 + LUNoMarker) * m * sizeof(Index) + (w + 1) * m * sizeof(Scalar);
@@ -59,9 +58,9 @@
   * \param keep_prev  1: use length  and do not expand the vector; 0: compute new_len and expand
   * \param[in,out] num_expansions Number of times the memory has been expanded
   */
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 template <typename VectorType>
-Index  SparseLUImpl<Scalar,Index>::expand(VectorType& vec, Index& length, Index nbElts, Index keep_prev, Index& num_expansions) 
+Index  SparseLUImpl<Scalar,StorageIndex>::expand(VectorType& vec, Index& length, Index nbElts, Index keep_prev, Index& num_expansions) 
 {
   
   float alpha = 1.5; // Ratio of the memory increase 
@@ -148,13 +147,13 @@
  * \return an estimated size of the required memory if lwork = -1; otherwise, return the size of actually allocated memory when allocation failed, and 0 on success
  * \note Unlike SuperLU, this routine does not support successive factorization with the same pattern and the same row permutation
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::memInit(Index m, Index n, Index annz, Index lwork, Index fillratio, Index panel_size,  GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::memInit(Index m, Index n, Index annz, Index lwork, Index fillratio, Index panel_size,  GlobalLU_t& glu)
 {
   Index& num_expansions = glu.num_expansions; //No memory expansions so far
   num_expansions = 0;
-  glu.nzumax = glu.nzlumax = (std::min)(fillratio * annz / n, m) * n; // estimated number of nonzeros in U 
-  glu.nzlmax = (std::max)(Index(4), fillratio) * annz / 4; // estimated  nnz in L factor
+  glu.nzumax = glu.nzlumax = (std::min)(fillratio * (annz+1) / n, m) * n; // estimated number of nonzeros in U 
+  glu.nzlmax = (std::max)(Index(4), fillratio) * (annz+1) / 4; // estimated  nnz in L factor
   // Return the estimated size to the user if necessary
   Index tempSpace;
   tempSpace = (2*panel_size + 4 + LUNoMarker) * m * sizeof(Index) + (panel_size + 1) * m * sizeof(Scalar);
@@ -205,9 +204,9 @@
  * \param num_expansions Number of expansions 
  * \return 0 on success, > 0 size of the memory allocated so far
  */
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 template <typename VectorType>
-Index SparseLUImpl<Scalar,Index>::memXpand(VectorType& vec, Index& maxlen, Index nbElts, MemType memtype, Index& num_expansions)
+Index SparseLUImpl<Scalar,StorageIndex>::memXpand(VectorType& vec, Index& maxlen, Index nbElts, MemType memtype, Index& num_expansions)
 {
   Index failed_size; 
   if (memtype == USUB)
diff --git a/Eigen/src/SparseLU/SparseLU_Structs.h b/Eigen/src/SparseLU/SparseLU_Structs.h
index 24d6bf1..cf5ec44 100644
--- a/Eigen/src/SparseLU/SparseLU_Structs.h
+++ b/Eigen/src/SparseLU/SparseLU_Structs.h
@@ -75,7 +75,7 @@
 
 template <typename IndexVector, typename ScalarVector>
 struct LU_GlobalLU_t {
-  typedef typename IndexVector::Scalar Index; 
+  typedef typename IndexVector::Scalar StorageIndex; 
   IndexVector xsup; //First supernode column ... xsup(s) points to the beginning of the s-th supernode
   IndexVector supno; // Supernode number corresponding to this column (column to supernode mapping)
   ScalarVector  lusup; // nonzero values of L ordered by columns 
@@ -93,7 +93,6 @@
 };
 
 // Values to set for performance
-template <typename Index>
 struct perfvalues {
   Index panel_size; // a panel consists of at most <panel_size> consecutive columns
   Index relax; // To control degree of relaxing supernodes. If the number of nodes (columns) 
diff --git a/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h b/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h
index b16afd6..721e188 100644
--- a/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h
+++ b/Eigen/src/SparseLU/SparseLU_SupernodalMatrix.h
@@ -29,20 +29,20 @@
  * SuperInnerIterator to iterate through all supernodes 
  * Function for triangular solve
  */
-template <typename _Scalar, typename _Index>
+template <typename _Scalar, typename _StorageIndex>
 class MappedSuperNodalMatrix
 {
   public:
     typedef _Scalar Scalar; 
-    typedef _Index Index;
-    typedef Matrix<Index,Dynamic,1> IndexVector; 
+    typedef _StorageIndex StorageIndex;
+    typedef Matrix<StorageIndex,Dynamic,1> IndexVector;
     typedef Matrix<Scalar,Dynamic,1> ScalarVector;
   public:
     MappedSuperNodalMatrix()
     {
       
     }
-    MappedSuperNodalMatrix(Index m, Index n,  ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind, 
+    MappedSuperNodalMatrix(Index m, Index n,  ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind,
              IndexVector& rowind_colptr, IndexVector& col_to_sup, IndexVector& sup_to_col )
     {
       setInfos(m, n, nzval, nzval_colptr, rowind, rowind_colptr, col_to_sup, sup_to_col);
@@ -58,7 +58,7 @@
      * FIXME This class will be modified such that it can be use in the course 
      * of the factorization.
      */
-    void setInfos(Index m, Index n, ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind, 
+    void setInfos(Index m, Index n, ScalarVector& nzval, IndexVector& nzval_colptr, IndexVector& rowind,
              IndexVector& rowind_colptr, IndexVector& col_to_sup, IndexVector& sup_to_col )
     {
       m_row = m;
@@ -96,12 +96,12 @@
     /**
      * Return the pointers to the beginning of each column in \ref valuePtr()
      */
-    Index* colIndexPtr()
+    StorageIndex* colIndexPtr()
     {
       return m_nzval_colptr; 
     }
     
-    const Index* colIndexPtr() const
+    const StorageIndex* colIndexPtr() const
     {
       return m_nzval_colptr; 
     }
@@ -109,9 +109,9 @@
     /**
      * Return the array of compressed row indices of all supernodes
      */
-    Index* rowIndex()  { return m_rowind; }
+    StorageIndex* rowIndex()  { return m_rowind; }
     
-    const Index* rowIndex() const
+    const StorageIndex* rowIndex() const
     {
       return m_rowind; 
     }
@@ -119,9 +119,9 @@
     /**
      * Return the location in \em rowvaluePtr() which starts each column
      */
-    Index* rowIndexPtr() { return m_rowind_colptr; }
+    StorageIndex* rowIndexPtr() { return m_rowind_colptr; }
     
-    const Index* rowIndexPtr() const 
+    const StorageIndex* rowIndexPtr() const
     {
       return m_rowind_colptr; 
     }
@@ -129,18 +129,18 @@
     /** 
      * Return the array of column-to-supernode mapping 
      */
-    Index* colToSup()  { return m_col_to_sup; }
+    StorageIndex* colToSup()  { return m_col_to_sup; }
     
-    const Index* colToSup() const
+    const StorageIndex* colToSup() const
     {
       return m_col_to_sup;       
     }
     /**
      * Return the array of supernode-to-column mapping
      */
-    Index* supToCol() { return m_sup_to_col; }
+    StorageIndex* supToCol() { return m_sup_to_col; }
     
-    const Index* supToCol() const 
+    const StorageIndex* supToCol() const
     {
       return m_sup_to_col;
     }
@@ -148,7 +148,7 @@
     /**
      * Return the number of supernodes
      */
-    Index nsuper() const 
+    Index nsuper() const
     {
       return m_nsuper; 
     }
@@ -162,14 +162,14 @@
     
   protected:
     Index m_row; // Number of rows
-    Index m_col; // Number of columns 
-    Index m_nsuper; // Number of supernodes 
+    Index m_col; // Number of columns
+    Index m_nsuper; // Number of supernodes
     Scalar* m_nzval; //array of nonzero values packed by column
-    Index* m_nzval_colptr; //nzval_colptr[j] Stores the location in nzval[] which starts column j 
-    Index* m_rowind; // Array of compressed row indices of rectangular supernodes
-    Index* m_rowind_colptr; //rowind_colptr[j] stores the location in rowind[] which starts column j
-    Index* m_col_to_sup; // col_to_sup[j] is the supernode number to which column j belongs
-    Index* m_sup_to_col; //sup_to_col[s] points to the starting column of the s-th supernode
+    StorageIndex* m_nzval_colptr; //nzval_colptr[j] Stores the location in nzval[] which starts column j
+    StorageIndex* m_rowind; // Array of compressed row indices of rectangular supernodes
+    StorageIndex* m_rowind_colptr; //rowind_colptr[j] stores the location in rowind[] which starts column j
+    StorageIndex* m_col_to_sup; // col_to_sup[j] is the supernode number to which column j belongs
+    StorageIndex* m_sup_to_col; //sup_to_col[s] points to the starting column of the s-th supernode
     
   private :
 };
@@ -178,13 +178,13 @@
   * \brief InnerIterator class to iterate over nonzero values of the current column in the supernodal matrix L
   * 
   */
-template<typename Scalar, typename Index>
-class MappedSuperNodalMatrix<Scalar,Index>::InnerIterator
+template<typename Scalar, typename StorageIndex>
+class MappedSuperNodalMatrix<Scalar,StorageIndex>::InnerIterator
 {
   public:
      InnerIterator(const MappedSuperNodalMatrix& mat, Index outer)
       : m_matrix(mat),
-        m_outer(outer), 
+        m_outer(outer),
         m_supno(mat.colToSup()[outer]),
         m_idval(mat.colIndexPtr()[outer]),
         m_startidval(m_idval),
@@ -229,14 +229,17 @@
  * \brief Solve with the supernode triangular matrix
  * 
  */
-template<typename Scalar, typename Index>
+template<typename Scalar, typename Index_>
 template<typename Dest>
-void MappedSuperNodalMatrix<Scalar,Index>::solveInPlace( MatrixBase<Dest>&X) const
+void MappedSuperNodalMatrix<Scalar,Index_>::solveInPlace( MatrixBase<Dest>&X) const
 {
-    Index n = X.rows(); 
-    Index nrhs = X.cols(); 
+    /* Explicit type conversion as the Index type of MatrixBase<Dest> may be wider than Index */
+//    eigen_assert(X.rows() <= NumTraits<Index>::highest());
+//    eigen_assert(X.cols() <= NumTraits<Index>::highest());
+    Index n    = int(X.rows());
+    Index nrhs = Index(X.cols());
     const Scalar * Lval = valuePtr();                 // Nonzero values 
-    Matrix<Scalar,Dynamic,Dynamic> work(n, nrhs);     // working vector
+    Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor> work(n, nrhs);     // working vector
     work.setZero();
     for (Index k = 0; k <= nsuper(); k ++)
     {
@@ -267,13 +270,13 @@
         Index lda = colIndexPtr()[fsupc+1] - luptr;
         
         // Triangular solve 
-        Map<const Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > A( &(Lval[luptr]), nsupc, nsupc, OuterStride<>(lda) ); 
-        Map< Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) ); 
+        Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > A( &(Lval[luptr]), nsupc, nsupc, OuterStride<>(lda) );
+        Map< Matrix<Scalar,Dynamic,Dest::ColsAtCompileTime, ColMajor>, 0, OuterStride<> > U (&(X(fsupc,0)), nsupc, nrhs, OuterStride<>(n) );
         U = A.template triangularView<UnitLower>().solve(U); 
         
         // Matrix-vector product 
-        new (&A) Map<const Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > ( &(Lval[luptr+nsupc]), nrow, nsupc, OuterStride<>(lda) ); 
-        work.block(0, 0, nrow, nrhs) = A * U; 
+        new (&A) Map<const Matrix<Scalar,Dynamic,Dynamic, ColMajor>, 0, OuterStride<> > ( &(Lval[luptr+nsupc]), nrow, nsupc, OuterStride<>(lda) );
+        work.topRows(nrow).noalias() = A * U;
         
         //Begin Scatter 
         for (Index j = 0; j < nrhs; j++)
diff --git a/Eigen/src/SparseLU/SparseLU_Utils.h b/Eigen/src/SparseLU/SparseLU_Utils.h
index 15352ac..9e3dab4 100644
--- a/Eigen/src/SparseLU/SparseLU_Utils.h
+++ b/Eigen/src/SparseLU/SparseLU_Utils.h
@@ -17,8 +17,8 @@
 /**
  * \brief Count Nonzero elements in the factors
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::countnz(const Index n, Index& nnzL, Index& nnzU, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::countnz(const Index n, Index& nnzL, Index& nnzU, GlobalLU_t& glu)
 {
  nnzL = 0; 
  nnzU = (glu.xusub)(n); 
@@ -48,12 +48,12 @@
  * and applies permutation to the remaining subscripts
  * 
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::fixupL(const Index n, const IndexVector& perm_r, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::fixupL(const Index n, const IndexVector& perm_r, GlobalLU_t& glu)
 {
   Index fsupc, i, j, k, jstart; 
   
-  Index nextl = 0; 
+  StorageIndex nextl = 0; 
   Index nsuper = (glu.supno)(n); 
   
   // For each supernode 
diff --git a/Eigen/src/SparseLU/SparseLU_column_bmod.h b/Eigen/src/SparseLU/SparseLU_column_bmod.h
index f24bd87..b57f068 100644
--- a/Eigen/src/SparseLU/SparseLU_column_bmod.h
+++ b/Eigen/src/SparseLU/SparseLU_column_bmod.h
@@ -49,8 +49,9 @@
  *         > 0 - number of bytes allocated when run out of space
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::column_bmod(const Index jcol, const Index nseg, BlockScalarVector dense, ScalarVector& tempv, BlockIndexVector segrep, BlockIndexVector repfnz, Index fpanelc, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::column_bmod(const Index jcol, const Index nseg, BlockScalarVector dense, ScalarVector& tempv,
+                                                     BlockIndexVector segrep, BlockIndexVector repfnz, Index fpanelc, GlobalLU_t& glu)
 {
   Index  jsupno, k, ksub, krep, ksupno; 
   Index lptr, nrow, isub, irow, nextlu, new_next, ufirst; 
@@ -137,7 +138,7 @@
     glu.lusup.segment(nextlu,offset).setZero();
     nextlu += offset;
   }
-  glu.xlusup(jcol + 1) = nextlu;  // close L\U(*,jcol); 
+  glu.xlusup(jcol + 1) = StorageIndex(nextlu);  // close L\U(*,jcol); 
   
   /* For more updates within the panel (also within the current supernode),
    * should start from the first column of the panel, or the first column
@@ -162,11 +163,11 @@
     // points to the beginning of jcol in snode L\U(jsupno) 
     ufirst = glu.xlusup(jcol) + d_fsupc; 
     Index lda = glu.xlusup(jcol+1) - glu.xlusup(jcol);
-    Map<Matrix<Scalar,Dynamic,Dynamic>, 0,  OuterStride<> > A( &(glu.lusup.data()[luptr]), nsupc, nsupc, OuterStride<>(lda) ); 
+    MappedMatrixBlock A( &(glu.lusup.data()[luptr]), nsupc, nsupc, OuterStride<>(lda) );
     VectorBlock<ScalarVector> u(glu.lusup, ufirst, nsupc); 
     u = A.template triangularView<UnitLower>().solve(u); 
     
-    new (&A) Map<Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > ( &(glu.lusup.data()[luptr+nsupc]), nrow, nsupc, OuterStride<>(lda) ); 
+    new (&A) MappedMatrixBlock ( &(glu.lusup.data()[luptr+nsupc]), nrow, nsupc, OuterStride<>(lda) );
     VectorBlock<ScalarVector> l(glu.lusup, ufirst+nsupc, nrow); 
     l.noalias() -= A * u;
     
diff --git a/Eigen/src/SparseLU/SparseLU_column_dfs.h b/Eigen/src/SparseLU/SparseLU_column_dfs.h
index 4c04b0e..c98b30e 100644
--- a/Eigen/src/SparseLU/SparseLU_column_dfs.h
+++ b/Eigen/src/SparseLU/SparseLU_column_dfs.h
@@ -30,7 +30,7 @@
 #ifndef SPARSELU_COLUMN_DFS_H
 #define SPARSELU_COLUMN_DFS_H
 
-template <typename Scalar, typename Index> class SparseLUImpl;
+template <typename Scalar, typename StorageIndex> class SparseLUImpl;
 namespace Eigen {
 
 namespace internal {
@@ -39,8 +39,8 @@
 struct column_dfs_traits : no_assignment_operator
 {
   typedef typename ScalarVector::Scalar Scalar;
-  typedef typename IndexVector::Scalar Index;
-  column_dfs_traits(Index jcol, Index& jsuper, typename SparseLUImpl<Scalar, Index>::GlobalLU_t& glu, SparseLUImpl<Scalar, Index>& luImpl)
+  typedef typename IndexVector::Scalar StorageIndex;
+  column_dfs_traits(Index jcol, Index& jsuper, typename SparseLUImpl<Scalar, StorageIndex>::GlobalLU_t& glu, SparseLUImpl<Scalar, StorageIndex>& luImpl)
    : m_jcol(jcol), m_jsuper_ref(jsuper), m_glu(glu), m_luImpl(luImpl)
  {}
   bool update_segrep(Index /*krep*/, Index /*jj*/)
@@ -57,8 +57,8 @@
   
   Index m_jcol;
   Index& m_jsuper_ref;
-  typename SparseLUImpl<Scalar, Index>::GlobalLU_t& m_glu;
-  SparseLUImpl<Scalar, Index>& m_luImpl;
+  typename SparseLUImpl<Scalar, StorageIndex>::GlobalLU_t& m_glu;
+  SparseLUImpl<Scalar, StorageIndex>& m_luImpl;
 };
 
 
@@ -89,8 +89,10 @@
  *         > 0 number of bytes allocated when run out of space
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::column_dfs(const Index m, const Index jcol, IndexVector& perm_r, Index maxsuper, Index& nseg,  BlockIndexVector lsub_col, IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune, IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::column_dfs(const Index m, const Index jcol, IndexVector& perm_r, Index maxsuper, Index& nseg,
+                                                    BlockIndexVector lsub_col, IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune,
+                                                    IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
 {
   
   Index jsuper = glu.supno(jcol); 
@@ -110,13 +112,13 @@
     // krow was visited before, go to the next nonz; 
     if (kmark == jcol) continue;
     
-    dfs_kernel(jcol, perm_r, nseg, glu.lsub, segrep, repfnz, xprune, marker2, parent,
+    dfs_kernel(StorageIndex(jcol), perm_r, nseg, glu.lsub, segrep, repfnz, xprune, marker2, parent,
                    xplore, glu, nextl, krow, traits);
   } // for each nonzero ... 
   
-  Index fsupc, jptr, jm1ptr, ito, ifrom, istop;
-  Index nsuper = glu.supno(jcol);
-  Index jcolp1 = jcol + 1;
+  Index fsupc;
+  StorageIndex nsuper = glu.supno(jcol);
+  StorageIndex jcolp1 = StorageIndex(jcol) + 1;
   Index jcolm1 = jcol - 1;
   
   // check to see if j belongs in the same supernode as j-1
@@ -127,8 +129,8 @@
   else 
   {
     fsupc = glu.xsup(nsuper); 
-    jptr = glu.xlsub(jcol); // Not yet compressed
-    jm1ptr = glu.xlsub(jcolm1); 
+    StorageIndex jptr = glu.xlsub(jcol); // Not yet compressed
+    StorageIndex jm1ptr = glu.xlsub(jcolm1); 
     
     // Use supernodes of type T2 : see SuperLU paper
     if ( (nextl-jptr != jptr-jm1ptr-1) ) jsuper = emptyIdxLU;
@@ -146,13 +148,13 @@
     { // starts a new supernode 
       if ( (fsupc < jcolm1-1) ) 
       { // >= 3 columns in nsuper
-        ito = glu.xlsub(fsupc+1);
+        StorageIndex ito = glu.xlsub(fsupc+1);
         glu.xlsub(jcolm1) = ito; 
-        istop = ito + jptr - jm1ptr; 
+        StorageIndex istop = ito + jptr - jm1ptr; 
         xprune(jcolm1) = istop; // intialize xprune(jcol-1)
         glu.xlsub(jcol) = istop; 
         
-        for (ifrom = jm1ptr; ifrom < nextl; ++ifrom, ++ito)
+        for (StorageIndex ifrom = jm1ptr; ifrom < nextl; ++ifrom, ++ito)
           glu.lsub(ito) = glu.lsub(ifrom); 
         nextl = ito;  // = istop + length(jcol)
       }
@@ -164,8 +166,8 @@
   // Tidy up the pointers before exit
   glu.xsup(nsuper+1) = jcolp1; 
   glu.supno(jcolp1) = nsuper; 
-  xprune(jcol) = nextl;  // Intialize upper bound for pruning
-  glu.xlsub(jcolp1) = nextl; 
+  xprune(jcol) = StorageIndex(nextl);  // Intialize upper bound for pruning
+  glu.xlsub(jcolp1) = StorageIndex(nextl); 
   
   return 0; 
 }
diff --git a/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h b/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h
index 170610d..c32d8d8 100644
--- a/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h
+++ b/Eigen/src/SparseLU/SparseLU_copy_to_ucol.h
@@ -46,8 +46,9 @@
  *         > 0 - number of bytes allocated when run out of space
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::copy_to_ucol(const Index jcol, const Index nseg, IndexVector& segrep, BlockIndexVector repfnz ,IndexVector& perm_r, BlockScalarVector dense, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::copy_to_ucol(const Index jcol, const Index nseg, IndexVector& segrep,
+                                                      BlockIndexVector repfnz ,IndexVector& perm_r, BlockScalarVector dense, GlobalLU_t& glu)
 {  
   Index ksub, krep, ksupno; 
     
@@ -55,7 +56,7 @@
   
   // For each nonzero supernode segment of U[*,j] in topological order 
   Index k = nseg - 1, i; 
-  Index nextu = glu.xusub(jcol); 
+  StorageIndex nextu = glu.xusub(jcol); 
   Index kfnz, isub, segsize; 
   Index new_next,irow; 
   Index fsupc, mem; 
diff --git a/Eigen/src/SparseLU/SparseLU_gemm_kernel.h b/Eigen/src/SparseLU/SparseLU_gemm_kernel.h
index 9e4e3e7..95ba741 100644
--- a/Eigen/src/SparseLU/SparseLU_gemm_kernel.h
+++ b/Eigen/src/SparseLU/SparseLU_gemm_kernel.h
@@ -21,7 +21,7 @@
   *  - lda and ldc must be multiples of the respective packet size
   *  - C must have the same alignment as A
   */
-template<typename Scalar,typename Index>
+template<typename Scalar>
 EIGEN_DONT_INLINE
 void sparselu_gemm(Index m, Index n, Index d, const Scalar* A, Index lda, const Scalar* B, Index ldb, Scalar* C, Index ldc)
 {
@@ -39,9 +39,9 @@
   };
   Index d_end = (d/RK)*RK;    // number of columns of A (rows of B) suitable for full register blocking
   Index n_end = (n/RN)*RN;    // number of columns of B-C suitable for processing RN columns at once
-  Index i0 = internal::first_aligned(A,m);
+  Index i0 = internal::first_default_aligned(A,m);
   
-  eigen_internal_assert(((lda%PacketSize)==0) && ((ldc%PacketSize)==0) && (i0==internal::first_aligned(C,m)));
+  eigen_internal_assert(((lda%PacketSize)==0) && ((ldc%PacketSize)==0) && (i0==internal::first_default_aligned(C,m)));
   
   // handle the non aligned rows of A and C without any optimization:
   for(Index i=0; i<i0; ++i)
@@ -72,14 +72,14 @@
         
         // load and expand a RN x RK block of B
         Packet b00, b10, b20, b30, b01, b11, b21, b31;
-                  b00 = pset1<Packet>(Bc0[0]);
-                  b10 = pset1<Packet>(Bc0[1]);
-        if(RK==4) b20 = pset1<Packet>(Bc0[2]);
-        if(RK==4) b30 = pset1<Packet>(Bc0[3]);
-                  b01 = pset1<Packet>(Bc1[0]);
-                  b11 = pset1<Packet>(Bc1[1]);
-        if(RK==4) b21 = pset1<Packet>(Bc1[2]);
-        if(RK==4) b31 = pset1<Packet>(Bc1[3]);
+                  { b00 = pset1<Packet>(Bc0[0]); }
+                  { b10 = pset1<Packet>(Bc0[1]); }
+        if(RK==4) { b20 = pset1<Packet>(Bc0[2]); }
+        if(RK==4) { b30 = pset1<Packet>(Bc0[3]); }
+                  { b01 = pset1<Packet>(Bc1[0]); }
+                  { b11 = pset1<Packet>(Bc1[1]); }
+        if(RK==4) { b21 = pset1<Packet>(Bc1[2]); }
+        if(RK==4) { b31 = pset1<Packet>(Bc1[3]); }
         
         Packet a0, a1, a2, a3, c0, c1, t0, t1;
         
@@ -106,22 +106,22 @@
         
 #define KMADD(c, a, b, tmp) {tmp = b; tmp = pmul(a,tmp); c = padd(c,tmp);}
 #define WORK(I)  \
-                    c0 = pload<Packet>(C0+i+(I)*PacketSize);   \
-                    c1 = pload<Packet>(C1+i+(I)*PacketSize);   \
-                    KMADD(c0, a0, b00, t0)      \
-                    KMADD(c1, a0, b01, t1)      \
-                    a0 = pload<Packet>(A0+i+(I+1)*PacketSize); \
-                    KMADD(c0, a1, b10, t0)      \
-                    KMADD(c1, a1, b11, t1)       \
-                    a1 = pload<Packet>(A1+i+(I+1)*PacketSize); \
-          if(RK==4) KMADD(c0, a2, b20, t0)       \
-          if(RK==4) KMADD(c1, a2, b21, t1)       \
-          if(RK==4) a2 = pload<Packet>(A2+i+(I+1)*PacketSize); \
-          if(RK==4) KMADD(c0, a3, b30, t0)       \
-          if(RK==4) KMADD(c1, a3, b31, t1)       \
-          if(RK==4) a3 = pload<Packet>(A3+i+(I+1)*PacketSize); \
-                    pstore(C0+i+(I)*PacketSize, c0);           \
-                    pstore(C1+i+(I)*PacketSize, c1)
+                     c0 = pload<Packet>(C0+i+(I)*PacketSize);    \
+                     c1 = pload<Packet>(C1+i+(I)*PacketSize);    \
+                     KMADD(c0, a0, b00, t0)                      \
+                     KMADD(c1, a0, b01, t1)                      \
+                     a0 = pload<Packet>(A0+i+(I+1)*PacketSize);  \
+                     KMADD(c0, a1, b10, t0)                      \
+                     KMADD(c1, a1, b11, t1)                      \
+                     a1 = pload<Packet>(A1+i+(I+1)*PacketSize);  \
+          if(RK==4){ KMADD(c0, a2, b20, t0)                     }\
+          if(RK==4){ KMADD(c1, a2, b21, t1)                     }\
+          if(RK==4){ a2 = pload<Packet>(A2+i+(I+1)*PacketSize); }\
+          if(RK==4){ KMADD(c0, a3, b30, t0)                     }\
+          if(RK==4){ KMADD(c1, a3, b31, t1)                     }\
+          if(RK==4){ a3 = pload<Packet>(A3+i+(I+1)*PacketSize); }\
+                     pstore(C0+i+(I)*PacketSize, c0);            \
+                     pstore(C1+i+(I)*PacketSize, c1)
         
         // process rows of A' - C' with aggressive vectorization and peeling 
         for(Index i=0; i<actual_b_end1; i+=PacketSize*8)
@@ -131,14 +131,15 @@
                     prefetch((A1+i+(5)*PacketSize));
           if(RK==4) prefetch((A2+i+(5)*PacketSize));
           if(RK==4) prefetch((A3+i+(5)*PacketSize));
-                    WORK(0);
-                    WORK(1);
-                    WORK(2);
-                    WORK(3);
-                    WORK(4);
-                    WORK(5);
-                    WORK(6);
-                    WORK(7);
+
+          WORK(0);
+          WORK(1);
+          WORK(2);
+          WORK(3);
+          WORK(4);
+          WORK(5);
+          WORK(6);
+          WORK(7);
         }
         // process the remaining rows with vectorization only
         for(Index i=actual_b_end1; i<actual_b_end2; i+=PacketSize)
@@ -165,7 +166,7 @@
         Bc1 += RK;
       } // peeled loop on k
     } // peeled loop on the columns j
-    // process the last column (we now perform a matrux-vector product)
+    // process the last column (we now perform a matrix-vector product)
     if((n-n_end)>0)
     {
       const Scalar* Bc0 = B+(n-1)*ldb;
@@ -203,16 +204,16 @@
         }
         
 #define WORK(I) \
-                  c0 = pload<Packet>(C0+i+(I)*PacketSize);   \
-                  KMADD(c0, a0, b00, t0)       \
-                  a0 = pload<Packet>(A0+i+(I+1)*PacketSize); \
-                  KMADD(c0, a1, b10, t0)       \
-                  a1 = pload<Packet>(A1+i+(I+1)*PacketSize); \
-        if(RK==4) KMADD(c0, a2, b20, t0)       \
-        if(RK==4) a2 = pload<Packet>(A2+i+(I+1)*PacketSize); \
-        if(RK==4) KMADD(c0, a3, b30, t0)       \
-        if(RK==4) a3 = pload<Packet>(A3+i+(I+1)*PacketSize); \
-                  pstore(C0+i+(I)*PacketSize, c0);
+                   c0 = pload<Packet>(C0+i+(I)*PacketSize);     \
+                   KMADD(c0, a0, b00, t0)                       \
+                   a0 = pload<Packet>(A0+i+(I+1)*PacketSize);   \
+                   KMADD(c0, a1, b10, t0)                       \
+                   a1 = pload<Packet>(A1+i+(I+1)*PacketSize);   \
+        if(RK==4){ KMADD(c0, a2, b20, t0)                      }\
+        if(RK==4){ a2 = pload<Packet>(A2+i+(I+1)*PacketSize);  }\
+        if(RK==4){ KMADD(c0, a3, b30, t0)                      }\
+        if(RK==4){ a3 = pload<Packet>(A3+i+(I+1)*PacketSize);  }\
+                   pstore(C0+i+(I)*PacketSize, c0);
         
         // agressive vectorization and peeling
         for(Index i=0; i<actual_b_end1; i+=PacketSize*8)
diff --git a/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h b/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h
index 7a4e430..6f75d50 100644
--- a/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h
+++ b/Eigen/src/SparseLU/SparseLU_heap_relax_snode.h
@@ -42,21 +42,20 @@
  * \param descendants Number of descendants of each node in the etree
  * \param relax_end last column in a supernode
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::heap_relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::heap_relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
 {
   
   // The etree may not be postordered, but its heap ordered  
   IndexVector post;
-  internal::treePostorder(n, et, post); // Post order etree
+  internal::treePostorder(StorageIndex(n), et, post); // Post order etree
   IndexVector inv_post(n+1); 
-  Index i;
-  for (i = 0; i < n+1; ++i) inv_post(post(i)) = i; // inv_post = post.inverse()???
+  for (StorageIndex i = 0; i < n+1; ++i) inv_post(post(i)) = i; // inv_post = post.inverse()???
   
   // Renumber etree in postorder 
   IndexVector iwork(n);
   IndexVector et_save(n+1);
-  for (i = 0; i < n; ++i)
+  for (Index i = 0; i < n; ++i)
   {
     iwork(post(i)) = post(et(i));
   }
@@ -75,10 +74,10 @@
   }
   // Identify the relaxed supernodes by postorder traversal of the etree
   Index snode_start; // beginning of a snode 
-  Index k;
+  StorageIndex k;
   Index nsuper_et_post = 0; // Number of relaxed snodes in postordered etree 
   Index nsuper_et = 0; // Number of relaxed snodes in the original etree 
-  Index l; 
+  StorageIndex l; 
   for (j = 0; j < n; )
   {
     parent = et(j);
@@ -90,8 +89,8 @@
     }
     // Found a supernode in postordered etree, j is the last column 
     ++nsuper_et_post;
-    k = n;
-    for (i = snode_start; i <= j; ++i)
+    k = StorageIndex(n);
+    for (Index i = snode_start; i <= j; ++i)
       k = (std::min)(k, inv_post(i));
     l = inv_post(j);
     if ( (l - k) == (j - snode_start) )  // Same number of columns in the snode
@@ -102,7 +101,7 @@
     }
     else 
     {
-      for (i = snode_start; i <= j; ++i) 
+      for (Index i = snode_start; i <= j; ++i) 
       {
         l = inv_post(i);
         if (descendants(i) == 0) 
diff --git a/Eigen/src/SparseLU/SparseLU_kernel_bmod.h b/Eigen/src/SparseLU/SparseLU_kernel_bmod.h
index 0d0283b..8c1b3e8 100644
--- a/Eigen/src/SparseLU/SparseLU_kernel_bmod.h
+++ b/Eigen/src/SparseLU/SparseLU_kernel_bmod.h
@@ -14,30 +14,29 @@
 namespace Eigen {
 namespace internal {
   
-/**
- * \brief Performs numeric block updates from a given supernode to a single column
- * 
- * \param segsize Size of the segment (and blocks ) to use for updates
- * \param[in,out] dense Packed values of the original matrix
- * \param tempv temporary vector to use for updates
- * \param lusup array containing the supernodes
- * \param lda Leading dimension in the supernode
- * \param nrow Number of rows in the rectangular part of the supernode
- * \param lsub compressed row subscripts of supernodes
- * \param lptr pointer to the first column of the current supernode in lsub
- * \param no_zeros Number of nonzeros elements before the diagonal part of the supernode
- * \return 0 on success
- */
 template <int SegSizeAtCompileTime> struct LU_kernel_bmod
 {
-  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-  static EIGEN_DONT_INLINE void run(const int segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
+  /** \internal
+    * \brief Performs numeric block updates from a given supernode to a single column
+    *
+    * \param segsize Size of the segment (and blocks ) to use for updates
+    * \param[in,out] dense Packed values of the original matrix
+    * \param tempv temporary vector to use for updates
+    * \param lusup array containing the supernodes
+    * \param lda Leading dimension in the supernode
+    * \param nrow Number of rows in the rectangular part of the supernode
+    * \param lsub compressed row subscripts of supernodes
+    * \param lptr pointer to the first column of the current supernode in lsub
+    * \param no_zeros Number of nonzeros elements before the diagonal part of the supernode
+    */
+  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+  static EIGEN_DONT_INLINE void run(const Index segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
                                     const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros);
 };
 
 template <int SegSizeAtCompileTime>
-template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-EIGEN_DONT_INLINE void LU_kernel_bmod<SegSizeAtCompileTime>::run(const int segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
+template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+EIGEN_DONT_INLINE void LU_kernel_bmod<SegSizeAtCompileTime>::run(const Index segsize, BlockScalarVector& dense, ScalarVector& tempv, ScalarVector& lusup, Index& luptr, const Index lda,
                                                                   const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros)
 {
   typedef typename ScalarVector::Scalar Scalar;
@@ -45,7 +44,7 @@
   // The result of triangular solve is in tempv[*]; 
     // The result of matric-vector update is in dense[*]
   Index isub = lptr + no_zeros; 
-  int i;
+  Index i;
   Index irow;
   for (i = 0; i < ((SegSizeAtCompileTime==Dynamic)?segsize:SegSizeAtCompileTime); i++)
   {
@@ -56,7 +55,7 @@
   // Dense triangular solve -- start effective triangle
   luptr += lda * no_zeros + no_zeros; 
   // Form Eigen matrix and vector 
-  Map<Matrix<Scalar,SegSizeAtCompileTime,SegSizeAtCompileTime>, 0, OuterStride<> > A( &(lusup.data()[luptr]), segsize, segsize, OuterStride<>(lda) );
+  Map<Matrix<Scalar,SegSizeAtCompileTime,SegSizeAtCompileTime, ColMajor>, 0, OuterStride<> > A( &(lusup.data()[luptr]), segsize, segsize, OuterStride<>(lda) );
   Map<Matrix<Scalar,SegSizeAtCompileTime,1> > u(tempv.data(), segsize);
   
   u = A.template triangularView<UnitLower>().solve(u); 
@@ -65,9 +64,9 @@
   luptr += segsize;
   const Index PacketSize = internal::packet_traits<Scalar>::size;
   Index ldl = internal::first_multiple(nrow, PacketSize);
-  Map<Matrix<Scalar,Dynamic,SegSizeAtCompileTime>, 0, OuterStride<> > B( &(lusup.data()[luptr]), nrow, segsize, OuterStride<>(lda) );
-  Index aligned_offset = internal::first_aligned(tempv.data()+segsize, PacketSize);
-  Index aligned_with_B_offset = (PacketSize-internal::first_aligned(B.data(), PacketSize))%PacketSize;
+  Map<Matrix<Scalar,Dynamic,SegSizeAtCompileTime, ColMajor>, 0, OuterStride<> > B( &(lusup.data()[luptr]), nrow, segsize, OuterStride<>(lda) );
+  Index aligned_offset = internal::first_default_aligned(tempv.data()+segsize, PacketSize);
+  Index aligned_with_B_offset = (PacketSize-internal::first_default_aligned(B.data(), PacketSize))%PacketSize;
   Map<Matrix<Scalar,Dynamic,1>, 0, OuterStride<> > l(tempv.data()+segsize+aligned_offset+aligned_with_B_offset, nrow, OuterStride<>(ldl) );
   
   l.setZero();
@@ -91,21 +90,22 @@
 
 template <> struct LU_kernel_bmod<1>
 {
-  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-  static EIGEN_DONT_INLINE void run(const int /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
+  template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+  static EIGEN_DONT_INLINE void run(const Index /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
                                     const Index lda, const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros);
 };
 
 
-template <typename BlockScalarVector, typename ScalarVector, typename IndexVector, typename Index>
-EIGEN_DONT_INLINE void LU_kernel_bmod<1>::run(const int /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
+template <typename BlockScalarVector, typename ScalarVector, typename IndexVector>
+EIGEN_DONT_INLINE void LU_kernel_bmod<1>::run(const Index /*segsize*/, BlockScalarVector& dense, ScalarVector& /*tempv*/, ScalarVector& lusup, Index& luptr,
                                               const Index lda, const Index nrow, IndexVector& lsub, const Index lptr, const Index no_zeros)
 {
   typedef typename ScalarVector::Scalar Scalar;
+  typedef typename IndexVector::Scalar StorageIndex;
   Scalar f = dense(lsub(lptr + no_zeros));
   luptr += lda * no_zeros + no_zeros + 1;
   const Scalar* a(lusup.data() + luptr);
-  const /*typename IndexVector::Scalar*/Index*  irow(lsub.data()+lptr + no_zeros + 1);
+  const StorageIndex*  irow(lsub.data()+lptr + no_zeros + 1);
   Index i = 0;
   for (; i+1 < nrow; i+=2)
   {
diff --git a/Eigen/src/SparseLU/SparseLU_panel_bmod.h b/Eigen/src/SparseLU/SparseLU_panel_bmod.h
index da0e0fc..822cf32 100644
--- a/Eigen/src/SparseLU/SparseLU_panel_bmod.h
+++ b/Eigen/src/SparseLU/SparseLU_panel_bmod.h
@@ -52,8 +52,8 @@
  * 
  * 
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::panel_bmod(const Index m, const Index w, const Index jcol, 
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::panel_bmod(const Index m, const Index w, const Index jcol, 
                                             const Index nseg, ScalarVector& dense, ScalarVector& tempv,
                                             IndexVector& segrep, IndexVector& repfnz, GlobalLU_t& glu)
 {
@@ -102,7 +102,7 @@
     if(nsupc >= 2)
     { 
       Index ldu = internal::first_multiple<Index>(u_rows, PacketSize);
-      Map<Matrix<Scalar,Dynamic,Dynamic>, Aligned,  OuterStride<> > U(tempv.data(), u_rows, u_cols, OuterStride<>(ldu));
+      Map<ScalarMatrix, Aligned,  OuterStride<> > U(tempv.data(), u_rows, u_cols, OuterStride<>(ldu));
       
       // gather U
       Index u_col = 0;
@@ -136,17 +136,17 @@
       Index lda = glu.xlusup(fsupc+1) - glu.xlusup(fsupc);
       no_zeros = (krep - u_rows + 1) - fsupc;
       luptr += lda * no_zeros + no_zeros;
-      Map<Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > A(glu.lusup.data()+luptr, u_rows, u_rows, OuterStride<>(lda) );
+      MappedMatrixBlock A(glu.lusup.data()+luptr, u_rows, u_rows, OuterStride<>(lda) );
       U = A.template triangularView<UnitLower>().solve(U);
       
       // update
       luptr += u_rows;
-      Map<Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > B(glu.lusup.data()+luptr, nrow, u_rows, OuterStride<>(lda) );
+      MappedMatrixBlock B(glu.lusup.data()+luptr, nrow, u_rows, OuterStride<>(lda) );
       eigen_assert(tempv.size()>w*ldu + nrow*w + 1);
       
       Index ldl = internal::first_multiple<Index>(nrow, PacketSize);
-      Index offset = (PacketSize-internal::first_aligned(B.data(), PacketSize)) % PacketSize;
-      Map<Matrix<Scalar,Dynamic,Dynamic>, 0, OuterStride<> > L(tempv.data()+w*ldu+offset, nrow, u_cols, OuterStride<>(ldl));
+      Index offset = (PacketSize-internal::first_default_aligned(B.data(), PacketSize)) % PacketSize;
+      MappedMatrixBlock L(tempv.data()+w*ldu+offset, nrow, u_cols, OuterStride<>(ldl));
       
       L.setZero();
       internal::sparselu_gemm<Scalar>(L.rows(), L.cols(), B.cols(), B.data(), B.outerStride(), U.data(), U.outerStride(), L.data(), L.outerStride());
diff --git a/Eigen/src/SparseLU/SparseLU_panel_dfs.h b/Eigen/src/SparseLU/SparseLU_panel_dfs.h
index dc0054e..155df73 100644
--- a/Eigen/src/SparseLU/SparseLU_panel_dfs.h
+++ b/Eigen/src/SparseLU/SparseLU_panel_dfs.h
@@ -37,11 +37,11 @@
 template<typename IndexVector>
 struct panel_dfs_traits
 {
-  typedef typename IndexVector::Scalar Index;
-  panel_dfs_traits(Index jcol, Index* marker)
+  typedef typename IndexVector::Scalar StorageIndex;
+  panel_dfs_traits(Index jcol, StorageIndex* marker)
     : m_jcol(jcol), m_marker(marker)
   {}
-  bool update_segrep(Index krep, Index jj)
+  bool update_segrep(Index krep, StorageIndex jj)
   {
     if(m_marker[krep]<m_jcol)
     {
@@ -53,13 +53,13 @@
   void mem_expand(IndexVector& /*glu.lsub*/, Index /*nextl*/, Index /*chmark*/) {}
   enum { ExpandMem = false };
   Index m_jcol;
-  Index* m_marker;
+  StorageIndex* m_marker;
 };
 
 
-template <typename Scalar, typename Index>
+template <typename Scalar, typename StorageIndex>
 template <typename Traits>
-void SparseLUImpl<Scalar,Index>::dfs_kernel(const Index jj, IndexVector& perm_r,
+void SparseLUImpl<Scalar,StorageIndex>::dfs_kernel(const StorageIndex jj, IndexVector& perm_r,
                    Index& nseg, IndexVector& panel_lsub, IndexVector& segrep,
                    Ref<IndexVector> repfnz_col, IndexVector& xprune, Ref<IndexVector> marker, IndexVector& parent,
                    IndexVector& xplore, GlobalLU_t& glu,
@@ -67,14 +67,14 @@
                   )
 {
   
-  Index kmark = marker(krow);
+  StorageIndex kmark = marker(krow);
       
   // For each unmarked krow of jj
   marker(krow) = jj; 
-  Index kperm = perm_r(krow); 
+  StorageIndex kperm = perm_r(krow); 
   if (kperm == emptyIdxLU ) {
     // krow is in L : place it in structure of L(*, jj)
-    panel_lsub(nextl_col++) = krow;  // krow is indexed into A
+    panel_lsub(nextl_col++) = StorageIndex(krow);  // krow is indexed into A
     
     traits.mem_expand(panel_lsub, nextl_col, kmark);
   }
@@ -83,9 +83,9 @@
     // krow is in U : if its supernode-representative krep
     // has been explored, update repfnz(*)
     // krep = supernode representative of the current row
-    Index krep = glu.xsup(glu.supno(kperm)+1) - 1; 
+    StorageIndex krep = glu.xsup(glu.supno(kperm)+1) - 1; 
     // First nonzero element in the current column:
-    Index myfnz = repfnz_col(krep); 
+    StorageIndex myfnz = repfnz_col(krep); 
     
     if (myfnz != emptyIdxLU )
     {
@@ -96,26 +96,26 @@
     else 
     {
       // Otherwise, perform dfs starting at krep
-      Index oldrep = emptyIdxLU; 
+      StorageIndex oldrep = emptyIdxLU; 
       parent(krep) = oldrep; 
       repfnz_col(krep) = kperm; 
-      Index xdfs =  glu.xlsub(krep); 
+      StorageIndex xdfs =  glu.xlsub(krep); 
       Index maxdfs = xprune(krep); 
       
-      Index kpar;
+      StorageIndex kpar;
       do 
       {
         // For each unmarked kchild of krep
         while (xdfs < maxdfs) 
         {
-          Index kchild = glu.lsub(xdfs); 
+          StorageIndex kchild = glu.lsub(xdfs); 
           xdfs++; 
-          Index chmark = marker(kchild); 
+          StorageIndex chmark = marker(kchild); 
           
           if (chmark != jj ) 
           {
             marker(kchild) = jj; 
-            Index chperm = perm_r(kchild); 
+            StorageIndex chperm = perm_r(kchild); 
             
             if (chperm == emptyIdxLU) 
             {
@@ -128,7 +128,7 @@
               // case kchild is in U :
               // chrep = its supernode-rep. If its rep has been explored, 
               // update its repfnz(*)
-              Index chrep = glu.xsup(glu.supno(chperm)+1) - 1; 
+              StorageIndex chrep = glu.xsup(glu.supno(chperm)+1) - 1; 
               myfnz = repfnz_col(chrep); 
               
               if (myfnz != emptyIdxLU) 
@@ -215,8 +215,8 @@
  * 
  */
 
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::panel_dfs(const Index m, const Index w, const Index jcol, MatrixType& A, IndexVector& perm_r, Index& nseg, ScalarVector& dense, IndexVector& panel_lsub, IndexVector& segrep, IndexVector& repfnz, IndexVector& xprune, IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::panel_dfs(const Index m, const Index w, const Index jcol, MatrixType& A, IndexVector& perm_r, Index& nseg, ScalarVector& dense, IndexVector& panel_lsub, IndexVector& segrep, IndexVector& repfnz, IndexVector& xprune, IndexVector& marker, IndexVector& parent, IndexVector& xplore, GlobalLU_t& glu)
 {
   Index nextl_col; // Next available position in panel_lsub[*,jj] 
   
@@ -227,7 +227,7 @@
   panel_dfs_traits<IndexVector> traits(jcol, marker1.data());
   
   // For each column in the panel 
-  for (Index jj = jcol; jj < jcol + w; jj++) 
+  for (StorageIndex jj = StorageIndex(jcol); jj < jcol + w; jj++) 
   {
     nextl_col = (jj - jcol) * m; 
     
@@ -241,7 +241,7 @@
       Index krow = it.row(); 
       dense_col(krow) = it.value();
       
-      Index kmark = marker(krow); 
+      StorageIndex kmark = marker(krow); 
       if (kmark == jj) 
         continue; // krow visited before, go to the next nonzero
       
diff --git a/Eigen/src/SparseLU/SparseLU_pivotL.h b/Eigen/src/SparseLU/SparseLU_pivotL.h
index 457789c..a86dac9 100644
--- a/Eigen/src/SparseLU/SparseLU_pivotL.h
+++ b/Eigen/src/SparseLU/SparseLU_pivotL.h
@@ -56,8 +56,8 @@
  * \return 0 if success, i > 0 if U(i,i) is exactly zero 
  * 
  */
-template <typename Scalar, typename Index>
-Index SparseLUImpl<Scalar,Index>::pivotL(const Index jcol, const RealScalar& diagpivotthresh, IndexVector& perm_r, IndexVector& iperm_c, Index& pivrow, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+Index SparseLUImpl<Scalar,StorageIndex>::pivotL(const Index jcol, const RealScalar& diagpivotthresh, IndexVector& perm_r, IndexVector& iperm_c, Index& pivrow, GlobalLU_t& glu)
 {
   
   Index fsupc = (glu.xsup)((glu.supno)(jcol)); // First column in the supernode containing the column jcol
@@ -67,11 +67,11 @@
   Index lda = glu.xlusup(fsupc+1) - glu.xlusup(fsupc); // leading dimension
   Scalar* lu_sup_ptr = &(glu.lusup.data()[glu.xlusup(fsupc)]); // Start of the current supernode
   Scalar* lu_col_ptr = &(glu.lusup.data()[glu.xlusup(jcol)]); // Start of jcol in the supernode
-  Index* lsub_ptr = &(glu.lsub.data()[lptr]); // Start of row indices of the supernode
+  StorageIndex* lsub_ptr = &(glu.lsub.data()[lptr]); // Start of row indices of the supernode
   
   // Determine the largest abs numerical value for partial pivoting 
   Index diagind = iperm_c(jcol); // diagonal index 
-  RealScalar pivmax = 0.0; 
+  RealScalar pivmax(-1.0);
   Index pivptr = nsupc; 
   Index diag = emptyIdxLU; 
   RealScalar rtemp;
@@ -87,9 +87,10 @@
   }
   
   // Test for singularity
-  if ( pivmax == 0.0 ) {
-    pivrow = lsub_ptr[pivptr];
-    perm_r(pivrow) = jcol;
+  if ( pivmax <= RealScalar(0.0) ) {
+    // if pivmax == -1, the column is structurally empty, otherwise it is only numerically zero
+    pivrow = pivmax < RealScalar(0.0) ? diagind : lsub_ptr[pivptr];
+    perm_r(pivrow) = StorageIndex(jcol);
     return (jcol+1);
   }
   
@@ -104,13 +105,13 @@
       // Diagonal element exists
       using std::abs;
       rtemp = abs(lu_col_ptr[diag]);
-      if (rtemp != 0.0 && rtemp >= thresh) pivptr = diag;
+      if (rtemp != RealScalar(0.0) && rtemp >= thresh) pivptr = diag;
     }
     pivrow = lsub_ptr[pivptr];
   }
   
   // Record pivot row
-  perm_r(pivrow) = jcol; 
+  perm_r(pivrow) = StorageIndex(jcol);
   // Interchange row subscripts
   if (pivptr != nsupc )
   {
diff --git a/Eigen/src/SparseLU/SparseLU_pruneL.h b/Eigen/src/SparseLU/SparseLU_pruneL.h
index 66460d1..ad32fed 100644
--- a/Eigen/src/SparseLU/SparseLU_pruneL.h
+++ b/Eigen/src/SparseLU/SparseLU_pruneL.h
@@ -49,8 +49,9 @@
  * \param glu Global LU data
  * 
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::pruneL(const Index jcol, const IndexVector& perm_r, const Index pivrow, const Index nseg, const IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune, GlobalLU_t& glu)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::pruneL(const Index jcol, const IndexVector& perm_r, const Index pivrow, const Index nseg,
+                                               const IndexVector& segrep, BlockIndexVector repfnz, IndexVector& xprune, GlobalLU_t& glu)
 {
   // For each supernode-rep irep in U(*,j]
   Index jsupno = glu.supno(jcol); 
@@ -123,7 +124,7 @@
           }
         } // end while 
         
-        xprune(irep) = kmin;  //Pruning 
+        xprune(irep) = StorageIndex(kmin);  //Pruning 
       } // end if do_prune 
     } // end pruning 
   } // End for each U-segment
diff --git a/Eigen/src/SparseLU/SparseLU_relax_snode.h b/Eigen/src/SparseLU/SparseLU_relax_snode.h
index 58ec32e..c408d01 100644
--- a/Eigen/src/SparseLU/SparseLU_relax_snode.h
+++ b/Eigen/src/SparseLU/SparseLU_relax_snode.h
@@ -43,15 +43,15 @@
  * \param descendants Number of descendants of each node in the etree
  * \param relax_end last column in a supernode
  */
-template <typename Scalar, typename Index>
-void SparseLUImpl<Scalar,Index>::relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
+template <typename Scalar, typename StorageIndex>
+void SparseLUImpl<Scalar,StorageIndex>::relax_snode (const Index n, IndexVector& et, const Index relax_columns, IndexVector& descendants, IndexVector& relax_end)
 {
   
   // compute the number of descendants of each node in the etree
-  Index j, parent; 
+  Index parent; 
   relax_end.setConstant(emptyIdxLU);
   descendants.setZero();
-  for (j = 0; j < n; j++) 
+  for (Index j = 0; j < n; j++) 
   {
     parent = et(j);
     if (parent != n) // not the dummy root
@@ -59,7 +59,7 @@
   }
   // Identify the relaxed supernodes by postorder traversal of the etree
   Index snode_start; // beginning of a snode 
-  for (j = 0; j < n; )
+  for (Index j = 0; j < n; )
   {
     parent = et(j);
     snode_start = j; 
@@ -69,7 +69,7 @@
       parent = et(j);
     }
     // Found a supernode in postordered etree, j is the last column 
-    relax_end(snode_start) = j; // Record last column
+    relax_end(snode_start) = StorageIndex(j); // Record last column
     j++;
     // Search for a new leaf
     while (descendants(j) != 0 && j < n) j++;
diff --git a/Eigen/src/SparseQR/CMakeLists.txt b/Eigen/src/SparseQR/CMakeLists.txt
deleted file mode 100644
index f9ddf2b..0000000
--- a/Eigen/src/SparseQR/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_SparseQR_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_SparseQR_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/SparseQR/ COMPONENT Devel
-  )
diff --git a/Eigen/src/SparseQR/SparseQR.h b/Eigen/src/SparseQR/SparseQR.h
index a00bd5d..7409fca 100644
--- a/Eigen/src/SparseQR/SparseQR.h
+++ b/Eigen/src/SparseQR/SparseQR.h
@@ -21,8 +21,12 @@
   template <typename SparseQRType> struct traits<SparseQRMatrixQReturnType<SparseQRType> >
   {
     typedef typename SparseQRType::MatrixType ReturnType;
-    typedef typename ReturnType::Index Index;
+    typedef typename ReturnType::StorageIndex StorageIndex;
     typedef typename ReturnType::StorageKind StorageKind;
+    enum {
+      RowsAtCompileTime = Dynamic,
+      ColsAtCompileTime = Dynamic
+    };
   };
   template <typename SparseQRType> struct traits<SparseQRMatrixQTransposeReturnType<SparseQRType> >
   {
@@ -48,7 +52,7 @@
   * rank-revealing permutations. Use colsPermutation() to get it.
   * 
   * Q is the orthogonal matrix represented as products of Householder reflectors. 
-  * Use matrixQ() to get an expression and matrixQ().transpose() to get the transpose.
+  * Use matrixQ() to get an expression and matrixQ().adjoint() to get the adjoint.
   * You can then apply it to a vector.
   * 
   * R is the sparse triangular or trapezoidal matrix. The later occurs when A is rank-deficient.
@@ -58,24 +62,37 @@
   * \tparam _OrderingType The fill-reducing ordering method. See the \link OrderingMethods_Module 
   *  OrderingMethods \endlink module for the list of built-in and external ordering methods.
   * 
+  * \implsparsesolverconcept
+  *
   * \warning The input sparse matrix A must be in compressed mode (see SparseMatrix::makeCompressed()).
+  * \warning For complex matrices matrixQ().transpose() will actually return the adjoint matrix.
   * 
   */
 template<typename _MatrixType, typename _OrderingType>
-class SparseQR
+class SparseQR : public SparseSolverBase<SparseQR<_MatrixType,_OrderingType> >
 {
+  protected:
+    typedef SparseSolverBase<SparseQR<_MatrixType,_OrderingType> > Base;
+    using Base::m_isInitialized;
   public:
+    using Base::_solve_impl;
     typedef _MatrixType MatrixType;
     typedef _OrderingType OrderingType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
-    typedef SparseMatrix<Scalar,ColMajor,Index> QRMatrixType;
-    typedef Matrix<Index, Dynamic, 1> IndexVector;
+    typedef typename MatrixType::StorageIndex StorageIndex;
+    typedef SparseMatrix<Scalar,ColMajor,StorageIndex> QRMatrixType;
+    typedef Matrix<StorageIndex, Dynamic, 1> IndexVector;
     typedef Matrix<Scalar, Dynamic, 1> ScalarVector;
-    typedef PermutationMatrix<Dynamic, Dynamic, Index> PermutationType;
+    typedef PermutationMatrix<Dynamic, Dynamic, StorageIndex> PermutationType;
+
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
+    
   public:
-    SparseQR () : m_isInitialized(false), m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
+    SparseQR () :  m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
     { }
     
     /** Construct a QR factorization of the matrix \a mat.
@@ -84,7 +101,7 @@
       * 
       * \sa compute()
       */
-    SparseQR(const MatrixType& mat) : m_isInitialized(false), m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
+    explicit SparseQR(const MatrixType& mat) : m_analysisIsok(false), m_lastError(""), m_useDefaultThreshold(true),m_isQSorted(false),m_isEtreeOk(false)
     {
       compute(mat);
     }
@@ -112,6 +129,17 @@
     inline Index cols() const { return m_pmat.cols();}
     
     /** \returns a const reference to the \b sparse upper triangular matrix R of the QR factorization.
+      * \warning The entries of the returned matrix are not sorted. This means that using it in algorithms
+      *          expecting sorted entries will fail. This include random coefficient accesses (SpaseMatrix::coeff()),
+      *          and coefficient-wise operations. Matrix products and triangular solves are fine though.
+      *
+      * To sort the entries, you can assign it to a row-major matrix, and if a column-major matrix
+      * is required, you can copy it again:
+      * \code
+      * SparseMatrix<double>          R  = qr.matrixR();  // column-major, not sorted!
+      * SparseMatrix<double,RowMajor> Rr = qr.matrixR();  // row-major, sorted
+      * SparseMatrix<double>          Rc = Rr;            // column-major, sorted
+      * \endcode
       */
     const QRMatrixType& matrixR() const { return m_R; }
     
@@ -119,7 +147,7 @@
       *
       * \sa setPivotThreshold()
       */
-    Index rank() const 
+    Index rank() const
     {
       eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
       return m_nonzeropivots; 
@@ -162,23 +190,23 @@
     
     /** \internal */
     template<typename Rhs, typename Dest>
-    bool _solve(const MatrixBase<Rhs> &B, MatrixBase<Dest> &dest) const
+    bool _solve_impl(const MatrixBase<Rhs> &B, MatrixBase<Dest> &dest) const
     {
       eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
       eigen_assert(this->rows() == B.rows() && "SparseQR::solve() : invalid number of rows in the right hand side matrix");
 
       Index rank = this->rank();
       
-      // Compute Q^T * b;
+      // Compute Q^* * b;
       typename Dest::PlainObject y, b;
-      y = this->matrixQ().transpose() * B; 
+      y = this->matrixQ().adjoint() * B;
       b = y;
       
       // Solve with the triangular matrix R
-      y.resize((std::max)(cols(),Index(y.rows())),y.cols());
+      y.resize((std::max<Index>)(cols(),y.rows()),y.cols());
       y.topRows(rank) = this->matrixR().topLeftCorner(rank, rank).template triangularView<Upper>().solve(b.topRows(rank));
       y.bottomRows(y.rows()-rank).setZero();
-
+      
       // Apply the column permutation
       if (m_perm_c.size())  dest = colsPermutation() * y.topRows(cols());
       else                  dest = y.topRows(cols());
@@ -186,7 +214,6 @@
       m_info = Success;
       return true;
     }
-    
 
     /** Sets the threshold that is used to determine linearly dependent columns during the factorization.
       *
@@ -204,18 +231,18 @@
       * \sa compute()
       */
     template<typename Rhs>
-    inline const internal::solve_retval<SparseQR, Rhs> solve(const MatrixBase<Rhs>& B) const 
+    inline const Solve<SparseQR, Rhs> solve(const MatrixBase<Rhs>& B) const 
     {
       eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
       eigen_assert(this->rows() == B.rows() && "SparseQR::solve() : invalid number of rows in the right hand side matrix");
-      return internal::solve_retval<SparseQR, Rhs>(*this, B.derived());
+      return Solve<SparseQR, Rhs>(*this, B.derived());
     }
     template<typename Rhs>
-    inline const internal::sparse_solve_retval<SparseQR, Rhs> solve(const SparseMatrixBase<Rhs>& B) const
+    inline const Solve<SparseQR, Rhs> solve(const SparseMatrixBase<Rhs>& B) const
     {
           eigen_assert(m_isInitialized && "The factorization should be called first, use compute()");
           eigen_assert(this->rows() == B.rows() && "SparseQR::solve() : invalid number of rows in the right hand side matrix");
-          return internal::sparse_solve_retval<SparseQR, Rhs>(*this, B.derived());
+          return Solve<SparseQR, Rhs>(*this, B.derived());
     }
     
     /** \brief Reports whether previous computation was successful.
@@ -232,8 +259,9 @@
       return m_info;
     }
 
-  protected:
-    inline void sort_matrix_Q()
+
+    /** \internal */
+    inline void _sort_matrix_Q()
     {
       if(this->m_isQSorted) return;
       // The matrix Q is sorted during the transposition
@@ -244,7 +272,6 @@
 
     
   protected:
-    bool m_isInitialized;
     bool m_analysisIsok;
     bool m_factorizationIsok;
     mutable ComputationInfo m_info;
@@ -258,14 +285,13 @@
     PermutationType m_outputPerm_c; // The final column permutation
     RealScalar m_threshold;         // Threshold to determine null Householder reflections
     bool m_useDefaultThreshold;     // Use default threshold
-    Index m_nonzeropivots;          // Number of non zero pivots found 
+    Index m_nonzeropivots;          // Number of non zero pivots found
     IndexVector m_etree;            // Column elimination tree
     IndexVector m_firstRowElt;      // First element in each row
     bool m_isQSorted;               // whether Q is sorted or not
     bool m_isEtreeOk;               // whether the elimination tree match the initial input matrix
     
     template <typename, typename > friend struct SparseQR_QProduct;
-    template <typename > friend struct SparseQRMatrixQReturnType;
     
 };
 
@@ -294,7 +320,7 @@
   if (!m_perm_c.size())
   {
     m_perm_c.resize(n);
-    m_perm_c.indices().setLinSpaced(n, 0,n-1);
+    m_perm_c.indices().setLinSpaced(n, 0,StorageIndex(n-1));
   }
   
   // Compute the column elimination tree of the permuted matrix
@@ -323,12 +349,11 @@
 void SparseQR<MatrixType,OrderingType>::factorize(const MatrixType& mat)
 {
   using std::abs;
-  using std::max;
   
   eigen_assert(m_analysisIsok && "analyzePattern() should be called before this step");
-  Index m = mat.rows();
-  Index n = mat.cols();
-  Index diagSize = (std::min)(m,n);
+  StorageIndex m = StorageIndex(mat.rows());
+  StorageIndex n = StorageIndex(mat.cols());
+  StorageIndex diagSize = (std::min)(m,n);
   IndexVector mark((std::max)(m,n)); mark.setConstant(-1);  // Record the visited nodes
   IndexVector Ridx(n), Qidx(m);                             // Store temporarily the row indexes for the current column of R and Q
   Index nzcolR, nzcolQ;                                     // Number of nonzero for the current column of R and Q
@@ -353,7 +378,7 @@
     // otherwise directly use the input matrix
     // 
     IndexVector originalOuterIndicesCpy;
-    const Index *originalOuterIndices = mat.outerIndexPtr();
+    const StorageIndex *originalOuterIndices = mat.outerIndexPtr();
     if(MatrixType::IsRowMajor)
     {
       originalOuterIndicesCpy = IndexVector::Map(m_pmat.outerIndexPtr(),n+1);
@@ -375,7 +400,7 @@
   if(m_useDefaultThreshold) 
   {
     RealScalar max2Norm = 0.0;
-    for (int j = 0; j < n; j++) max2Norm = (max)(max2Norm, m_pmat.col(j).norm());
+    for (int j = 0; j < n; j++) max2Norm = numext::maxi(max2Norm, m_pmat.col(j).norm());
     if(max2Norm==RealScalar(0))
       max2Norm = RealScalar(1);
     pivotThreshold = 20 * (m + n) * max2Norm * NumTraits<RealScalar>::epsilon();
@@ -384,11 +409,11 @@
   // Initialize the numerical permutation
   m_pivotperm.setIdentity(n);
   
-  Index nonzeroCol = 0; // Record the number of valid pivots
+  StorageIndex nonzeroCol = 0; // Record the number of valid pivots
   m_Q.startVec(0);
 
   // Left looking rank-revealing QR factorization: compute a column of R and Q at a time
-  for (Index col = 0; col < n; ++col)
+  for (StorageIndex col = 0; col < n; ++col)
   {
     mark.setConstant(-1);
     m_R.startVec(col);
@@ -404,12 +429,12 @@
     // thus the trick with found_diag that permits to do one more iteration on the diagonal element if this one has not been found.
     for (typename QRMatrixType::InnerIterator itp(m_pmat, col); itp || !found_diag; ++itp)
     {
-      Index curIdx = nonzeroCol;
-      if(itp) curIdx = itp.row();
+      StorageIndex curIdx = nonzeroCol;
+      if(itp) curIdx = StorageIndex(itp.row());
       if(curIdx == nonzeroCol) found_diag = true;
       
       // Get the nonzeros indexes of the current column of R
-      Index st = m_firstRowElt(curIdx); // The traversal of the etree starts here 
+      StorageIndex st = m_firstRowElt(curIdx); // The traversal of the etree starts here
       if (st < 0 )
       {
         m_lastError = "Empty row found during numerical factorization";
@@ -466,7 +491,7 @@
       {
         for (typename QRMatrixType::InnerIterator itq(m_Q, curIdx); itq; ++itq)
         {
-          Index iQ = itq.row();
+          StorageIndex iQ = StorageIndex(itq.row());
           if (mark(iQ) != col)
           {
             Qidx(nzcolQ++) = iQ;  // Add this row to the pattern of Q,
@@ -476,7 +501,7 @@
       }
     } // End update current column
     
-    Scalar tau = 0;
+    Scalar tau = RealScalar(0);
     RealScalar beta = 0;
     
     if(nonzeroCol < diagSize)
@@ -572,44 +597,15 @@
   m_info = Success;
 }
 
-namespace internal {
-  
-template<typename _MatrixType, typename OrderingType, typename Rhs>
-struct solve_retval<SparseQR<_MatrixType,OrderingType>, Rhs>
-  : solve_retval_base<SparseQR<_MatrixType,OrderingType>, Rhs>
-{
-  typedef SparseQR<_MatrixType,OrderingType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-template<typename _MatrixType, typename OrderingType, typename Rhs>
-struct sparse_solve_retval<SparseQR<_MatrixType, OrderingType>, Rhs>
- : sparse_solve_retval_base<SparseQR<_MatrixType, OrderingType>, Rhs>
-{
-  typedef SparseQR<_MatrixType, OrderingType> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec, Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-} // end namespace internal
-
 template <typename SparseQRType, typename Derived>
 struct SparseQR_QProduct : ReturnByValue<SparseQR_QProduct<SparseQRType, Derived> >
 {
   typedef typename SparseQRType::QRMatrixType MatrixType;
   typedef typename SparseQRType::Scalar Scalar;
-  typedef typename SparseQRType::Index Index;
   // Get the references 
   SparseQR_QProduct(const SparseQRType& qr, const Derived& other, bool transpose) : 
   m_qr(qr),m_other(other),m_transpose(transpose) {}
-  inline Index rows() const { return m_transpose ? m_qr.rows() : m_qr.cols(); }
+  inline Index rows() const { return m_qr.matrixQ().rows(); }
   inline Index cols() const { return m_other.cols(); }
   
   // Assign to a vector
@@ -637,7 +633,10 @@
     }
     else
     {
-      eigen_assert(m_qr.m_Q.rows() == m_other.rows() && "Non conforming object sizes");
+      eigen_assert(m_qr.matrixQ().cols() == m_other.rows() && "Non conforming object sizes");
+
+      res.conservativeResize(rows(), cols());
+
       // Compute res = Q * other column by column
       for(Index j = 0; j < res.cols(); j++)
       {
@@ -646,7 +645,7 @@
           Scalar tau = Scalar(0);
           tau = m_qr.m_Q.col(k).dot(res.col(j));
           if(tau==Scalar(0)) continue;
-          tau = tau * m_qr.m_hcoeffs(k);
+          tau = tau * numext::conj(m_qr.m_hcoeffs(k));
           res.col(j) -= tau * m_qr.m_Q.col(k);
         }
       }
@@ -655,52 +654,44 @@
   
   const SparseQRType& m_qr;
   const Derived& m_other;
-  bool m_transpose;
+  bool m_transpose; // TODO this actually means adjoint
 };
 
 template<typename SparseQRType>
 struct SparseQRMatrixQReturnType : public EigenBase<SparseQRMatrixQReturnType<SparseQRType> >
 {  
-  typedef typename SparseQRType::Index Index;
   typedef typename SparseQRType::Scalar Scalar;
   typedef Matrix<Scalar,Dynamic,Dynamic> DenseMatrix;
-  SparseQRMatrixQReturnType(const SparseQRType& qr) : m_qr(qr) {}
+  enum {
+    RowsAtCompileTime = Dynamic,
+    ColsAtCompileTime = Dynamic
+  };
+  explicit SparseQRMatrixQReturnType(const SparseQRType& qr) : m_qr(qr) {}
   template<typename Derived>
   SparseQR_QProduct<SparseQRType, Derived> operator*(const MatrixBase<Derived>& other)
   {
     return SparseQR_QProduct<SparseQRType,Derived>(m_qr,other.derived(),false);
   }
+  // To use for operations with the adjoint of Q
   SparseQRMatrixQTransposeReturnType<SparseQRType> adjoint() const
   {
     return SparseQRMatrixQTransposeReturnType<SparseQRType>(m_qr);
   }
   inline Index rows() const { return m_qr.rows(); }
-  inline Index cols() const { return (std::min)(m_qr.rows(),m_qr.cols()); }
-  // To use for operations with the transpose of Q
+  inline Index cols() const { return m_qr.rows(); }
+  // To use for operations with the transpose of Q FIXME this is the same as adjoint at the moment
   SparseQRMatrixQTransposeReturnType<SparseQRType> transpose() const
   {
     return SparseQRMatrixQTransposeReturnType<SparseQRType>(m_qr);
   }
-  template<typename Dest> void evalTo(MatrixBase<Dest>& dest) const
-  {
-    dest.derived() = m_qr.matrixQ() * Dest::Identity(m_qr.rows(), m_qr.rows());
-  }
-  template<typename Dest> void evalTo(SparseMatrixBase<Dest>& dest) const
-  {
-    Dest idMat(m_qr.rows(), m_qr.rows());
-    idMat.setIdentity();
-    // Sort the sparse householder reflectors if needed
-    const_cast<SparseQRType *>(&m_qr)->sort_matrix_Q();
-    dest.derived() = SparseQR_QProduct<SparseQRType, Dest>(m_qr, idMat, false);
-  }
-
   const SparseQRType& m_qr;
 };
 
+// TODO this actually represents the adjoint of Q
 template<typename SparseQRType>
 struct SparseQRMatrixQTransposeReturnType
 {
-  SparseQRMatrixQTransposeReturnType(const SparseQRType& qr) : m_qr(qr) {}
+  explicit SparseQRMatrixQTransposeReturnType(const SparseQRType& qr) : m_qr(qr) {}
   template<typename Derived>
   SparseQR_QProduct<SparseQRType,Derived> operator*(const MatrixBase<Derived>& other)
   {
@@ -709,6 +700,46 @@
   const SparseQRType& m_qr;
 };
 
+namespace internal {
+  
+template<typename SparseQRType>
+struct evaluator_traits<SparseQRMatrixQReturnType<SparseQRType> >
+{
+  typedef typename SparseQRType::MatrixType MatrixType;
+  typedef typename storage_kind_to_evaluator_kind<typename MatrixType::StorageKind>::Kind Kind;
+  typedef SparseShape Shape;
+};
+
+template< typename DstXprType, typename SparseQRType>
+struct Assignment<DstXprType, SparseQRMatrixQReturnType<SparseQRType>, internal::assign_op<typename DstXprType::Scalar,typename DstXprType::Scalar>, Sparse2Sparse>
+{
+  typedef SparseQRMatrixQReturnType<SparseQRType> SrcXprType;
+  typedef typename DstXprType::Scalar Scalar;
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &/*func*/)
+  {
+    typename DstXprType::PlainObject idMat(src.rows(), src.cols());
+    idMat.setIdentity();
+    // Sort the sparse householder reflectors if needed
+    const_cast<SparseQRType *>(&src.m_qr)->_sort_matrix_Q();
+    dst = SparseQR_QProduct<SparseQRType, DstXprType>(src.m_qr, idMat, false);
+  }
+};
+
+template< typename DstXprType, typename SparseQRType>
+struct Assignment<DstXprType, SparseQRMatrixQReturnType<SparseQRType>, internal::assign_op<typename DstXprType::Scalar,typename DstXprType::Scalar>, Sparse2Dense>
+{
+  typedef SparseQRMatrixQReturnType<SparseQRType> SrcXprType;
+  typedef typename DstXprType::Scalar Scalar;
+  typedef typename DstXprType::StorageIndex StorageIndex;
+  static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op<Scalar,Scalar> &/*func*/)
+  {
+    dst = src.m_qr.matrixQ() * DstXprType::Identity(src.m_qr.rows(), src.m_qr.rows());
+  }
+};
+
+} // end namespace internal
+
 } // end namespace Eigen
 
 #endif
diff --git a/Eigen/src/StlSupport/CMakeLists.txt b/Eigen/src/StlSupport/CMakeLists.txt
deleted file mode 100644
index 0f094f6..0000000
--- a/Eigen/src/StlSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_StlSupport_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_StlSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/StlSupport COMPONENT Devel
-  )
diff --git a/Eigen/src/StlSupport/StdDeque.h b/Eigen/src/StlSupport/StdDeque.h
index aaf6633..cf1fedf 100644
--- a/Eigen/src/StlSupport/StdDeque.h
+++ b/Eigen/src/StlSupport/StdDeque.h
@@ -13,32 +13,24 @@
 
 #include "details.h"
 
-// Define the explicit instantiation (e.g. necessary for the Intel compiler)
-#if defined(__INTEL_COMPILER) || defined(__GNUC__)
-  #define EIGEN_EXPLICIT_STL_DEQUE_INSTANTIATION(...) template class std::deque<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> >;
-#else
-  #define EIGEN_EXPLICIT_STL_DEQUE_INSTANTIATION(...)
-#endif
-
 /**
  * This section contains a convenience MACRO which allows an easy specialization of
  * std::deque such that for data types with alignment issues the correct allocator
  * is used automatically.
  */
 #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...) \
-EIGEN_EXPLICIT_STL_DEQUE_INSTANTIATION(__VA_ARGS__) \
 namespace std \
 { \
-  template<typename _Ay> \
-  class deque<__VA_ARGS__, _Ay>  \
+  template<> \
+  class deque<__VA_ARGS__, std::allocator<__VA_ARGS__> >           \
     : public deque<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > \
   { \
     typedef deque<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > deque_base; \
   public: \
     typedef __VA_ARGS__ value_type; \
-    typedef typename deque_base::allocator_type allocator_type; \
-    typedef typename deque_base::size_type size_type;  \
-    typedef typename deque_base::iterator iterator;  \
+    typedef deque_base::allocator_type allocator_type; \
+    typedef deque_base::size_type size_type;  \
+    typedef deque_base::iterator iterator;  \
     explicit deque(const allocator_type& a = allocator_type()) : deque_base(a) {}  \
     template<typename InputIterator> \
     deque(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) : deque_base(first, last, a) {} \
@@ -53,7 +45,7 @@
 }
 
 // check whether we really need the std::deque specialization
-#if !(defined(_GLIBCXX_DEQUE) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::deque::resize(size_type,const T&). */
+#if !EIGEN_HAS_CXX11_CONTAINERS && !(defined(_GLIBCXX_DEQUE) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::deque::resize(size_type,const T&). */
 
 namespace std {
 
diff --git a/Eigen/src/StlSupport/StdList.h b/Eigen/src/StlSupport/StdList.h
index 3c74243..e1eba49 100644
--- a/Eigen/src/StlSupport/StdList.h
+++ b/Eigen/src/StlSupport/StdList.h
@@ -12,32 +12,24 @@
 
 #include "details.h"
 
-// Define the explicit instantiation (e.g. necessary for the Intel compiler)
-#if defined(__INTEL_COMPILER) || defined(__GNUC__)
-  #define EIGEN_EXPLICIT_STL_LIST_INSTANTIATION(...) template class std::list<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> >;
-#else
-  #define EIGEN_EXPLICIT_STL_LIST_INSTANTIATION(...)
-#endif
-
 /**
  * This section contains a convenience MACRO which allows an easy specialization of
  * std::list such that for data types with alignment issues the correct allocator
  * is used automatically.
  */
 #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...) \
-EIGEN_EXPLICIT_STL_LIST_INSTANTIATION(__VA_ARGS__) \
 namespace std \
 { \
-  template<typename _Ay> \
-  class list<__VA_ARGS__, _Ay>  \
+  template<> \
+  class list<__VA_ARGS__, std::allocator<__VA_ARGS__> >           \
     : public list<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > \
   { \
     typedef list<__VA_ARGS__, EIGEN_ALIGNED_ALLOCATOR<__VA_ARGS__> > list_base; \
   public: \
     typedef __VA_ARGS__ value_type; \
-    typedef typename list_base::allocator_type allocator_type; \
-    typedef typename list_base::size_type size_type;  \
-    typedef typename list_base::iterator iterator;  \
+    typedef list_base::allocator_type allocator_type; \
+    typedef list_base::size_type size_type;  \
+    typedef list_base::iterator iterator;  \
     explicit list(const allocator_type& a = allocator_type()) : list_base(a) {}  \
     template<typename InputIterator> \
     list(InputIterator first, InputIterator last, const allocator_type& a = allocator_type()) : list_base(first, last, a) {} \
@@ -51,8 +43,8 @@
   }; \
 }
 
-// check whether we really need the std::vector specialization
-#if !(defined(_GLIBCXX_VECTOR) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::list::resize(size_type,const T&). */
+// check whether we really need the std::list specialization
+#if !EIGEN_HAS_CXX11_CONTAINERS && !(defined(_GLIBCXX_LIST) && (!EIGEN_GNUC_AT_LEAST(4,1))) /* Note that before gcc-4.1 we already have: std::list::resize(size_type,const T&). */
 
 namespace std
 {
diff --git a/Eigen/src/StlSupport/StdVector.h b/Eigen/src/StlSupport/StdVector.h
index 611664a..ec22821 100644
--- a/Eigen/src/StlSupport/StdVector.h
+++ b/Eigen/src/StlSupport/StdVector.h
@@ -44,6 +44,9 @@
   }; \
 }
 
+// Don't specialize if containers are implemented according to C++11
+#if !EIGEN_HAS_CXX11_CONTAINERS
+
 namespace std {
 
 #define EIGEN_STD_VECTOR_SPECIALIZATION_BODY \
@@ -122,5 +125,7 @@
 #endif
   };
 }
+#endif // !EIGEN_HAS_CXX11_CONTAINERS
+
 
 #endif // EIGEN_STDVECTOR_H
diff --git a/Eigen/src/StlSupport/details.h b/Eigen/src/StlSupport/details.h
index d8debc7..2cfd13e 100644
--- a/Eigen/src/StlSupport/details.h
+++ b/Eigen/src/StlSupport/details.h
@@ -22,13 +22,13 @@
   class aligned_allocator_indirection : public EIGEN_ALIGNED_ALLOCATOR<T>
   {
   public:
-    typedef size_t    size_type;
-    typedef ptrdiff_t difference_type;
-    typedef T*        pointer;
-    typedef const T*  const_pointer;
-    typedef T&        reference;
-    typedef const T&  const_reference;
-    typedef T         value_type;
+    typedef std::size_t     size_type;
+    typedef std::ptrdiff_t  difference_type;
+    typedef T*              pointer;
+    typedef const T*        const_pointer;
+    typedef T&              reference;
+    typedef const T&        const_reference;
+    typedef T               value_type;
 
     template<class U>
     struct rebind
@@ -46,7 +46,7 @@
     ~aligned_allocator_indirection() {}
   };
 
-#ifdef _MSC_VER
+#if EIGEN_COMP_MSVC
 
   // sometimes, MSVC detects, at compile time, that the argument x
   // in std::vector::resize(size_t s,T x) won't be aligned and generate an error
diff --git a/Eigen/src/SuperLUSupport/CMakeLists.txt b/Eigen/src/SuperLUSupport/CMakeLists.txt
deleted file mode 100644
index b28ebe5..0000000
--- a/Eigen/src/SuperLUSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_SuperLUSupport_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_SuperLUSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/SuperLUSupport COMPONENT Devel
-  )
diff --git a/Eigen/src/SuperLUSupport/SuperLUSupport.h b/Eigen/src/SuperLUSupport/SuperLUSupport.h
index bcb3557..7261c7d 100644
--- a/Eigen/src/SuperLUSupport/SuperLUSupport.h
+++ b/Eigen/src/SuperLUSupport/SuperLUSupport.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2011 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2015 Gael Guennebaud <gael.guennebaud@inria.fr>
 //
 // This Source Code Form is subject to the terms of the Mozilla
 // Public License v. 2.0. If a copy of the MPL was not distributed
@@ -10,16 +10,16 @@
 #ifndef EIGEN_SUPERLUSUPPORT_H
 #define EIGEN_SUPERLUSUPPORT_H
 
-namespace Eigen { 
+namespace Eigen {
 
+#if defined(SUPERLU_MAJOR_VERSION) && (SUPERLU_MAJOR_VERSION >= 5)
 #define DECL_GSSVX(PREFIX,FLOATTYPE,KEYTYPE)		\
     extern "C" {                                                                                          \
-      typedef struct { FLOATTYPE for_lu; FLOATTYPE total_needed; int expansions; } PREFIX##mem_usage_t;   \
       extern void PREFIX##gssvx(superlu_options_t *, SuperMatrix *, int *, int *, int *,                  \
                                 char *, FLOATTYPE *, FLOATTYPE *, SuperMatrix *, SuperMatrix *,           \
                                 void *, int, SuperMatrix *, SuperMatrix *,                                \
                                 FLOATTYPE *, FLOATTYPE *, FLOATTYPE *, FLOATTYPE *,                       \
-                                PREFIX##mem_usage_t *, SuperLUStat_t *, int *);                           \
+                                GlobalLU_t *, mem_usage_t *, SuperLUStat_t *, int *);                     \
     }                                                                                                     \
     inline float SuperLU_gssvx(superlu_options_t *options, SuperMatrix *A,                                \
          int *perm_c, int *perm_r, int *etree, char *equed,                                               \
@@ -29,12 +29,37 @@
          FLOATTYPE *recip_pivot_growth,                                                                   \
          FLOATTYPE *rcond, FLOATTYPE *ferr, FLOATTYPE *berr,                                              \
          SuperLUStat_t *stats, int *info, KEYTYPE) {                                                      \
-    PREFIX##mem_usage_t mem_usage;                                                                        \
+    mem_usage_t mem_usage;                                                                                \
+    GlobalLU_t gLU;                                                                                       \
+    PREFIX##gssvx(options, A, perm_c, perm_r, etree, equed, R, C, L,                                      \
+         U, work, lwork, B, X, recip_pivot_growth, rcond,                                                 \
+         ferr, berr, &gLU, &mem_usage, stats, info);                                                      \
+    return mem_usage.for_lu; /* bytes used by the factor storage */                                       \
+  }
+#else // version < 5.0
+#define DECL_GSSVX(PREFIX,FLOATTYPE,KEYTYPE)		\
+    extern "C" {                                                                                          \
+      extern void PREFIX##gssvx(superlu_options_t *, SuperMatrix *, int *, int *, int *,                  \
+                                char *, FLOATTYPE *, FLOATTYPE *, SuperMatrix *, SuperMatrix *,           \
+                                void *, int, SuperMatrix *, SuperMatrix *,                                \
+                                FLOATTYPE *, FLOATTYPE *, FLOATTYPE *, FLOATTYPE *,                       \
+                                mem_usage_t *, SuperLUStat_t *, int *);                                   \
+    }                                                                                                     \
+    inline float SuperLU_gssvx(superlu_options_t *options, SuperMatrix *A,                                \
+         int *perm_c, int *perm_r, int *etree, char *equed,                                               \
+         FLOATTYPE *R, FLOATTYPE *C, SuperMatrix *L,                                                      \
+         SuperMatrix *U, void *work, int lwork,                                                           \
+         SuperMatrix *B, SuperMatrix *X,                                                                  \
+         FLOATTYPE *recip_pivot_growth,                                                                   \
+         FLOATTYPE *rcond, FLOATTYPE *ferr, FLOATTYPE *berr,                                              \
+         SuperLUStat_t *stats, int *info, KEYTYPE) {                                                      \
+    mem_usage_t mem_usage;                                                                                \
     PREFIX##gssvx(options, A, perm_c, perm_r, etree, equed, R, C, L,                                      \
          U, work, lwork, B, X, recip_pivot_growth, rcond,                                                 \
          ferr, berr, &mem_usage, stats, info);                                                            \
     return mem_usage.for_lu; /* bytes used by the factor storage */                                       \
   }
+#endif
 
 DECL_GSSVX(s,float,float)
 DECL_GSSVX(c,float,std::complex<float>)
@@ -53,7 +78,7 @@
       extern void PREFIX##gsisx(superlu_options_t *, SuperMatrix *, int *, int *, int *,        \
                          char *, FLOATTYPE *, FLOATTYPE *, SuperMatrix *, SuperMatrix *,        \
                          void *, int, SuperMatrix *, SuperMatrix *, FLOATTYPE *, FLOATTYPE *,   \
-                         PREFIX##mem_usage_t *, SuperLUStat_t *, int *);                        \
+                         mem_usage_t *, SuperLUStat_t *, int *);                        \
     }                                                                                           \
     inline float SuperLU_gsisx(superlu_options_t *options, SuperMatrix *A,                      \
          int *perm_c, int *perm_r, int *etree, char *equed,                                     \
@@ -63,7 +88,7 @@
          FLOATTYPE *recip_pivot_growth,                                                         \
          FLOATTYPE *rcond,                                                                      \
          SuperLUStat_t *stats, int *info, KEYTYPE) {                                            \
-    PREFIX##mem_usage_t mem_usage;                                                              \
+    mem_usage_t mem_usage;                                                              \
     PREFIX##gsisx(options, A, perm_c, perm_r, etree, equed, R, C, L,                            \
          U, work, lwork, B, X, recip_pivot_growth, rcond,                                       \
          &mem_usage, stats, info);                                                              \
@@ -156,37 +181,38 @@
     res.setScalarType<typename MatrixType::Scalar>();
     res.Mtype     = SLU_GE;
 
-    res.nrow      = mat.rows();
-    res.ncol      = mat.cols();
+    res.nrow      = internal::convert_index<int>(mat.rows());
+    res.ncol      = internal::convert_index<int>(mat.cols());
 
-    res.storage.lda       = MatrixType::IsVectorAtCompileTime ? mat.size() : mat.outerStride();
+    res.storage.lda       = internal::convert_index<int>(MatrixType::IsVectorAtCompileTime ? mat.size() : mat.outerStride());
     res.storage.values    = (void*)(mat.data());
     return res;
   }
 
   template<typename MatrixType>
-  static SluMatrix Map(SparseMatrixBase<MatrixType>& mat)
+  static SluMatrix Map(SparseMatrixBase<MatrixType>& a_mat)
   {
+    MatrixType &mat(a_mat.derived());
     SluMatrix res;
     if ((MatrixType::Flags&RowMajorBit)==RowMajorBit)
     {
       res.setStorageType(SLU_NR);
-      res.nrow      = mat.cols();
-      res.ncol      = mat.rows();
+      res.nrow      = internal::convert_index<int>(mat.cols());
+      res.ncol      = internal::convert_index<int>(mat.rows());
     }
     else
     {
       res.setStorageType(SLU_NC);
-      res.nrow      = mat.rows();
-      res.ncol      = mat.cols();
+      res.nrow      = internal::convert_index<int>(mat.rows());
+      res.ncol      = internal::convert_index<int>(mat.cols());
     }
 
     res.Mtype       = SLU_GE;
 
-    res.storage.nnz       = mat.nonZeros();
-    res.storage.values    = mat.derived().valuePtr();
-    res.storage.innerInd  = mat.derived().innerIndexPtr();
-    res.storage.outerInd  = mat.derived().outerIndexPtr();
+    res.storage.nnz       = internal::convert_index<int>(mat.nonZeros());
+    res.storage.values    = mat.valuePtr();
+    res.storage.innerInd  = mat.innerIndexPtr();
+    res.storage.outerInd  = mat.outerIndexPtr();
 
     res.setScalarType<typename MatrixType::Scalar>();
 
@@ -271,8 +297,8 @@
 template<typename Scalar, int Flags, typename Index>
 MappedSparseMatrix<Scalar,Flags,Index> map_superlu(SluMatrix& sluMat)
 {
-  eigen_assert((Flags&RowMajor)==RowMajor && sluMat.Stype == SLU_NR
-         || (Flags&ColMajor)==ColMajor && sluMat.Stype == SLU_NC);
+  eigen_assert(((Flags&RowMajor)==RowMajor && sluMat.Stype == SLU_NR)
+         || ((Flags&ColMajor)==ColMajor && sluMat.Stype == SLU_NC));
 
   Index outerSize = (Flags&RowMajor)==RowMajor ? sluMat.ncol : sluMat.nrow;
 
@@ -288,17 +314,26 @@
   * \brief The base class for the direct and incomplete LU factorization of SuperLU
   */
 template<typename _MatrixType, typename Derived>
-class SuperLUBase : internal::noncopyable
+class SuperLUBase : public SparseSolverBase<Derived>
 {
+  protected:
+    typedef SparseSolverBase<Derived> Base;
+    using Base::derived;
+    using Base::m_isInitialized;
   public:
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar,Dynamic,1> Vector;
     typedef Matrix<int, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
     typedef Matrix<int, MatrixType::RowsAtCompileTime, 1> IntColVectorType;    
+    typedef Map<PermutationMatrix<Dynamic,Dynamic,int> > PermutationMap;
     typedef SparseMatrix<Scalar> LUMatrixType;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
 
@@ -309,9 +344,6 @@
       clearFactors();
     }
     
-    Derived& derived() { return *static_cast<Derived*>(this); }
-    const Derived& derived() const { return *static_cast<const Derived*>(this); }
-    
     inline Index rows() const { return m_matrix.rows(); }
     inline Index cols() const { return m_matrix.cols(); }
     
@@ -335,33 +367,7 @@
       derived().analyzePattern(matrix);
       derived().factorize(matrix);
     }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<SuperLUBase, Rhs> solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "SuperLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SuperLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<SuperLUBase, Rhs>(*this, b.derived());
-    }
-    
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<SuperLUBase, Rhs> solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "SuperLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "SuperLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<SuperLUBase, Rhs>(*this, b.derived());
-    }
-    
+
     /** Performs a symbolic decomposition on the sparcity of \a matrix.
       *
       * This function is particularly useful when solving for several problems having the same structure.
@@ -386,7 +392,7 @@
     {
       set_default_options(&this->m_sluOptions);
       
-      const int size = a.rows();
+      const Index size = a.rows();
       m_matrix = a;
 
       m_sluA = internal::asSluMatrix(m_matrix);
@@ -405,7 +411,7 @@
       m_sluB.storage.values = 0;
       m_sluB.nrow           = 0;
       m_sluB.ncol           = 0;
-      m_sluB.storage.lda    = size;
+      m_sluB.storage.lda    = internal::convert_index<int>(size);
       m_sluX                = m_sluB;
       
       m_extractedDataAreDirty = true;
@@ -453,7 +459,6 @@
     mutable char m_sluEqued;
 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
     mutable bool m_extractedDataAreDirty;
@@ -473,7 +478,11 @@
   *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \warning This class is only for the 4.x versions of SuperLU. The 3.x and 5.x versions are not supported.
+  *
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   */
 template<typename _MatrixType>
 class SuperLU : public SuperLUBase<_MatrixType,SuperLU<_MatrixType> >
@@ -483,18 +492,20 @@
     typedef _MatrixType MatrixType;
     typedef typename Base::Scalar Scalar;
     typedef typename Base::RealScalar RealScalar;
-    typedef typename Base::Index Index;
+    typedef typename Base::StorageIndex StorageIndex;
     typedef typename Base::IntRowVectorType IntRowVectorType;
-    typedef typename Base::IntColVectorType IntColVectorType;    
+    typedef typename Base::IntColVectorType IntColVectorType;   
+    typedef typename Base::PermutationMap PermutationMap;
     typedef typename Base::LUMatrixType LUMatrixType;
     typedef TriangularView<LUMatrixType, Lower|UnitDiag>  LMatrixType;
-    typedef TriangularView<LUMatrixType,  Upper>           UMatrixType;
+    typedef TriangularView<LUMatrixType,  Upper>          UMatrixType;
 
   public:
+    using Base::_solve_impl;
 
     SuperLU() : Base() { init(); }
 
-    SuperLU(const MatrixType& matrix) : Base()
+    explicit SuperLU(const MatrixType& matrix) : Base()
     {
       init();
       Base::compute(matrix);
@@ -525,11 +536,9 @@
       */
     void factorize(const MatrixType& matrix);
     
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
-    #endif // EIGEN_PARSED_BY_DOXYGEN
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
     
     inline const LMatrixType& matrixL() const
     {
@@ -637,12 +646,12 @@
 
 template<typename MatrixType>
 template<typename Rhs,typename Dest>
-void SuperLU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
+void SuperLU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
 {
   eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or analyzePattern()/factorize()");
 
-  const int size = m_matrix.rows();
-  const int rhsCols = b.cols();
+  const Index size = m_matrix.rows();
+  const Index rhsCols = b.cols();
   eigen_assert(size==b.rows());
 
   m_sluOptions.Trans = NOTRANS;
@@ -652,8 +661,12 @@
 
   m_sluFerr.resize(rhsCols);
   m_sluBerr.resize(rhsCols);
-  m_sluB = SluMatrix::Map(b.const_cast_derived());
-  m_sluX = SluMatrix::Map(x.derived());
+  
+  Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b);
+  Ref<const Matrix<typename Dest::Scalar,Dynamic,Dynamic,ColMajor> > x_ref(x);
+  
+  m_sluB = SluMatrix::Map(b_ref.const_cast_derived());
+  m_sluX = SluMatrix::Map(x_ref.const_cast_derived());
   
   typename Rhs::PlainObject b_cpy;
   if(m_sluEqued!='N')
@@ -676,6 +689,10 @@
                 &m_sluFerr[0], &m_sluBerr[0],
                 &m_sluStat, &info, Scalar());
   StatFree(&m_sluStat);
+  
+  if(x.derived().data() != x_ref.data())
+    x = x_ref;
+  
   m_info = info==0 ? Success : NumericalIssue;
 }
 
@@ -699,7 +716,7 @@
     NCformat    *Ustore = static_cast<NCformat*>(m_sluU.Store);
     Scalar      *SNptr;
 
-    const int size = m_matrix.rows();
+    const Index size = m_matrix.rows();
     m_l.resize(size,size);
     m_l.resizeNonZeros(Lstore->nnz);
     m_u.resize(size,size);
@@ -791,6 +808,8 @@
         det *= m_u.valuePtr()[lastId];
     }
   }
+  if(PermutationMap(m_p.data(),m_p.size()).determinant()*PermutationMap(m_q.data(),m_q.size()).determinant()<0)
+    det = -det;
   if(m_sluEqued!='N')
     return det/m_sluRscale.prod()/m_sluCscale.prod();
   else
@@ -810,11 +829,13 @@
   * This class allows to solve for an approximate solution of A.X = B sparse linear problems via an incomplete LU factorization
   * using the SuperLU library. This class is aimed to be used as a preconditioner of the iterative linear solvers.
   *
-  * \warning This class requires SuperLU 4 or later.
+  * \warning This class is only for the 4.x versions of SuperLU. The 3.x and 5.x versions are not supported.
   *
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers, class ConjugateGradient, class BiCGSTAB
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class IncompleteLUT, class ConjugateGradient, class BiCGSTAB
   */
 
 template<typename _MatrixType>
@@ -825,9 +846,9 @@
     typedef _MatrixType MatrixType;
     typedef typename Base::Scalar Scalar;
     typedef typename Base::RealScalar RealScalar;
-    typedef typename Base::Index Index;
 
   public:
+    using Base::_solve_impl;
 
     SuperILU() : Base() { init(); }
 
@@ -863,7 +884,7 @@
     #ifndef EIGEN_PARSED_BY_DOXYGEN
     /** \internal */
     template<typename Rhs,typename Dest>
-    void _solve(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
+    void _solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest> &dest) const;
     #endif // EIGEN_PARSED_BY_DOXYGEN
     
   protected:
@@ -946,9 +967,10 @@
   m_factorizationIsOk = true;
 }
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
 template<typename MatrixType>
 template<typename Rhs,typename Dest>
-void SuperILU<MatrixType>::_solve(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
+void SuperILU<MatrixType>::_solve_impl(const MatrixBase<Rhs> &b, MatrixBase<Dest>& x) const
 {
   eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or analyzePattern()/factorize()");
 
@@ -962,8 +984,12 @@
 
   m_sluFerr.resize(rhsCols);
   m_sluBerr.resize(rhsCols);
-  m_sluB = SluMatrix::Map(b.const_cast_derived());
-  m_sluX = SluMatrix::Map(x.derived());
+  
+  Ref<const Matrix<typename Rhs::Scalar,Dynamic,Dynamic,ColMajor> > b_ref(b);
+  Ref<const Matrix<typename Dest::Scalar,Dynamic,Dynamic,ColMajor> > x_ref(x);
+  
+  m_sluB = SluMatrix::Map(b_ref.const_cast_derived());
+  m_sluX = SluMatrix::Map(x_ref.const_cast_derived());
 
   typename Rhs::PlainObject b_cpy;
   if(m_sluEqued!='N')
@@ -986,40 +1012,15 @@
                 &recip_pivot_growth, &rcond,
                 &m_sluStat, &info, Scalar());
   StatFree(&m_sluStat);
+  
+  if(x.derived().data() != x_ref.data())
+    x = x_ref;
 
   m_info = info==0 ? Success : NumericalIssue;
 }
 #endif
 
-namespace internal {
-  
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct solve_retval<SuperLUBase<_MatrixType,Derived>, Rhs>
-  : solve_retval_base<SuperLUBase<_MatrixType,Derived>, Rhs>
-{
-  typedef SuperLUBase<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec().derived()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Derived, typename Rhs>
-struct sparse_solve_retval<SuperLUBase<_MatrixType,Derived>, Rhs>
-  : sparse_solve_retval_base<SuperLUBase<_MatrixType,Derived>, Rhs>
-{
-  typedef SuperLUBase<_MatrixType,Derived> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
+#endif
 
 } // end namespace Eigen
 
diff --git a/Eigen/src/UmfPackSupport/CMakeLists.txt b/Eigen/src/UmfPackSupport/CMakeLists.txt
deleted file mode 100644
index a57de00..0000000
--- a/Eigen/src/UmfPackSupport/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_UmfPackSupport_SRCS "*.h")
-
-INSTALL(FILES 
-  ${Eigen_UmfPackSupport_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/UmfPackSupport COMPONENT Devel
-  )
diff --git a/Eigen/src/UmfPackSupport/UmfPackSupport.h b/Eigen/src/UmfPackSupport/UmfPackSupport.h
index 29c60c3..91c09ab 100644
--- a/Eigen/src/UmfPackSupport/UmfPackSupport.h
+++ b/Eigen/src/UmfPackSupport/UmfPackSupport.h
@@ -10,12 +10,37 @@
 #ifndef EIGEN_UMFPACKSUPPORT_H
 #define EIGEN_UMFPACKSUPPORT_H
 
-namespace Eigen { 
+namespace Eigen {
 
 /* TODO extract L, extract U, compute det, etc... */
 
 // generic double/complex<double> wrapper functions:
 
+
+inline void umfpack_defaults(double control[UMFPACK_CONTROL], double)
+{ umfpack_di_defaults(control); }
+
+inline void umfpack_defaults(double control[UMFPACK_CONTROL], std::complex<double>)
+{ umfpack_zi_defaults(control); }
+
+inline void umfpack_report_info(double control[UMFPACK_CONTROL], double info[UMFPACK_INFO], double)
+{ umfpack_di_report_info(control, info);}
+
+inline void umfpack_report_info(double control[UMFPACK_CONTROL], double info[UMFPACK_INFO], std::complex<double>)
+{ umfpack_zi_report_info(control, info);}
+
+inline void umfpack_report_status(double control[UMFPACK_CONTROL], int status, double)
+{ umfpack_di_report_status(control, status);}
+
+inline void umfpack_report_status(double control[UMFPACK_CONTROL], int status, std::complex<double>)
+{ umfpack_zi_report_status(control, status);}
+
+inline void umfpack_report_control(double control[UMFPACK_CONTROL], double)
+{ umfpack_di_report_control(control);}
+
+inline void umfpack_report_control(double control[UMFPACK_CONTROL], std::complex<double>)
+{ umfpack_zi_report_control(control);}
+
 inline void umfpack_free_numeric(void **Numeric, double)
 { umfpack_di_free_numeric(Numeric); *Numeric = 0; }
 
@@ -107,15 +132,6 @@
   return umfpack_zi_get_determinant(&mx_real,0,Ex,NumericHandle,User_Info);
 }
 
-namespace internal {
-  template<typename T> struct umfpack_helper_is_sparse_plain : false_type {};
-  template<typename Scalar, int Options, typename StorageIndex>
-  struct umfpack_helper_is_sparse_plain<SparseMatrix<Scalar,Options,StorageIndex> >
-    : true_type {};
-  template<typename Scalar, int Options, typename StorageIndex>
-  struct umfpack_helper_is_sparse_plain<MappedSparseMatrix<Scalar,Options,StorageIndex> >
-    : true_type {};
-}
 
 /** \ingroup UmfPackSupport_Module
   * \brief A sparse LU factorization and solver based on UmfPack
@@ -128,27 +144,47 @@
   * Otherwise an expensive copy will be made. You can call the inexpensive makeCompressed() to get a compressed matrix.
   * \tparam _MatrixType the type of the sparse matrix A, it must be a SparseMatrix<>
   *
-  * \sa \ref TutorialSparseDirectSolvers
+  * \implsparsesolverconcept
+  *
+  * \sa \ref TutorialSparseSolverConcept, class SparseLU
   */
 template<typename _MatrixType>
-class UmfPackLU : internal::noncopyable
+class UmfPackLU : public SparseSolverBase<UmfPackLU<_MatrixType> >
 {
+  protected:
+    typedef SparseSolverBase<UmfPackLU<_MatrixType> > Base;
+    using Base::m_isInitialized;
   public:
+    using Base::_solve_impl;
     typedef _MatrixType MatrixType;
     typedef typename MatrixType::Scalar Scalar;
     typedef typename MatrixType::RealScalar RealScalar;
-    typedef typename MatrixType::Index Index;
+    typedef typename MatrixType::StorageIndex StorageIndex;
     typedef Matrix<Scalar,Dynamic,1> Vector;
     typedef Matrix<int, 1, MatrixType::ColsAtCompileTime> IntRowVectorType;
     typedef Matrix<int, MatrixType::RowsAtCompileTime, 1> IntColVectorType;
     typedef SparseMatrix<Scalar> LUMatrixType;
     typedef SparseMatrix<Scalar,ColMajor,int> UmfpackMatrixType;
+    typedef Ref<const UmfpackMatrixType, StandardCompressedFormat> UmfpackMatrixRef;
+    enum {
+      ColsAtCompileTime = MatrixType::ColsAtCompileTime,
+      MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime
+    };
 
   public:
 
-    UmfPackLU() { init(); }
+    typedef Array<double, UMFPACK_CONTROL, 1> UmfpackControl;
+    typedef Array<double, UMFPACK_INFO, 1> UmfpackInfo;
 
-    UmfPackLU(const MatrixType& matrix)
+    UmfPackLU()
+      : m_dummy(0,0), mp_matrix(m_dummy)
+    {
+      init();
+    }
+
+    template<typename InputMatrixType>
+    explicit UmfPackLU(const InputMatrixType& matrix)
+      : mp_matrix(matrix)
     {
       init();
       compute(matrix);
@@ -160,8 +196,8 @@
       if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
     }
 
-    inline Index rows() const { return m_copyMatrix.rows(); }
-    inline Index cols() const { return m_copyMatrix.cols(); }
+    inline Index rows() const { return mp_matrix.rows(); }
+    inline Index cols() const { return mp_matrix.cols(); }
 
     /** \brief Reports whether previous computation was successful.
       *
@@ -198,7 +234,7 @@
       return m_q;
     }
 
-    /** Computes the sparse Cholesky decomposition of \a matrix 
+    /** Computes the sparse Cholesky decomposition of \a matrix
      *  Note that the matrix should be column-major, and in compressed format for best performance.
      *  \sa SparseMatrix::makeCompressed().
      */
@@ -207,37 +243,11 @@
     {
       if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
       if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
-      grapInput(matrix.derived());
+      grab(matrix.derived());
       analyzePattern_impl();
       factorize_impl();
     }
 
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::solve_retval<UmfPackLU, Rhs> solve(const MatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "UmfPackLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "UmfPackLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::solve_retval<UmfPackLU, Rhs>(*this, b.derived());
-    }
-
-    /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A.
-      *
-      * \sa compute()
-      */
-    template<typename Rhs>
-    inline const internal::sparse_solve_retval<UmfPackLU, Rhs> solve(const SparseMatrixBase<Rhs>& b) const
-    {
-      eigen_assert(m_isInitialized && "UmfPackLU is not initialized.");
-      eigen_assert(rows()==b.rows()
-                && "UmfPackLU::solve(): invalid number of rows of the right hand side matrix b");
-      return internal::sparse_solve_retval<UmfPackLU, Rhs>(*this, b.derived());
-    }
-
     /** Performs a symbolic decomposition on the sparcity of \a matrix.
       *
       * This function is particularly useful when solving for several problems having the same structure.
@@ -249,12 +259,45 @@
     {
       if(m_symbolic) umfpack_free_symbolic(&m_symbolic,Scalar());
       if(m_numeric)  umfpack_free_numeric(&m_numeric,Scalar());
-      
-      grapInput(matrix.derived());
+
+      grab(matrix.derived());
 
       analyzePattern_impl();
     }
 
+    /** Provides the return status code returned by UmfPack during the numeric
+      * factorization.
+      *
+      * \sa factorize(), compute()
+      */
+    inline int umfpackFactorizeReturncode() const
+    {
+      eigen_assert(m_numeric && "UmfPackLU: you must first call factorize()");
+      return m_fact_errorCode;
+    }
+
+    /** Provides access to the control settings array used by UmfPack.
+      *
+      * If this array contains NaN's, the default values are used.
+      *
+      * See UMFPACK documentation for details.
+      */
+    inline const UmfpackControl& umfpackControl() const
+    {
+      return m_control;
+    }
+
+    /** Provides access to the control settings array used by UmfPack.
+      *
+      * If this array contains NaN's, the default values are used.
+      *
+      * See UMFPACK documentation for details.
+      */
+    inline UmfpackControl& umfpackControl()
+    {
+      return m_control;
+    }
+
     /** Performs a numeric decomposition of \a matrix
       *
       * The given matrix must has the same sparcity than the matrix on which the pattern anylysis has been performed.
@@ -268,16 +311,42 @@
       if(m_numeric)
         umfpack_free_numeric(&m_numeric,Scalar());
 
-      grapInput(matrix.derived());
-      
+      grab(matrix.derived());
+
       factorize_impl();
     }
 
-    #ifndef EIGEN_PARSED_BY_DOXYGEN
+    /** Prints the current UmfPack control settings.
+      *
+      * \sa umfpackControl()
+      */
+    void umfpackReportControl()
+    {
+      umfpack_report_control(m_control.data(), Scalar());
+    }
+
+    /** Prints statistics collected by UmfPack.
+      *
+      * \sa analyzePattern(), compute()
+      */
+    void umfpackReportInfo()
+    {
+      eigen_assert(m_analysisIsOk && "UmfPackLU: you must first call analyzePattern()");
+      umfpack_report_info(m_control.data(), m_umfpackInfo.data(), Scalar());
+    }
+
+    /** Prints the status of the previous factorization operation performed by UmfPack (symbolic or numerical factorization).
+      *
+      * \sa analyzePattern(), compute()
+      */
+    void umfpackReportStatus() {
+      eigen_assert(m_analysisIsOk && "UmfPackLU: you must first call analyzePattern()");
+      umfpack_report_status(m_control.data(), m_fact_errorCode, Scalar());
+    }
+
     /** \internal */
     template<typename BDerived,typename XDerived>
-    bool _solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const;
-    #endif
+    bool _solve_impl(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const;
 
     Scalar determinant() const;
 
@@ -291,92 +360,75 @@
       m_isInitialized         = false;
       m_numeric               = 0;
       m_symbolic              = 0;
-      m_outerIndexPtr         = 0;
-      m_innerIndexPtr         = 0;
-      m_valuePtr              = 0;
       m_extractedDataAreDirty = true;
+
+      umfpack_defaults(m_control.data(), Scalar());
     }
-    
-    template<typename InputMatrixType>
-    void grapInput_impl(const InputMatrixType& mat, internal::true_type)
-    {
-      m_copyMatrix.resize(mat.rows(), mat.cols());
-      if( ((MatrixType::Flags&RowMajorBit)==RowMajorBit) || sizeof(typename MatrixType::Index)!=sizeof(int) || !mat.isCompressed() )
-      {
-        // non supported input -> copy
-        m_copyMatrix = mat;
-        m_outerIndexPtr = m_copyMatrix.outerIndexPtr();
-        m_innerIndexPtr = m_copyMatrix.innerIndexPtr();
-        m_valuePtr      = m_copyMatrix.valuePtr();
-      }
-      else
-      {
-        m_outerIndexPtr = mat.outerIndexPtr();
-        m_innerIndexPtr = mat.innerIndexPtr();
-        m_valuePtr      = mat.valuePtr();
-      }
-    }
-    
-    template<typename InputMatrixType>
-    void grapInput_impl(const InputMatrixType& mat, internal::false_type)
-    {
-      m_copyMatrix = mat;
-      m_outerIndexPtr = m_copyMatrix.outerIndexPtr();
-      m_innerIndexPtr = m_copyMatrix.innerIndexPtr();
-      m_valuePtr      = m_copyMatrix.valuePtr();
-    }
-    
-    template<typename InputMatrixType>
-    void grapInput(const InputMatrixType& mat)
-    {
-      grapInput_impl(mat, internal::umfpack_helper_is_sparse_plain<InputMatrixType>());
-    }
-    
+
     void analyzePattern_impl()
     {
-      int errorCode = 0;
-      errorCode = umfpack_symbolic(m_copyMatrix.rows(), m_copyMatrix.cols(), m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
-                                   &m_symbolic, 0, 0);
+      m_fact_errorCode = umfpack_symbolic(internal::convert_index<int>(mp_matrix.rows()),
+                                          internal::convert_index<int>(mp_matrix.cols()),
+                                          mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
+                                          &m_symbolic, m_control.data(), m_umfpackInfo.data());
 
       m_isInitialized = true;
-      m_info = errorCode ? InvalidInput : Success;
+      m_info = m_fact_errorCode ? InvalidInput : Success;
       m_analysisIsOk = true;
       m_factorizationIsOk = false;
       m_extractedDataAreDirty = true;
     }
-    
+
     void factorize_impl()
     {
-      int errorCode;
-      errorCode = umfpack_numeric(m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
-                                  m_symbolic, &m_numeric, 0, 0);
 
-      m_info = errorCode ? NumericalIssue : Success;
+      m_fact_errorCode = umfpack_numeric(mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
+                                         m_symbolic, &m_numeric, m_control.data(), m_umfpackInfo.data());
+
+      m_info = m_fact_errorCode == UMFPACK_OK ? Success : NumericalIssue;
       m_factorizationIsOk = true;
       m_extractedDataAreDirty = true;
     }
 
+    template<typename MatrixDerived>
+    void grab(const EigenBase<MatrixDerived> &A)
+    {
+      mp_matrix.~UmfpackMatrixRef();
+      ::new (&mp_matrix) UmfpackMatrixRef(A.derived());
+    }
+
+    void grab(const UmfpackMatrixRef &A)
+    {
+      if(&(A.derived()) != &mp_matrix)
+      {
+        mp_matrix.~UmfpackMatrixRef();
+        ::new (&mp_matrix) UmfpackMatrixRef(A);
+      }
+    }
+
     // cached data to reduce reallocation, etc.
     mutable LUMatrixType m_l;
+    int m_fact_errorCode;
+    UmfpackControl m_control;
+    mutable UmfpackInfo m_umfpackInfo;
+
     mutable LUMatrixType m_u;
     mutable IntColVectorType m_p;
     mutable IntRowVectorType m_q;
 
-    UmfpackMatrixType m_copyMatrix;
-    const Scalar* m_valuePtr;
-    const int* m_outerIndexPtr;
-    const int* m_innerIndexPtr;
+    UmfpackMatrixType m_dummy;
+    UmfpackMatrixRef mp_matrix;
+
     void* m_numeric;
     void* m_symbolic;
 
     mutable ComputationInfo m_info;
-    bool m_isInitialized;
     int m_factorizationIsOk;
     int m_analysisIsOk;
     mutable bool m_extractedDataAreDirty;
-    
+
   private:
-    UmfPackLU(UmfPackLU& ) { }
+    UmfPackLU(const UmfPackLU& ) { }
 };
 
 
@@ -418,19 +470,30 @@
 
 template<typename MatrixType>
 template<typename BDerived,typename XDerived>
-bool UmfPackLU<MatrixType>::_solve(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const
+bool UmfPackLU<MatrixType>::_solve_impl(const MatrixBase<BDerived> &b, MatrixBase<XDerived> &x) const
 {
-  const int rhsCols = b.cols();
+  Index rhsCols = b.cols();
   eigen_assert((BDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major rhs yet");
   eigen_assert((XDerived::Flags&RowMajorBit)==0 && "UmfPackLU backend does not support non col-major result yet");
   eigen_assert(b.derived().data() != x.derived().data() && " Umfpack does not support inplace solve");
-  
+
   int errorCode;
+  Scalar* x_ptr = 0;
+  Matrix<Scalar,Dynamic,1> x_tmp;
+  if(x.innerStride()!=1)
+  {
+    x_tmp.resize(x.rows());
+    x_ptr = x_tmp.data();
+  }
   for (int j=0; j<rhsCols; ++j)
   {
+    if(x.innerStride()==1)
+      x_ptr = &x.col(j).coeffRef(0);
     errorCode = umfpack_solve(UMFPACK_A,
-        m_outerIndexPtr, m_innerIndexPtr, m_valuePtr,
-        &x.col(j).coeffRef(0), &b.const_cast_derived().col(j).coeffRef(0), m_numeric, 0, 0);
+        mp_matrix.outerIndexPtr(), mp_matrix.innerIndexPtr(), mp_matrix.valuePtr(),
+        x_ptr, &b.const_cast_derived().col(j).coeffRef(0), m_numeric, m_control.data(), m_umfpackInfo.data());
+    if(x.innerStride()!=1)
+      x.col(j) = x_tmp;
     if (errorCode!=0)
       return false;
   }
@@ -438,37 +501,6 @@
   return true;
 }
 
-
-namespace internal {
-
-template<typename _MatrixType, typename Rhs>
-struct solve_retval<UmfPackLU<_MatrixType>, Rhs>
-  : solve_retval_base<UmfPackLU<_MatrixType>, Rhs>
-{
-  typedef UmfPackLU<_MatrixType> Dec;
-  EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    dec()._solve(rhs(),dst);
-  }
-};
-
-template<typename _MatrixType, typename Rhs>
-struct sparse_solve_retval<UmfPackLU<_MatrixType>, Rhs>
-  : sparse_solve_retval_base<UmfPackLU<_MatrixType>, Rhs>
-{
-  typedef UmfPackLU<_MatrixType> Dec;
-  EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs)
-
-  template<typename Dest> void evalTo(Dest& dst) const
-  {
-    this->defaultEvalTo(dst);
-  }
-};
-
-} // end namespace internal
-
 } // end namespace Eigen
 
 #endif // EIGEN_UMFPACKSUPPORT_H
diff --git a/Eigen/src/misc/CMakeLists.txt b/Eigen/src/misc/CMakeLists.txt
deleted file mode 100644
index a58ffb7..0000000
--- a/Eigen/src/misc/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_misc_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_misc_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/misc COMPONENT Devel
-  )
diff --git a/Eigen/src/misc/Image.h b/Eigen/src/misc/Image.h
index 75c5f43..b8b8a04 100644
--- a/Eigen/src/misc/Image.h
+++ b/Eigen/src/misc/Image.h
@@ -38,7 +38,6 @@
   typedef _DecompositionType DecompositionType;
   typedef typename DecompositionType::MatrixType MatrixType;
   typedef ReturnByValue<image_retval_base> Base;
-  typedef typename Base::Index Index;
 
   image_retval_base(const DecompositionType& dec, const MatrixType& originalMatrix)
     : m_dec(dec), m_rank(dec.rank()),
@@ -69,7 +68,6 @@
   typedef typename DecompositionType::MatrixType MatrixType; \
   typedef typename MatrixType::Scalar Scalar; \
   typedef typename MatrixType::RealScalar RealScalar; \
-  typedef typename MatrixType::Index Index; \
   typedef Eigen::internal::image_retval_base<DecompositionType> Base; \
   using Base::dec; \
   using Base::originalMatrix; \
diff --git a/Eigen/src/misc/Kernel.h b/Eigen/src/misc/Kernel.h
index b9e1518..bef5d6f 100644
--- a/Eigen/src/misc/Kernel.h
+++ b/Eigen/src/misc/Kernel.h
@@ -39,9 +39,8 @@
 {
   typedef _DecompositionType DecompositionType;
   typedef ReturnByValue<kernel_retval_base> Base;
-  typedef typename Base::Index Index;
 
-  kernel_retval_base(const DecompositionType& dec)
+  explicit kernel_retval_base(const DecompositionType& dec)
     : m_dec(dec),
       m_rank(dec.rank()),
       m_cols(m_rank==dec.cols() ? 1 : dec.cols() - m_rank)
@@ -68,7 +67,6 @@
   typedef typename DecompositionType::MatrixType MatrixType; \
   typedef typename MatrixType::Scalar Scalar; \
   typedef typename MatrixType::RealScalar RealScalar; \
-  typedef typename MatrixType::Index Index; \
   typedef Eigen::internal::kernel_retval_base<DecompositionType> Base; \
   using Base::dec; \
   using Base::rank; \
diff --git a/Eigen/src/misc/RealSvd2x2.h b/Eigen/src/misc/RealSvd2x2.h
new file mode 100644
index 0000000..abb4d3c
--- /dev/null
+++ b/Eigen/src/misc/RealSvd2x2.h
@@ -0,0 +1,55 @@
+// This file is part of Eigen, a lightweight C++ template library
+// for linear algebra.
+//
+// Copyright (C) 2009-2010 Benoit Jacob <jacob.benoit.1@gmail.com>
+// Copyright (C) 2013-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
+//
+// This Source Code Form is subject to the terms of the Mozilla
+// Public License v. 2.0. If a copy of the MPL was not distributed
+// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
+
+#ifndef EIGEN_REALSVD2X2_H
+#define EIGEN_REALSVD2X2_H
+
+namespace Eigen {
+
+namespace internal {
+
+template<typename MatrixType, typename RealScalar, typename Index>
+void real_2x2_jacobi_svd(const MatrixType& matrix, Index p, Index q,
+                         JacobiRotation<RealScalar> *j_left,
+                         JacobiRotation<RealScalar> *j_right)
+{
+  using std::sqrt;
+  using std::abs;
+  Matrix<RealScalar,2,2> m;
+  m << numext::real(matrix.coeff(p,p)), numext::real(matrix.coeff(p,q)),
+       numext::real(matrix.coeff(q,p)), numext::real(matrix.coeff(q,q));
+  JacobiRotation<RealScalar> rot1;
+  RealScalar t = m.coeff(0,0) + m.coeff(1,1);
+  RealScalar d = m.coeff(1,0) - m.coeff(0,1);
+
+  if(abs(d) < (std::numeric_limits<RealScalar>::min)())
+  {
+    rot1.s() = RealScalar(0);
+    rot1.c() = RealScalar(1);
+  }
+  else
+  {
+    // If d!=0, then t/d cannot overflow because the magnitude of the
+    // entries forming d are not too small compared to the ones forming t.
+    RealScalar u = t / d;
+    RealScalar tmp = sqrt(RealScalar(1) + numext::abs2(u));
+    rot1.s() = RealScalar(1) / tmp;
+    rot1.c() = u / tmp;
+  }
+  m.applyOnTheLeft(0,1,rot1);
+  j_right->makeJacobi(m,0,1);
+  *j_left = rot1 * j_right->transpose();
+}
+
+} // end namespace internal
+
+} // end namespace Eigen
+
+#endif // EIGEN_REALSVD2X2_H
diff --git a/Eigen/src/misc/Solve.h b/Eigen/src/misc/Solve.h
deleted file mode 100644
index 7f70d60..0000000
--- a/Eigen/src/misc/Solve.h
+++ /dev/null
@@ -1,76 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2009 Benoit Jacob <jacob.benoit.1@gmail.com>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_MISC_SOLVE_H
-#define EIGEN_MISC_SOLVE_H
-
-namespace Eigen { 
-
-namespace internal {
-
-/** \class solve_retval_base
-  *
-  */
-template<typename DecompositionType, typename Rhs>
-struct traits<solve_retval_base<DecompositionType, Rhs> >
-{
-  typedef typename DecompositionType::MatrixType MatrixType;
-  typedef Matrix<typename Rhs::Scalar,
-                 MatrixType::ColsAtCompileTime,
-                 Rhs::ColsAtCompileTime,
-                 Rhs::PlainObject::Options,
-                 MatrixType::MaxColsAtCompileTime,
-                 Rhs::MaxColsAtCompileTime> ReturnType;
-};
-
-template<typename _DecompositionType, typename Rhs> struct solve_retval_base
- : public ReturnByValue<solve_retval_base<_DecompositionType, Rhs> >
-{
-  typedef typename remove_all<typename Rhs::Nested>::type RhsNestedCleaned;
-  typedef _DecompositionType DecompositionType;
-  typedef ReturnByValue<solve_retval_base> Base;
-  typedef typename Base::Index Index;
-
-  solve_retval_base(const DecompositionType& dec, const Rhs& rhs)
-    : m_dec(dec), m_rhs(rhs)
-  {}
-
-  inline Index rows() const { return m_dec.cols(); }
-  inline Index cols() const { return m_rhs.cols(); }
-  inline const DecompositionType& dec() const { return m_dec; }
-  inline const RhsNestedCleaned& rhs() const { return m_rhs; }
-
-  template<typename Dest> inline void evalTo(Dest& dst) const
-  {
-    static_cast<const solve_retval<DecompositionType,Rhs>*>(this)->evalTo(dst);
-  }
-
-  protected:
-    const DecompositionType& m_dec;
-    typename Rhs::Nested m_rhs;
-};
-
-} // end namespace internal
-
-#define EIGEN_MAKE_SOLVE_HELPERS(DecompositionType,Rhs) \
-  typedef typename DecompositionType::MatrixType MatrixType; \
-  typedef typename MatrixType::Scalar Scalar; \
-  typedef typename MatrixType::RealScalar RealScalar; \
-  typedef typename MatrixType::Index Index; \
-  typedef Eigen::internal::solve_retval_base<DecompositionType,Rhs> Base; \
-  using Base::dec; \
-  using Base::rhs; \
-  using Base::rows; \
-  using Base::cols; \
-  solve_retval(const DecompositionType& dec, const Rhs& rhs) \
-    : Base(dec, rhs) {}
-
-} // end namespace Eigen
-
-#endif // EIGEN_MISC_SOLVE_H
diff --git a/Eigen/src/misc/SparseSolve.h b/Eigen/src/misc/SparseSolve.h
deleted file mode 100644
index 244bb8e..0000000
--- a/Eigen/src/misc/SparseSolve.h
+++ /dev/null
@@ -1,128 +0,0 @@
-// This file is part of Eigen, a lightweight C++ template library
-// for linear algebra.
-//
-// Copyright (C) 2010 Gael Guennebaud <gael.guennebaud@inria.fr>
-//
-// This Source Code Form is subject to the terms of the Mozilla
-// Public License v. 2.0. If a copy of the MPL was not distributed
-// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
-
-#ifndef EIGEN_SPARSE_SOLVE_H
-#define EIGEN_SPARSE_SOLVE_H
-
-namespace Eigen { 
-
-namespace internal {
-
-template<typename _DecompositionType, typename Rhs> struct sparse_solve_retval_base;
-template<typename _DecompositionType, typename Rhs> struct sparse_solve_retval;
-  
-template<typename DecompositionType, typename Rhs>
-struct traits<sparse_solve_retval_base<DecompositionType, Rhs> >
-{
-  typedef typename DecompositionType::MatrixType MatrixType;
-  typedef SparseMatrix<typename Rhs::Scalar, Rhs::Options, typename Rhs::Index> ReturnType;
-};
-
-template<typename _DecompositionType, typename Rhs> struct sparse_solve_retval_base
- : public ReturnByValue<sparse_solve_retval_base<_DecompositionType, Rhs> >
-{
-  typedef typename remove_all<typename Rhs::Nested>::type RhsNestedCleaned;
-  typedef _DecompositionType DecompositionType;
-  typedef ReturnByValue<sparse_solve_retval_base> Base;
-  typedef typename Base::Index Index;
-
-  sparse_solve_retval_base(const DecompositionType& dec, const Rhs& rhs)
-    : m_dec(dec), m_rhs(rhs)
-  {}
-
-  inline Index rows() const { return m_dec.cols(); }
-  inline Index cols() const { return m_rhs.cols(); }
-  inline const DecompositionType& dec() const { return m_dec; }
-  inline const RhsNestedCleaned& rhs() const { return m_rhs; }
-
-  template<typename Dest> inline void evalTo(Dest& dst) const
-  {
-    static_cast<const sparse_solve_retval<DecompositionType,Rhs>*>(this)->evalTo(dst);
-  }
-
-  protected:
-    template<typename DestScalar, int DestOptions, typename DestIndex>
-    inline void defaultEvalTo(SparseMatrix<DestScalar,DestOptions,DestIndex>& dst) const
-    {
-      // we process the sparse rhs per block of NbColsAtOnce columns temporarily stored into a dense matrix.
-      static const int NbColsAtOnce = 4;
-      int rhsCols = m_rhs.cols();
-      int size = m_rhs.rows();
-      Eigen::Matrix<DestScalar,Dynamic,Dynamic> tmp(size,rhsCols);
-      Eigen::Matrix<DestScalar,Dynamic,Dynamic> tmpX(size,rhsCols);
-      for(int k=0; k<rhsCols; k+=NbColsAtOnce)
-      {
-        int actualCols = std::min<int>(rhsCols-k, NbColsAtOnce);
-        tmp.leftCols(actualCols) = m_rhs.middleCols(k,actualCols);
-        tmpX.leftCols(actualCols) = m_dec.solve(tmp.leftCols(actualCols));
-        dst.middleCols(k,actualCols) = tmpX.leftCols(actualCols).sparseView();
-      }
-    }
-    const DecompositionType& m_dec;
-    typename Rhs::Nested m_rhs;
-};
-
-#define EIGEN_MAKE_SPARSE_SOLVE_HELPERS(DecompositionType,Rhs) \
-  typedef typename DecompositionType::MatrixType MatrixType; \
-  typedef typename MatrixType::Scalar Scalar; \
-  typedef typename MatrixType::RealScalar RealScalar; \
-  typedef typename MatrixType::Index Index; \
-  typedef Eigen::internal::sparse_solve_retval_base<DecompositionType,Rhs> Base; \
-  using Base::dec; \
-  using Base::rhs; \
-  using Base::rows; \
-  using Base::cols; \
-  sparse_solve_retval(const DecompositionType& dec, const Rhs& rhs) \
-    : Base(dec, rhs) {}
-
-
-
-template<typename DecompositionType, typename Rhs, typename Guess> struct solve_retval_with_guess;
-
-template<typename DecompositionType, typename Rhs, typename Guess>
-struct traits<solve_retval_with_guess<DecompositionType, Rhs, Guess> >
-{
-  typedef typename DecompositionType::MatrixType MatrixType;
-  typedef Matrix<typename Rhs::Scalar,
-                 MatrixType::ColsAtCompileTime,
-                 Rhs::ColsAtCompileTime,
-                 Rhs::PlainObject::Options,
-                 MatrixType::MaxColsAtCompileTime,
-                 Rhs::MaxColsAtCompileTime> ReturnType;
-};
-
-template<typename DecompositionType, typename Rhs, typename Guess> struct solve_retval_with_guess
- : public ReturnByValue<solve_retval_with_guess<DecompositionType, Rhs, Guess> >
-{
-  typedef typename DecompositionType::Index Index;
-
-  solve_retval_with_guess(const DecompositionType& dec, const Rhs& rhs, const Guess& guess)
-    : m_dec(dec), m_rhs(rhs), m_guess(guess)
-  {}
-
-  inline Index rows() const { return m_dec.cols(); }
-  inline Index cols() const { return m_rhs.cols(); }
-
-  template<typename Dest> inline void evalTo(Dest& dst) const
-  {
-    dst = m_guess;
-    m_dec._solveWithGuess(m_rhs,dst);
-  }
-
-  protected:
-    const DecompositionType& m_dec;
-    const typename Rhs::Nested m_rhs;
-    const typename Guess::Nested m_guess;
-};
-
-} // namepsace internal
-
-} // end namespace Eigen
-
-#endif // EIGEN_SPARSE_SOLVE_H
diff --git a/Eigen/src/misc/blas.h b/Eigen/src/misc/blas.h
index 6fce99e..25215b1 100644
--- a/Eigen/src/misc/blas.h
+++ b/Eigen/src/misc/blas.h
@@ -30,15 +30,15 @@
 int  BLASFUNC(zdotuw)  (int *, double  *, int *, double  *, int *, double*);
 int  BLASFUNC(zdotcw)  (int *, double  *, int *, double  *, int *, double*);
 
-int    BLASFUNC(saxpy) (int *, float  *, float  *, int *, float  *, int *);
-int    BLASFUNC(daxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(qaxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(caxpy) (int *, float  *, float  *, int *, float  *, int *);
-int    BLASFUNC(zaxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(xaxpy) (int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(caxpyc)(int *, float  *, float  *, int *, float  *, int *);
-int    BLASFUNC(zaxpyc)(int *, double *, double *, int *, double *, int *);
-int    BLASFUNC(xaxpyc)(int *, double *, double *, int *, double *, int *);
+int    BLASFUNC(saxpy) (const int *, const float  *, const float  *, const int *, float  *, const int *);
+int    BLASFUNC(daxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(qaxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(caxpy) (const int *, const float  *, const float  *, const int *, float  *, const int *);
+int    BLASFUNC(zaxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(xaxpy) (const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(caxpyc)(const int *, const float  *, const float  *, const int *, float  *, const int *);
+int    BLASFUNC(zaxpyc)(const int *, const double *, const double *, const int *, double *, const int *);
+int    BLASFUNC(xaxpyc)(const int *, const double *, const double *, const int *, double *, const int *);
 
 int    BLASFUNC(scopy) (int *, float  *, int *, float  *, int *);
 int    BLASFUNC(dcopy) (int *, double *, int *, double *, int *);
@@ -177,31 +177,19 @@
 int BLASFUNC(xgerc)(int *,    int *, double *, double *, int *,
 		    double *, int *, double *, int *);
 
-int BLASFUNC(sgemv)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *, float  *, float  *, int *);
-int BLASFUNC(dgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
-int BLASFUNC(qgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
-int BLASFUNC(cgemv)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
-int BLASFUNC(xgemv)(char *, int *, int *, double *, double *, int *,
-		    double *, int *, double *, double *, int *);
+int BLASFUNC(sgemv)(const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(cgemv)(const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xgemv)(const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
-int BLASFUNC(strsv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(dtrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(qtrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(ctrsv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(ztrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(xtrsv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
+int BLASFUNC(strsv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dtrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrsv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(ztrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrsv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
 
 int BLASFUNC(stpsv) (char *, char *, char *, int *, float  *, float  *, int *);
 int BLASFUNC(dtpsv) (char *, char *, char *, int *, double *, double *, int *);
@@ -210,18 +198,12 @@
 int BLASFUNC(ztpsv) (char *, char *, char *, int *, double *, double *, int *);
 int BLASFUNC(xtpsv) (char *, char *, char *, int *, double *, double *, int *);
 
-int BLASFUNC(strmv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(dtrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(qtrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(ctrmv) (char *, char *, char *, int *, float  *, int *,
-		     float  *, int *);
-int BLASFUNC(ztrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
-int BLASFUNC(xtrmv) (char *, char *, char *, int *, double *, int *,
-		     double *, int *);
+int BLASFUNC(strmv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dtrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrmv) (const char *, const char *, const char *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(ztrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrmv) (const char *, const char *, const char *, const int *, const double *, const int *, double *, const int *);
 
 int BLASFUNC(stpmv) (char *, char *, char *, int *, float  *, float  *, int *);
 int BLASFUNC(dtpmv) (char *, char *, char *, int *, double *, double *, int *);
@@ -244,18 +226,9 @@
 int BLASFUNC(ztbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *);
 int BLASFUNC(xtbsv) (char *, char *, char *, int *, int *, double *, int *, double *, int *);
 
-int BLASFUNC(ssymv) (char *, int *, float  *, float *, int *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(dsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(qsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(csymv) (char *, int *, float  *, float *, int *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(zsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(xsymv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
+int BLASFUNC(ssymv) (const char *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(sspmv) (char *, int *, float  *, float *,
 		     float  *, int *, float *, float *, int *);
@@ -263,38 +236,17 @@
 		     double  *, int *, double *, double *, int *);
 int BLASFUNC(qspmv) (char *, int *, double  *, double *,
 		     double  *, int *, double *, double *, int *);
-int BLASFUNC(cspmv) (char *, int *, float  *, float *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(zspmv) (char *, int *, double  *, double *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(xspmv) (char *, int *, double  *, double *,
-		     double  *, int *, double *, double *, int *);
 
-int BLASFUNC(ssyr) (char *, int *, float   *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(dsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
-int BLASFUNC(qsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
-int BLASFUNC(csyr) (char *, int *, float   *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(zsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
-int BLASFUNC(xsyr) (char *, int *, double  *, double *, int *,
-		    double *, int *);
+int BLASFUNC(ssyr) (const char *, const int *, const float   *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dsyr) (const char *, const int *, const double  *, const double *, const int *, double *, const int *);
+int BLASFUNC(qsyr) (const char *, const int *, const double  *, const double *, const int *, double *, const int *);
 
-int BLASFUNC(ssyr2) (char *, int *, float   *,
-		     float  *, int *, float  *, int *, float  *, int *);
-int BLASFUNC(dsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
-int BLASFUNC(qsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
-int BLASFUNC(csyr2) (char *, int *, float   *,
-		     float  *, int *, float  *, int *, float  *, int *);
-int BLASFUNC(zsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
-int BLASFUNC(xsyr2) (char *, int *, double  *,
-		     double *, int *, double *, int *, double *, int *);
+int BLASFUNC(ssyr2) (const char *, const int *, const float   *, const float  *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(dsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(qsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(csyr2) (const char *, const int *, const float   *, const float  *, const int *, const float  *, const int *, float  *, const int *);
+int BLASFUNC(zsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
+int BLASFUNC(xsyr2) (const char *, const int *, const double  *, const double *, const int *, const double *, const int *, double *, const int *);
 
 int BLASFUNC(sspr) (char *, int *, float   *, float  *, int *,
 		    float  *);
@@ -302,12 +254,6 @@
 		    double *);
 int BLASFUNC(qspr) (char *, int *, double  *, double *, int *,
 		    double *);
-int BLASFUNC(cspr) (char *, int *, float   *, float  *, int *,
-		    float  *);
-int BLASFUNC(zspr) (char *, int *, double  *, double *, int *,
-		    double *);
-int BLASFUNC(xspr) (char *, int *, double  *, double *, int *,
-		    double *);
 
 int BLASFUNC(sspr2) (char *, int *, float   *,
 		     float  *, int *, float  *, int *, float  *);
@@ -347,12 +293,9 @@
 int BLASFUNC(xhpr2) (char *, int *, double  *,
 		     double *, int *, double *, int *, double *);
 
-int BLASFUNC(chemv) (char *, int *, float  *, float *, int *,
-		     float  *, int *, float *, float *, int *);
-int BLASFUNC(zhemv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
-int BLASFUNC(xhemv) (char *, int *, double  *, double *, int *,
-		     double  *, int *, double *, double *, int *);
+int BLASFUNC(chemv) (const char *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zhemv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xhemv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(chpmv) (char *, int *, float  *, float *,
 		     float  *, int *, float *, float *, int *);
@@ -401,18 +344,12 @@
 
 /* Level 3 routines */
 
-int BLASFUNC(sgemm)(char *, char *, int *, int *, int *, float *,
-	   float  *, int *, float  *, int *, float  *, float  *, int *);
-int BLASFUNC(dgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
-int BLASFUNC(qgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
-int BLASFUNC(cgemm)(char *, char *, int *, int *, int *, float *,
-	   float  *, int *, float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
-int BLASFUNC(xgemm)(char *, char *, int *, int *, int *, double *,
-	   double *, int *, double *, int *, double *, double *, int *);
+int BLASFUNC(sgemm)(const char *, const char *, const int *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(cgemm)(const char *, const char *, const int *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xgemm)(const char *, const char *, const int *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(cgemm3m)(char *, char *, int *, int *, int *, float *,
 	   float  *, int *, float  *, int *, float  *, float  *, int *);
@@ -434,84 +371,48 @@
 		     double *, double  *, int *, double  *, int *,
 		     double *, double  *, int *);
 
-int BLASFUNC(strsm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(dtrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(qtrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(ctrsm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(ztrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(xtrsm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
+int BLASFUNC(strsm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(dtrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(ztrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrsm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
 
-int BLASFUNC(strmm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(dtrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(qtrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(ctrmm)(char *, char *, char *, char *, int *, int *,
-	   float *,  float *, int *, float *, int *);
-int BLASFUNC(ztrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
-int BLASFUNC(xtrmm)(char *, char *, char *, char *, int *, int *,
-	   double *,  double *, int *, double *, int *);
+int BLASFUNC(strmm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(dtrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(qtrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(ctrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const float *,  const float *,  const int *, float *,  const int *);
+int BLASFUNC(ztrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
+int BLASFUNC(xtrmm)(const char *, const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, double *, const int *);
 
-int BLASFUNC(ssymm)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(dsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(qsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(csymm)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(xsymm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
+int BLASFUNC(ssymm)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(csymm)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xsymm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
-int BLASFUNC(csymm3m)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zsymm3m)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(xsymm3m)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
+int BLASFUNC(csymm3m)(char *, char *, int *, int *, float  *, float  *, int *, float  *, int *, float  *, float  *, int *);
+int BLASFUNC(zsymm3m)(char *, char *, int *, int *, double *, double *, int *, double *, int *, double *, double *, int *);
+int BLASFUNC(xsymm3m)(char *, char *, int *, int *, double *, double *, int *, double *, int *, double *, double *, int *);
 
-int BLASFUNC(ssyrk)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, float  *, int *);
-int BLASFUNC(dsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(qsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(csyrk)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, float  *, int *);
-int BLASFUNC(zsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(xsyrk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
+int BLASFUNC(ssyrk)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(qsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(csyrk)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xsyrk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
 
-int BLASFUNC(ssyr2k)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(dsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(qsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(csyr2k)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(zsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(xsyr2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
+int BLASFUNC(ssyr2k)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(dsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(qsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(csyr2k)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(xsyr2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
 
-int BLASFUNC(chemm)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, int *, float  *, float  *, int *);
-int BLASFUNC(zhemm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
-int BLASFUNC(xhemm)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, int *, double *, double *, int *);
+int BLASFUNC(chemm)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zhemm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xhemm)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
 
 int BLASFUNC(chemm3m)(char *, char *, int *, int *, float  *, float  *, int *,
 	   float  *, int *, float  *, float  *, int *);
@@ -520,136 +421,17 @@
 int BLASFUNC(xhemm3m)(char *, char *, int *, int *, double *, double *, int *,
 	   double *, int *, double *, double *, int *);
 
-int BLASFUNC(cherk)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float  *, float  *, int *);
-int BLASFUNC(zherk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
-int BLASFUNC(xherk)(char *, char *, int *, int *, double *, double *, int *,
-	   double *, double *, int *);
+int BLASFUNC(cherk)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zherk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xherk)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, double *, const int *);
 
-int BLASFUNC(cher2k)(char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(zher2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(xher2k)(char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(cher2m)(char *, char *, char *, int *, int *, float  *, float  *, int *,
-	   float *, int *, float  *, float  *, int *);
-int BLASFUNC(zher2m)(char *, char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
-int BLASFUNC(xher2m)(char *, char *, char *, int *, int *, double *, double *, int *,
-	   double*, int *, double *, double *, int *);
+int BLASFUNC(cher2k)(const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zher2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xher2k)(const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(cher2m)(const char *, const char *, const char *, const int *, const int *, const float  *, const float  *, const int *, const float *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zher2m)(const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
+int BLASFUNC(xher2m)(const char *, const char *, const char *, const int *, const int *, const double *, const double *, const int *, const double*, const int *, const double *, double *, const int *);
 
-int BLASFUNC(sgemt)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(dgemt)(char *, int *, int *, double *, double *, int *,
-		    double *, int *);
-int BLASFUNC(cgemt)(char *, int *, int *, float  *, float  *, int *,
-		    float  *, int *);
-int BLASFUNC(zgemt)(char *, int *, int *, double *, double *, int *,
-		    double *, int *);
-
-int BLASFUNC(sgema)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(dgema)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-int BLASFUNC(cgema)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(zgema)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-
-int BLASFUNC(sgems)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(dgems)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-int BLASFUNC(cgems)(char *, char *, int *, int *, float  *,
-		    float  *, int *, float *, float  *, int *, float *, int *);
-int BLASFUNC(zgems)(char *, char *, int *, int *, double *,
-		    double *, int *, double*, double *, int *, double*, int *);
-
-int BLASFUNC(sgetf2)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(dgetf2)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(qgetf2)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(cgetf2)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(zgetf2)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(xgetf2)(int *, int *, double *, int *, int *, int *);
-
-int BLASFUNC(sgetrf)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(dgetrf)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(qgetrf)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(cgetrf)(int *, int *, float  *, int *, int *, int *);
-int BLASFUNC(zgetrf)(int *, int *, double *, int *, int *, int *);
-int BLASFUNC(xgetrf)(int *, int *, double *, int *, int *, int *);
-
-int BLASFUNC(slaswp)(int *, float  *, int *, int *, int *, int *, int *);
-int BLASFUNC(dlaswp)(int *, double *, int *, int *, int *, int *, int *);
-int BLASFUNC(qlaswp)(int *, double *, int *, int *, int *, int *, int *);
-int BLASFUNC(claswp)(int *, float  *, int *, int *, int *, int *, int *);
-int BLASFUNC(zlaswp)(int *, double *, int *, int *, int *, int *, int *);
-int BLASFUNC(xlaswp)(int *, double *, int *, int *, int *, int *, int *);
-
-int BLASFUNC(sgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
-int BLASFUNC(dgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-int BLASFUNC(qgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-int BLASFUNC(cgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
-int BLASFUNC(zgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-int BLASFUNC(xgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
-
-int BLASFUNC(sgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
-int BLASFUNC(dgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-int BLASFUNC(qgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-int BLASFUNC(cgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
-int BLASFUNC(zgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-int BLASFUNC(xgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
-
-int BLASFUNC(spotf2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dpotf2)(char *, int *, double *, int *, int *);
-int BLASFUNC(qpotf2)(char *, int *, double *, int *, int *);
-int BLASFUNC(cpotf2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zpotf2)(char *, int *, double *, int *, int *);
-int BLASFUNC(xpotf2)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(spotrf)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dpotrf)(char *, int *, double *, int *, int *);
-int BLASFUNC(qpotrf)(char *, int *, double *, int *, int *);
-int BLASFUNC(cpotrf)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zpotrf)(char *, int *, double *, int *, int *);
-int BLASFUNC(xpotrf)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(slauu2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dlauu2)(char *, int *, double *, int *, int *);
-int BLASFUNC(qlauu2)(char *, int *, double *, int *, int *);
-int BLASFUNC(clauu2)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zlauu2)(char *, int *, double *, int *, int *);
-int BLASFUNC(xlauu2)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(slauum)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dlauum)(char *, int *, double *, int *, int *);
-int BLASFUNC(qlauum)(char *, int *, double *, int *, int *);
-int BLASFUNC(clauum)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zlauum)(char *, int *, double *, int *, int *);
-int BLASFUNC(xlauum)(char *, int *, double *, int *, int *);
-
-int BLASFUNC(strti2)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(dtrti2)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(qtrti2)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(ctrti2)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(ztrti2)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(xtrti2)(char *, char *, int *, double *, int *, int *);
-
-int BLASFUNC(strtri)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(dtrtri)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(qtrtri)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(ctrtri)(char *, char *, int *, float  *, int *, int *);
-int BLASFUNC(ztrtri)(char *, char *, int *, double *, int *, int *);
-int BLASFUNC(xtrtri)(char *, char *, int *, double *, int *, int *);
-
-int BLASFUNC(spotri)(char *, int *, float  *, int *, int *);
-int BLASFUNC(dpotri)(char *, int *, double *, int *, int *);
-int BLASFUNC(qpotri)(char *, int *, double *, int *, int *);
-int BLASFUNC(cpotri)(char *, int *, float  *, int *, int *);
-int BLASFUNC(zpotri)(char *, int *, double *, int *, int *);
-int BLASFUNC(xpotri)(char *, int *, double *, int *, int *);
 
 #ifdef __cplusplus
 }
diff --git a/Eigen/src/misc/lapack.h b/Eigen/src/misc/lapack.h
new file mode 100644
index 0000000..249f357
--- /dev/null
+++ b/Eigen/src/misc/lapack.h
@@ -0,0 +1,152 @@
+#ifndef LAPACK_H
+#define LAPACK_H
+
+#include "blas.h"
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+int BLASFUNC(csymv) (const char *, const int *, const float  *, const float  *, const int *, const float  *, const int *, const float  *, float  *, const int *);
+int BLASFUNC(zsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+int BLASFUNC(xsymv) (const char *, const int *, const double *, const double *, const int *, const double *, const int *, const double *, double *, const int *);
+
+
+int BLASFUNC(cspmv) (char *, int *, float  *, float *,
+         float  *, int *, float *, float *, int *);
+int BLASFUNC(zspmv) (char *, int *, double  *, double *,
+         double  *, int *, double *, double *, int *);
+int BLASFUNC(xspmv) (char *, int *, double  *, double *,
+         double  *, int *, double *, double *, int *);
+
+int BLASFUNC(csyr) (char *, int *, float   *, float  *, int *,
+        float  *, int *);
+int BLASFUNC(zsyr) (char *, int *, double  *, double *, int *,
+        double *, int *);
+int BLASFUNC(xsyr) (char *, int *, double  *, double *, int *,
+        double *, int *);
+
+int BLASFUNC(cspr) (char *, int *, float   *, float  *, int *,
+        float  *);
+int BLASFUNC(zspr) (char *, int *, double  *, double *, int *,
+        double *);
+int BLASFUNC(xspr) (char *, int *, double  *, double *, int *,
+        double *);
+
+int BLASFUNC(sgemt)(char *, int *, int *, float  *, float  *, int *,
+        float  *, int *);
+int BLASFUNC(dgemt)(char *, int *, int *, double *, double *, int *,
+        double *, int *);
+int BLASFUNC(cgemt)(char *, int *, int *, float  *, float  *, int *,
+        float  *, int *);
+int BLASFUNC(zgemt)(char *, int *, int *, double *, double *, int *,
+        double *, int *);
+
+int BLASFUNC(sgema)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(dgema)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+int BLASFUNC(cgema)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(zgema)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+
+int BLASFUNC(sgems)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(dgems)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+int BLASFUNC(cgems)(char *, char *, int *, int *, float  *,
+        float  *, int *, float *, float  *, int *, float *, int *);
+int BLASFUNC(zgems)(char *, char *, int *, int *, double *,
+        double *, int *, double*, double *, int *, double*, int *);
+
+int BLASFUNC(sgetf2)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(dgetf2)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(qgetf2)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(cgetf2)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(zgetf2)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(xgetf2)(int *, int *, double *, int *, int *, int *);
+
+int BLASFUNC(sgetrf)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(dgetrf)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(qgetrf)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(cgetrf)(int *, int *, float  *, int *, int *, int *);
+int BLASFUNC(zgetrf)(int *, int *, double *, int *, int *, int *);
+int BLASFUNC(xgetrf)(int *, int *, double *, int *, int *, int *);
+
+int BLASFUNC(slaswp)(int *, float  *, int *, int *, int *, int *, int *);
+int BLASFUNC(dlaswp)(int *, double *, int *, int *, int *, int *, int *);
+int BLASFUNC(qlaswp)(int *, double *, int *, int *, int *, int *, int *);
+int BLASFUNC(claswp)(int *, float  *, int *, int *, int *, int *, int *);
+int BLASFUNC(zlaswp)(int *, double *, int *, int *, int *, int *, int *);
+int BLASFUNC(xlaswp)(int *, double *, int *, int *, int *, int *, int *);
+
+int BLASFUNC(sgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
+int BLASFUNC(dgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+int BLASFUNC(qgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+int BLASFUNC(cgetrs)(char *, int *, int *, float  *, int *, int *, float  *, int *, int *);
+int BLASFUNC(zgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+int BLASFUNC(xgetrs)(char *, int *, int *, double *, int *, int *, double *, int *, int *);
+
+int BLASFUNC(sgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
+int BLASFUNC(dgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+int BLASFUNC(qgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+int BLASFUNC(cgesv)(int *, int *, float  *, int *, int *, float *, int *, int *);
+int BLASFUNC(zgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+int BLASFUNC(xgesv)(int *, int *, double *, int *, int *, double*, int *, int *);
+
+int BLASFUNC(spotf2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dpotf2)(char *, int *, double *, int *, int *);
+int BLASFUNC(qpotf2)(char *, int *, double *, int *, int *);
+int BLASFUNC(cpotf2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zpotf2)(char *, int *, double *, int *, int *);
+int BLASFUNC(xpotf2)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(spotrf)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dpotrf)(char *, int *, double *, int *, int *);
+int BLASFUNC(qpotrf)(char *, int *, double *, int *, int *);
+int BLASFUNC(cpotrf)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zpotrf)(char *, int *, double *, int *, int *);
+int BLASFUNC(xpotrf)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(slauu2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dlauu2)(char *, int *, double *, int *, int *);
+int BLASFUNC(qlauu2)(char *, int *, double *, int *, int *);
+int BLASFUNC(clauu2)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zlauu2)(char *, int *, double *, int *, int *);
+int BLASFUNC(xlauu2)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(slauum)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dlauum)(char *, int *, double *, int *, int *);
+int BLASFUNC(qlauum)(char *, int *, double *, int *, int *);
+int BLASFUNC(clauum)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zlauum)(char *, int *, double *, int *, int *);
+int BLASFUNC(xlauum)(char *, int *, double *, int *, int *);
+
+int BLASFUNC(strti2)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(dtrti2)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(qtrti2)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(ctrti2)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(ztrti2)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(xtrti2)(char *, char *, int *, double *, int *, int *);
+
+int BLASFUNC(strtri)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(dtrtri)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(qtrtri)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(ctrtri)(char *, char *, int *, float  *, int *, int *);
+int BLASFUNC(ztrtri)(char *, char *, int *, double *, int *, int *);
+int BLASFUNC(xtrtri)(char *, char *, int *, double *, int *, int *);
+
+int BLASFUNC(spotri)(char *, int *, float  *, int *, int *);
+int BLASFUNC(dpotri)(char *, int *, double *, int *, int *);
+int BLASFUNC(qpotri)(char *, int *, double *, int *, int *);
+int BLASFUNC(cpotri)(char *, int *, float  *, int *, int *);
+int BLASFUNC(zpotri)(char *, int *, double *, int *, int *);
+int BLASFUNC(xpotri)(char *, int *, double *, int *, int *);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/Eigen/src/misc/lapacke.h b/Eigen/src/misc/lapacke.h
new file mode 100755
index 0000000..8c7e79b
--- /dev/null
+++ b/Eigen/src/misc/lapacke.h
@@ -0,0 +1,16291 @@
+/*****************************************************************************
+  Copyright (c) 2010, Intel Corp.
+  All rights reserved.
+
+  Redistribution and use in source and binary forms, with or without
+  modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice,
+      this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+    * Neither the name of Intel Corporation nor the names of its contributors
+      may be used to endorse or promote products derived from this software
+      without specific prior written permission.
+
+  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+  IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+  ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
+  LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
+  THE POSSIBILITY OF SUCH DAMAGE.
+******************************************************************************
+* Contents: Native C interface to LAPACK
+* Author: Intel Corporation
+* Generated November, 2011
+*****************************************************************************/
+
+#ifndef _MKL_LAPACKE_H_
+
+#ifndef _LAPACKE_H_
+#define _LAPACKE_H_
+
+/*
+*  Turn on HAVE_LAPACK_CONFIG_H to redefine C-LAPACK datatypes
+*/
+#ifdef HAVE_LAPACK_CONFIG_H
+#include "lapacke_config.h"
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+
+#include <stdlib.h>
+
+#ifndef lapack_int
+#define lapack_int     int
+#endif
+
+#ifndef lapack_logical
+#define lapack_logical lapack_int
+#endif
+
+/* Complex types are structures equivalent to the
+* Fortran complex types COMPLEX(4) and COMPLEX(8).
+*
+* One can also redefine the types with his own types
+* for example by including in the code definitions like
+*
+* #define lapack_complex_float std::complex<float>
+* #define lapack_complex_double std::complex<double>
+*
+* or define these types in the command line:
+*
+* -Dlapack_complex_float="std::complex<float>"
+* -Dlapack_complex_double="std::complex<double>"
+*/
+
+#ifndef LAPACK_COMPLEX_CUSTOM
+
+/* Complex type (single precision) */
+#ifndef lapack_complex_float
+#include <complex.h>
+#define lapack_complex_float    float _Complex
+#endif
+
+#ifndef lapack_complex_float_real
+#define lapack_complex_float_real(z)       (creal(z))
+#endif
+
+#ifndef lapack_complex_float_imag
+#define lapack_complex_float_imag(z)       (cimag(z))
+#endif
+
+lapack_complex_float lapack_make_complex_float( float re, float im );
+
+/* Complex type (double precision) */
+#ifndef lapack_complex_double
+#include <complex.h>
+#define lapack_complex_double   double _Complex
+#endif
+
+#ifndef lapack_complex_double_real
+#define lapack_complex_double_real(z)      (creal(z))
+#endif
+
+#ifndef lapack_complex_double_imag
+#define lapack_complex_double_imag(z)       (cimag(z))
+#endif
+
+lapack_complex_double lapack_make_complex_double( double re, double im );
+
+#endif
+
+#ifndef LAPACKE_malloc
+#define LAPACKE_malloc( size ) malloc( size )
+#endif
+#ifndef LAPACKE_free
+#define LAPACKE_free( p )      free( p )
+#endif
+
+#define LAPACK_C2INT( x ) (lapack_int)(*((float*)&x ))
+#define LAPACK_Z2INT( x ) (lapack_int)(*((double*)&x ))
+
+#define LAPACK_ROW_MAJOR               101
+#define LAPACK_COL_MAJOR               102
+
+#define LAPACK_WORK_MEMORY_ERROR       -1010
+#define LAPACK_TRANSPOSE_MEMORY_ERROR  -1011
+
+/* Callback logical functions of one, two, or three arguments are used
+*  to select eigenvalues to sort to the top left of the Schur form.
+*  The value is selected if function returns TRUE (non-zero). */
+
+typedef lapack_logical (*LAPACK_S_SELECT2) ( const float*, const float* );
+typedef lapack_logical (*LAPACK_S_SELECT3)
+    ( const float*, const float*, const float* );
+typedef lapack_logical (*LAPACK_D_SELECT2) ( const double*, const double* );
+typedef lapack_logical (*LAPACK_D_SELECT3)
+    ( const double*, const double*, const double* );
+
+typedef lapack_logical (*LAPACK_C_SELECT1) ( const lapack_complex_float* );
+typedef lapack_logical (*LAPACK_C_SELECT2)
+    ( const lapack_complex_float*, const lapack_complex_float* );
+typedef lapack_logical (*LAPACK_Z_SELECT1) ( const lapack_complex_double* );
+typedef lapack_logical (*LAPACK_Z_SELECT2)
+    ( const lapack_complex_double*, const lapack_complex_double* );
+
+#include "lapacke_mangling.h"
+
+#define LAPACK_lsame LAPACK_GLOBAL(lsame,LSAME)
+lapack_logical LAPACK_lsame( char* ca,  char* cb,
+                              lapack_int lca, lapack_int lcb );
+
+/* C-LAPACK function prototypes */
+
+lapack_int LAPACKE_sbdsdc( int matrix_order, char uplo, char compq,
+                           lapack_int n, float* d, float* e, float* u,
+                           lapack_int ldu, float* vt, lapack_int ldvt, float* q,
+                           lapack_int* iq );
+lapack_int LAPACKE_dbdsdc( int matrix_order, char uplo, char compq,
+                           lapack_int n, double* d, double* e, double* u,
+                           lapack_int ldu, double* vt, lapack_int ldvt,
+                           double* q, lapack_int* iq );
+
+lapack_int LAPACKE_sbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           float* d, float* e, float* vt, lapack_int ldvt,
+                           float* u, lapack_int ldu, float* c, lapack_int ldc );
+lapack_int LAPACKE_dbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           double* d, double* e, double* vt, lapack_int ldvt,
+                           double* u, lapack_int ldu, double* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_cbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           float* d, float* e, lapack_complex_float* vt,
+                           lapack_int ldvt, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_zbdsqr( int matrix_order, char uplo, lapack_int n,
+                           lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                           double* d, double* e, lapack_complex_double* vt,
+                           lapack_int ldvt, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* c,
+                           lapack_int ldc );
+
+lapack_int LAPACKE_sdisna( char job, lapack_int m, lapack_int n, const float* d,
+                           float* sep );
+lapack_int LAPACKE_ddisna( char job, lapack_int m, lapack_int n,
+                           const double* d, double* sep );
+
+lapack_int LAPACKE_sgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, float* ab, lapack_int ldab, float* d,
+                           float* e, float* q, lapack_int ldq, float* pt,
+                           lapack_int ldpt, float* c, lapack_int ldc );
+lapack_int LAPACKE_dgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, double* ab, lapack_int ldab,
+                           double* d, double* e, double* q, lapack_int ldq,
+                           double* pt, lapack_int ldpt, double* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_cgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, lapack_complex_float* ab,
+                           lapack_int ldab, float* d, float* e,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* pt, lapack_int ldpt,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zgbbrd( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int ncc, lapack_int kl,
+                           lapack_int ku, lapack_complex_double* ab,
+                           lapack_int ldab, double* d, double* e,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* pt, lapack_int ldpt,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* ab,
+                           lapack_int ldab, const lapack_int* ipiv, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* ab,
+                           lapack_int ldab, const lapack_int* ipiv,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zgbcon( int matrix_order, char norm, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* ab,
+                           lapack_int ldab, float* r, float* c, float* rowcnd,
+                           float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* ab,
+                           lapack_int ldab, double* r, double* c,
+                           double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           float* r, float* c, float* rowcnd, float* colcnd,
+                           float* amax );
+lapack_int LAPACKE_zgbequ( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           double* r, double* c, double* rowcnd, double* colcnd,
+                           double* amax );
+
+lapack_int LAPACKE_sgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku, const float* ab,
+                            lapack_int ldab, float* r, float* c, float* rowcnd,
+                            float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku, const double* ab,
+                            lapack_int ldab, double* r, double* c,
+                            double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku,
+                            const lapack_complex_float* ab, lapack_int ldab,
+                            float* r, float* c, float* rowcnd, float* colcnd,
+                            float* amax );
+lapack_int LAPACKE_zgbequb( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_int kl, lapack_int ku,
+                            const lapack_complex_double* ab, lapack_int ldab,
+                            double* r, double* c, double* rowcnd,
+                            double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const float* ab, lapack_int ldab, const float* afb,
+                           lapack_int ldafb, const lapack_int* ipiv,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const double* ab, lapack_int ldab, const double* afb,
+                           lapack_int ldafb, const lapack_int* ipiv,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* afb, lapack_int ldafb,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgbrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* afb, lapack_int ldafb,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const float* ab, lapack_int ldab,
+                            const float* afb, lapack_int ldafb,
+                            const lapack_int* ipiv, const float* r,
+                            const float* c, const float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const double* ab, lapack_int ldab,
+                            const double* afb, lapack_int ldafb,
+                            const lapack_int* ipiv, const double* r,
+                            const double* c, const double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const lapack_complex_float* ab,
+                            lapack_int ldab, const lapack_complex_float* afb,
+                            lapack_int ldafb, const lapack_int* ipiv,
+                            const float* r, const float* c,
+                            const lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zgbrfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, const lapack_complex_double* ab,
+                            lapack_int ldab, const lapack_complex_double* afb,
+                            lapack_int ldafb, const lapack_int* ipiv,
+                            const double* r, const double* c,
+                            const lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_sgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs, float* ab,
+                          lapack_int ldab, lapack_int* ipiv, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs, double* ab,
+                          lapack_int ldab, lapack_int* ipiv, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs,
+                          lapack_complex_float* ab, lapack_int ldab,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zgbsv( int matrix_order, lapack_int n, lapack_int kl,
+                          lapack_int ku, lapack_int nrhs,
+                          lapack_complex_double* ab, lapack_int ldab,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_sgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, float* ab, lapack_int ldab,
+                           float* afb, lapack_int ldafb, lapack_int* ipiv,
+                           char* equed, float* r, float* c, float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr,
+                           float* rpivot );
+lapack_int LAPACKE_dgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, double* ab, lapack_int ldab,
+                           double* afb, lapack_int ldafb, lapack_int* ipiv,
+                           char* equed, double* r, double* c, double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr,
+                           double* rpivot );
+lapack_int LAPACKE_cgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, lapack_complex_float* ab,
+                           lapack_int ldab, lapack_complex_float* afb,
+                           lapack_int ldafb, lapack_int* ipiv, char* equed,
+                           float* r, float* c, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr, float* rpivot );
+lapack_int LAPACKE_zgbsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int kl, lapack_int ku,
+                           lapack_int nrhs, lapack_complex_double* ab,
+                           lapack_int ldab, lapack_complex_double* afb,
+                           lapack_int ldafb, lapack_int* ipiv, char* equed,
+                           double* r, double* c, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, double* rcond, double* ferr,
+                           double* berr, double* rpivot );
+
+lapack_int LAPACKE_sgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, float* ab, lapack_int ldab,
+                            float* afb, lapack_int ldafb, lapack_int* ipiv,
+                            char* equed, float* r, float* c, float* b,
+                            lapack_int ldb, float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, double* ab, lapack_int ldab,
+                            double* afb, lapack_int ldafb, lapack_int* ipiv,
+                            char* equed, double* r, double* c, double* b,
+                            lapack_int ldb, double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+lapack_int LAPACKE_cgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, lapack_complex_float* ab,
+                            lapack_int ldab, lapack_complex_float* afb,
+                            lapack_int ldafb, lapack_int* ipiv, char* equed,
+                            float* r, float* c, lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* rpvgrw,
+                            float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zgbsvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int kl, lapack_int ku,
+                            lapack_int nrhs, lapack_complex_double* ab,
+                            lapack_int ldab, lapack_complex_double* afb,
+                            lapack_int ldafb, lapack_int* ipiv, char* equed,
+                            double* r, double* c, lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* rpvgrw,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_sgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, float* ab,
+                           lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_dgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, double* ab,
+                           lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_cgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zgbtrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_sgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const float* ab, lapack_int ldab,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const double* ab, lapack_int ldab,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zgbtrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int kl, lapack_int ku, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_sgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* scale,
+                           lapack_int m, float* v, lapack_int ldv );
+lapack_int LAPACKE_dgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* scale,
+                           lapack_int m, double* v, lapack_int ldv );
+lapack_int LAPACKE_cgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* scale,
+                           lapack_int m, lapack_complex_float* v,
+                           lapack_int ldv );
+lapack_int LAPACKE_zgebak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* scale,
+                           lapack_int m, lapack_complex_double* v,
+                           lapack_int ldv );
+
+lapack_int LAPACKE_sgebal( int matrix_order, char job, lapack_int n, float* a,
+                           lapack_int lda, lapack_int* ilo, lapack_int* ihi,
+                           float* scale );
+lapack_int LAPACKE_dgebal( int matrix_order, char job, lapack_int n, double* a,
+                           lapack_int lda, lapack_int* ilo, lapack_int* ihi,
+                           double* scale );
+lapack_int LAPACKE_cgebal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ilo, lapack_int* ihi, float* scale );
+lapack_int LAPACKE_zgebal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ilo, lapack_int* ihi, double* scale );
+
+lapack_int LAPACKE_sgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* d, float* e,
+                           float* tauq, float* taup );
+lapack_int LAPACKE_dgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* d, double* e,
+                           double* tauq, double* taup );
+lapack_int LAPACKE_cgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda, float* d,
+                           float* e, lapack_complex_float* tauq,
+                           lapack_complex_float* taup );
+lapack_int LAPACKE_zgebrd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda, double* d,
+                           double* e, lapack_complex_double* tauq,
+                           lapack_complex_double* taup );
+
+lapack_int LAPACKE_sgecon( int matrix_order, char norm, lapack_int n,
+                           const float* a, lapack_int lda, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dgecon( int matrix_order, char norm, lapack_int n,
+                           const double* a, lapack_int lda, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_cgecon( int matrix_order, char norm, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_zgecon( int matrix_order, char norm, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double anorm, double* rcond );
+
+lapack_int LAPACKE_sgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const float* a, lapack_int lda, float* r, float* c,
+                           float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const double* a, lapack_int lda, double* r,
+                           double* c, double* rowcnd, double* colcnd,
+                           double* amax );
+lapack_int LAPACKE_cgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float* r, float* c, float* rowcnd, float* colcnd,
+                           float* amax );
+lapack_int LAPACKE_zgeequ( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double* r, double* c, double* rowcnd, double* colcnd,
+                           double* amax );
+
+lapack_int LAPACKE_sgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const float* a, lapack_int lda, float* r, float* c,
+                            float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const double* a, lapack_int lda, double* r,
+                            double* c, double* rowcnd, double* colcnd,
+                            double* amax );
+lapack_int LAPACKE_cgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* r, float* c, float* rowcnd, float* colcnd,
+                            float* amax );
+lapack_int LAPACKE_zgeequb( int matrix_order, lapack_int m, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* r, double* c, double* rowcnd,
+                            double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_S_SELECT2 select, lapack_int n, float* a,
+                          lapack_int lda, lapack_int* sdim, float* wr,
+                          float* wi, float* vs, lapack_int ldvs );
+lapack_int LAPACKE_dgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_D_SELECT2 select, lapack_int n, double* a,
+                          lapack_int lda, lapack_int* sdim, double* wr,
+                          double* wi, double* vs, lapack_int ldvs );
+lapack_int LAPACKE_cgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_C_SELECT1 select, lapack_int n,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_int* sdim, lapack_complex_float* w,
+                          lapack_complex_float* vs, lapack_int ldvs );
+lapack_int LAPACKE_zgees( int matrix_order, char jobvs, char sort,
+                          LAPACK_Z_SELECT1 select, lapack_int n,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_int* sdim, lapack_complex_double* w,
+                          lapack_complex_double* vs, lapack_int ldvs );
+
+lapack_int LAPACKE_sgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_S_SELECT2 select, char sense, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* sdim,
+                           float* wr, float* wi, float* vs, lapack_int ldvs,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_dgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_D_SELECT2 select, char sense, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* sdim,
+                           double* wr, double* wi, double* vs, lapack_int ldvs,
+                           double* rconde, double* rcondv );
+lapack_int LAPACKE_cgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_C_SELECT1 select, char sense, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* sdim, lapack_complex_float* w,
+                           lapack_complex_float* vs, lapack_int ldvs,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_zgeesx( int matrix_order, char jobvs, char sort,
+                           LAPACK_Z_SELECT1 select, char sense, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* sdim, lapack_complex_double* w,
+                           lapack_complex_double* vs, lapack_int ldvs,
+                           double* rconde, double* rcondv );
+
+lapack_int LAPACKE_sgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, float* a, lapack_int lda, float* wr,
+                          float* wi, float* vl, lapack_int ldvl, float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_dgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, double* a, lapack_int lda, double* wr,
+                          double* wi, double* vl, lapack_int ldvl, double* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_cgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* w, lapack_complex_float* vl,
+                          lapack_int ldvl, lapack_complex_float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_zgeev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* w,
+                          lapack_complex_double* vl, lapack_int ldvl,
+                          lapack_complex_double* vr, lapack_int ldvr );
+
+lapack_int LAPACKE_sgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, float* a,
+                           lapack_int lda, float* wr, float* wi, float* vl,
+                           lapack_int ldvl, float* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, float* scale,
+                           float* abnrm, float* rconde, float* rcondv );
+lapack_int LAPACKE_dgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, double* a,
+                           lapack_int lda, double* wr, double* wi, double* vl,
+                           lapack_int ldvl, double* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, double* scale,
+                           double* abnrm, double* rconde, double* rcondv );
+lapack_int LAPACKE_cgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* w, lapack_complex_float* vl,
+                           lapack_int ldvl, lapack_complex_float* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           float* scale, float* abnrm, float* rconde,
+                           float* rcondv );
+lapack_int LAPACKE_zgeevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* w, lapack_complex_double* vl,
+                           lapack_int ldvl, lapack_complex_double* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           double* scale, double* abnrm, double* rconde,
+                           double* rcondv );
+
+lapack_int LAPACKE_sgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, float* a, lapack_int lda,
+                           float* tau );
+lapack_int LAPACKE_dgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, double* a, lapack_int lda,
+                           double* tau );
+lapack_int LAPACKE_cgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* tau );
+lapack_int LAPACKE_zgehrd( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgejsv( int matrix_order, char joba, char jobu, char jobv,
+                           char jobr, char jobt, char jobp, lapack_int m,
+                           lapack_int n, float* a, lapack_int lda, float* sva,
+                           float* u, lapack_int ldu, float* v, lapack_int ldv,
+                           float* stat, lapack_int* istat );
+lapack_int LAPACKE_dgejsv( int matrix_order, char joba, char jobu, char jobv,
+                           char jobr, char jobt, char jobp, lapack_int m,
+                           lapack_int n, double* a, lapack_int lda, double* sva,
+                           double* u, lapack_int ldu, double* v, lapack_int ldv,
+                           double* stat, lapack_int* istat );
+
+lapack_int LAPACKE_sgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgelq2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgelqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs, float* a,
+                          lapack_int lda, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs, double* a,
+                          lapack_int lda, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgels( int matrix_order, char trans, lapack_int m,
+                          lapack_int n, lapack_int nrhs,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_dgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_cgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_zgelsd( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+
+lapack_int LAPACKE_sgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_dgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_cgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float* s, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_zgelss( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double* s, double rcond,
+                           lapack_int* rank );
+
+lapack_int LAPACKE_sgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, lapack_int* jpvt, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_dgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, lapack_int* jpvt,
+                           double rcond, lapack_int* rank );
+lapack_int LAPACKE_cgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_int* jpvt, float rcond,
+                           lapack_int* rank );
+lapack_int LAPACKE_zgelsy( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_int* jpvt, double rcond,
+                           lapack_int* rank );
+
+lapack_int LAPACKE_sgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqlf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* jpvt,
+                           float* tau );
+lapack_int LAPACKE_dgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* jpvt,
+                           double* tau );
+lapack_int LAPACKE_cgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqp3( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* jpvt,
+                           float* tau );
+lapack_int LAPACKE_dgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* jpvt,
+                           double* tau );
+lapack_int LAPACKE_cgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqpf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* jpvt, lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqr2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* tau );
+lapack_int LAPACKE_zgeqrfp( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const float* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const double* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgerfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs, const float* a,
+                            lapack_int lda, const float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* r,
+                            const float* c, const float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs, const double* a,
+                            lapack_int lda, const double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* r,
+                            const double* c, const double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* r,
+                            const float* c, const lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zgerfsx( int matrix_order, char trans, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* r,
+                            const double* c, const lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_sgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_cgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zgerqf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_sgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, float* a, lapack_int lda, float* s,
+                           float* u, lapack_int ldu, float* vt,
+                           lapack_int ldvt );
+lapack_int LAPACKE_dgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, double* a, lapack_int lda, double* s,
+                           double* u, lapack_int ldu, double* vt,
+                           lapack_int ldvt );
+lapack_int LAPACKE_cgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float* s, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* vt,
+                           lapack_int ldvt );
+lapack_int LAPACKE_zgesdd( int matrix_order, char jobz, lapack_int m,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double* s, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* vt,
+                           lapack_int ldvt );
+
+lapack_int LAPACKE_sgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* a, lapack_int lda, lapack_int* ipiv, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* a, lapack_int lda, lapack_int* ipiv,
+                          double* b, lapack_int ldb );
+lapack_int LAPACKE_cgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dsgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                           double* a, lapack_int lda, lapack_int* ipiv,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           lapack_int* iter );
+lapack_int LAPACKE_zcgesv( int matrix_order, lapack_int n, lapack_int nrhs,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, lapack_int* iter );
+
+lapack_int LAPACKE_sgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, float* a, lapack_int lda,
+                           float* s, float* u, lapack_int ldu, float* vt,
+                           lapack_int ldvt, float* superb );
+lapack_int LAPACKE_dgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, double* a,
+                           lapack_int lda, double* s, double* u, lapack_int ldu,
+                           double* vt, lapack_int ldvt, double* superb );
+lapack_int LAPACKE_cgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float* s, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* vt,
+                           lapack_int ldvt, float* superb );
+lapack_int LAPACKE_zgesvd( int matrix_order, char jobu, char jobvt,
+                           lapack_int m, lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double* s, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* vt,
+                           lapack_int ldvt, double* superb );
+
+lapack_int LAPACKE_sgesvj( int matrix_order, char joba, char jobu, char jobv,
+                           lapack_int m, lapack_int n, float* a, lapack_int lda,
+                           float* sva, lapack_int mv, float* v, lapack_int ldv,
+                           float* stat );
+lapack_int LAPACKE_dgesvj( int matrix_order, char joba, char jobu, char jobv,
+                           lapack_int m, lapack_int n, double* a,
+                           lapack_int lda, double* sva, lapack_int mv,
+                           double* v, lapack_int ldv, double* stat );
+
+lapack_int LAPACKE_sgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, float* a,
+                           lapack_int lda, float* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, float* r, float* c,
+                           float* b, lapack_int ldb, float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr,
+                           float* rpivot );
+lapack_int LAPACKE_dgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, double* a,
+                           lapack_int lda, double* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, double* r, double* c,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr,
+                           double* rpivot );
+lapack_int LAPACKE_cgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, float* r, float* c,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr,
+                           float* rpivot );
+lapack_int LAPACKE_zgesvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* af, lapack_int ldaf,
+                           lapack_int* ipiv, char* equed, double* r, double* c,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr,
+                           double* rpivot );
+
+lapack_int LAPACKE_sgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs, float* a,
+                            lapack_int lda, float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* r, float* c,
+                            float* b, lapack_int ldb, float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs, double* a,
+                            lapack_int lda, double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* r, double* c,
+                            double* b, lapack_int ldb, double* x,
+                            lapack_int ldx, double* rcond, double* rpvgrw,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* r, float* c,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zgesvxx( int matrix_order, char fact, char trans,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* r, double* c,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_sgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zgetf2( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zgetrf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetri( int matrix_order, lapack_int n, float* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dgetri( int matrix_order, lapack_int n, double* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_cgetri( int matrix_order, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_zgetri( int matrix_order, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgetrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* lscale,
+                           const float* rscale, lapack_int m, float* v,
+                           lapack_int ldv );
+lapack_int LAPACKE_dggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* lscale,
+                           const double* rscale, lapack_int m, double* v,
+                           lapack_int ldv );
+lapack_int LAPACKE_cggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const float* lscale,
+                           const float* rscale, lapack_int m,
+                           lapack_complex_float* v, lapack_int ldv );
+lapack_int LAPACKE_zggbak( int matrix_order, char job, char side, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, const double* lscale,
+                           const double* rscale, lapack_int m,
+                           lapack_complex_double* v, lapack_int ldv );
+
+lapack_int LAPACKE_sggbal( int matrix_order, char job, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, float* lscale,
+                           float* rscale );
+lapack_int LAPACKE_dggbal( int matrix_order, char job, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, double* lscale,
+                           double* rscale );
+lapack_int LAPACKE_cggbal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, float* lscale,
+                           float* rscale );
+lapack_int LAPACKE_zggbal( int matrix_order, char job, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_int* ilo, lapack_int* ihi, double* lscale,
+                           double* rscale );
+
+lapack_int LAPACKE_sgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_S_SELECT3 selctg, lapack_int n, float* a,
+                          lapack_int lda, float* b, lapack_int ldb,
+                          lapack_int* sdim, float* alphar, float* alphai,
+                          float* beta, float* vsl, lapack_int ldvsl, float* vsr,
+                          lapack_int ldvsr );
+lapack_int LAPACKE_dgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_D_SELECT3 selctg, lapack_int n, double* a,
+                          lapack_int lda, double* b, lapack_int ldb,
+                          lapack_int* sdim, double* alphar, double* alphai,
+                          double* beta, double* vsl, lapack_int ldvsl,
+                          double* vsr, lapack_int ldvsr );
+lapack_int LAPACKE_cgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_C_SELECT2 selctg, lapack_int n,
+                          lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* b, lapack_int ldb,
+                          lapack_int* sdim, lapack_complex_float* alpha,
+                          lapack_complex_float* beta, lapack_complex_float* vsl,
+                          lapack_int ldvsl, lapack_complex_float* vsr,
+                          lapack_int ldvsr );
+lapack_int LAPACKE_zgges( int matrix_order, char jobvsl, char jobvsr, char sort,
+                          LAPACK_Z_SELECT2 selctg, lapack_int n,
+                          lapack_complex_double* a, lapack_int lda,
+                          lapack_complex_double* b, lapack_int ldb,
+                          lapack_int* sdim, lapack_complex_double* alpha,
+                          lapack_complex_double* beta,
+                          lapack_complex_double* vsl, lapack_int ldvsl,
+                          lapack_complex_double* vsr, lapack_int ldvsr );
+
+lapack_int LAPACKE_sggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_S_SELECT3 selctg, char sense,
+                           lapack_int n, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, lapack_int* sdim, float* alphar,
+                           float* alphai, float* beta, float* vsl,
+                           lapack_int ldvsl, float* vsr, lapack_int ldvsr,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_dggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_D_SELECT3 selctg, char sense,
+                           lapack_int n, double* a, lapack_int lda, double* b,
+                           lapack_int ldb, lapack_int* sdim, double* alphar,
+                           double* alphai, double* beta, double* vsl,
+                           lapack_int ldvsl, double* vsr, lapack_int ldvsr,
+                           double* rconde, double* rcondv );
+lapack_int LAPACKE_cggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_C_SELECT2 selctg, char sense,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_int* sdim,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta,
+                           lapack_complex_float* vsl, lapack_int ldvsl,
+                           lapack_complex_float* vsr, lapack_int ldvsr,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_zggesx( int matrix_order, char jobvsl, char jobvsr,
+                           char sort, LAPACK_Z_SELECT2 selctg, char sense,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_int* sdim,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* vsl, lapack_int ldvsl,
+                           lapack_complex_double* vsr, lapack_int ldvsr,
+                           double* rconde, double* rcondv );
+
+lapack_int LAPACKE_sggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, float* a, lapack_int lda, float* b,
+                          lapack_int ldb, float* alphar, float* alphai,
+                          float* beta, float* vl, lapack_int ldvl, float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_dggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, double* a, lapack_int lda, double* b,
+                          lapack_int ldb, double* alphar, double* alphai,
+                          double* beta, double* vl, lapack_int ldvl, double* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_cggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_float* a, lapack_int lda,
+                          lapack_complex_float* b, lapack_int ldb,
+                          lapack_complex_float* alpha,
+                          lapack_complex_float* beta, lapack_complex_float* vl,
+                          lapack_int ldvl, lapack_complex_float* vr,
+                          lapack_int ldvr );
+lapack_int LAPACKE_zggev( int matrix_order, char jobvl, char jobvr,
+                          lapack_int n, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* b,
+                          lapack_int ldb, lapack_complex_double* alpha,
+                          lapack_complex_double* beta,
+                          lapack_complex_double* vl, lapack_int ldvl,
+                          lapack_complex_double* vr, lapack_int ldvr );
+
+lapack_int LAPACKE_sggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           float* alphar, float* alphai, float* beta, float* vl,
+                           lapack_int ldvl, float* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, float* lscale,
+                           float* rscale, float* abnrm, float* bbnrm,
+                           float* rconde, float* rcondv );
+lapack_int LAPACKE_dggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double* alphar, double* alphai, double* beta,
+                           double* vl, lapack_int ldvl, double* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           double* lscale, double* rscale, double* abnrm,
+                           double* bbnrm, double* rconde, double* rcondv );
+lapack_int LAPACKE_cggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta, lapack_complex_float* vl,
+                           lapack_int ldvl, lapack_complex_float* vr,
+                           lapack_int ldvr, lapack_int* ilo, lapack_int* ihi,
+                           float* lscale, float* rscale, float* abnrm,
+                           float* bbnrm, float* rconde, float* rcondv );
+lapack_int LAPACKE_zggevx( int matrix_order, char balanc, char jobvl,
+                           char jobvr, char sense, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* vl, lapack_int ldvl,
+                           lapack_complex_double* vr, lapack_int ldvr,
+                           lapack_int* ilo, lapack_int* ihi, double* lscale,
+                           double* rscale, double* abnrm, double* bbnrm,
+                           double* rconde, double* rcondv );
+
+lapack_int LAPACKE_sggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* d, float* x, float* y );
+lapack_int LAPACKE_dggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, double* a, lapack_int lda, double* b,
+                           lapack_int ldb, double* d, double* x, double* y );
+lapack_int LAPACKE_cggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* d,
+                           lapack_complex_float* x, lapack_complex_float* y );
+lapack_int LAPACKE_zggglm( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* d,
+                           lapack_complex_double* x, lapack_complex_double* y );
+
+lapack_int LAPACKE_sgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           float* a, lapack_int lda, float* b, lapack_int ldb,
+                           float* q, lapack_int ldq, float* z, lapack_int ldz );
+lapack_int LAPACKE_dgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           double* a, lapack_int lda, double* b, lapack_int ldb,
+                           double* q, lapack_int ldq, double* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_cgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zgghrd( int matrix_order, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, float* a, lapack_int lda, float* b,
+                           lapack_int ldb, float* c, float* d, float* x );
+lapack_int LAPACKE_dgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, double* a, lapack_int lda, double* b,
+                           lapack_int ldb, double* c, double* d, double* x );
+lapack_int LAPACKE_cgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* c,
+                           lapack_complex_float* d, lapack_complex_float* x );
+lapack_int LAPACKE_zgglse( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int p, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* c,
+                           lapack_complex_double* d, lapack_complex_double* x );
+
+lapack_int LAPACKE_sggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, float* a, lapack_int lda, float* taua,
+                           float* b, lapack_int ldb, float* taub );
+lapack_int LAPACKE_dggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, double* a, lapack_int lda,
+                           double* taua, double* b, lapack_int ldb,
+                           double* taub );
+lapack_int LAPACKE_cggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* taua,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* taub );
+lapack_int LAPACKE_zggqrf( int matrix_order, lapack_int n, lapack_int m,
+                           lapack_int p, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* taua,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* taub );
+
+lapack_int LAPACKE_sggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, float* a, lapack_int lda, float* taua,
+                           float* b, lapack_int ldb, float* taub );
+lapack_int LAPACKE_dggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, double* a, lapack_int lda,
+                           double* taua, double* b, lapack_int ldb,
+                           double* taub );
+lapack_int LAPACKE_cggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* taua,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* taub );
+lapack_int LAPACKE_zggrqf( int matrix_order, lapack_int m, lapack_int p,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* taua,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* taub );
+
+lapack_int LAPACKE_sggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           float* alpha, float* beta, float* u, lapack_int ldu,
+                           float* v, lapack_int ldv, float* q, lapack_int ldq,
+                           lapack_int* iwork );
+lapack_int LAPACKE_dggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double* alpha, double* beta, double* u,
+                           lapack_int ldu, double* v, lapack_int ldv, double* q,
+                           lapack_int ldq, lapack_int* iwork );
+lapack_int LAPACKE_cggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           float* alpha, float* beta, lapack_complex_float* u,
+                           lapack_int ldu, lapack_complex_float* v,
+                           lapack_int ldv, lapack_complex_float* q,
+                           lapack_int ldq, lapack_int* iwork );
+lapack_int LAPACKE_zggsvd( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int n, lapack_int p,
+                           lapack_int* k, lapack_int* l,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           double* alpha, double* beta,
+                           lapack_complex_double* u, lapack_int ldu,
+                           lapack_complex_double* v, lapack_int ldv,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_int* iwork );
+
+lapack_int LAPACKE_sggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb, float tola,
+                           float tolb, lapack_int* k, lapack_int* l, float* u,
+                           lapack_int ldu, float* v, lapack_int ldv, float* q,
+                           lapack_int ldq );
+lapack_int LAPACKE_dggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double tola, double tolb, lapack_int* k,
+                           lapack_int* l, double* u, lapack_int ldu, double* v,
+                           lapack_int ldv, double* q, lapack_int ldq );
+lapack_int LAPACKE_cggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb, float tola,
+                           float tolb, lapack_int* k, lapack_int* l,
+                           lapack_complex_float* u, lapack_int ldu,
+                           lapack_complex_float* v, lapack_int ldv,
+                           lapack_complex_float* q, lapack_int ldq );
+lapack_int LAPACKE_zggsvp( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           double tola, double tolb, lapack_int* k,
+                           lapack_int* l, lapack_complex_double* u,
+                           lapack_int ldu, lapack_complex_double* v,
+                           lapack_int ldv, lapack_complex_double* q,
+                           lapack_int ldq );
+
+lapack_int LAPACKE_sgtcon( char norm, lapack_int n, const float* dl,
+                           const float* d, const float* du, const float* du2,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_dgtcon( char norm, lapack_int n, const double* dl,
+                           const double* d, const double* du, const double* du2,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_cgtcon( char norm, lapack_int n,
+                           const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           const lapack_complex_float* du2,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zgtcon( char norm, lapack_int n,
+                           const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           const lapack_complex_double* du2,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* dl, const float* d,
+                           const float* du, const float* dlf, const float* df,
+                           const float* duf, const float* du2,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* dl, const double* d,
+                           const double* du, const double* dlf,
+                           const double* df, const double* duf,
+                           const double* du2, const lapack_int* ipiv,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           const lapack_complex_float* dlf,
+                           const lapack_complex_float* df,
+                           const lapack_complex_float* duf,
+                           const lapack_complex_float* du2,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgtrfs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           const lapack_complex_double* dlf,
+                           const lapack_complex_double* df,
+                           const lapack_complex_double* duf,
+                           const lapack_complex_double* du2,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* dl, float* d, float* du, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* dl, double* d, double* du, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_float* dl, lapack_complex_float* d,
+                          lapack_complex_float* du, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zgtsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          lapack_complex_double* dl, lapack_complex_double* d,
+                          lapack_complex_double* du, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_sgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, const float* dl,
+                           const float* d, const float* du, float* dlf,
+                           float* df, float* duf, float* du2, lapack_int* ipiv,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs, const double* dl,
+                           const double* d, const double* du, double* dlf,
+                           double* df, double* duf, double* du2,
+                           lapack_int* ipiv, const double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           lapack_complex_float* dlf, lapack_complex_float* df,
+                           lapack_complex_float* duf, lapack_complex_float* du2,
+                           lapack_int* ipiv, const lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zgtsvx( int matrix_order, char fact, char trans,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           lapack_complex_double* dlf,
+                           lapack_complex_double* df,
+                           lapack_complex_double* duf,
+                           lapack_complex_double* du2, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_sgttrf( lapack_int n, float* dl, float* d, float* du,
+                           float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_dgttrf( lapack_int n, double* dl, double* d, double* du,
+                           double* du2, lapack_int* ipiv );
+lapack_int LAPACKE_cgttrf( lapack_int n, lapack_complex_float* dl,
+                           lapack_complex_float* d, lapack_complex_float* du,
+                           lapack_complex_float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_zgttrf( lapack_int n, lapack_complex_double* dl,
+                           lapack_complex_double* d, lapack_complex_double* du,
+                           lapack_complex_double* du2, lapack_int* ipiv );
+
+lapack_int LAPACKE_sgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const float* dl, const float* d,
+                           const float* du, const float* du2,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const double* dl, const double* d,
+                           const double* du, const double* du2,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_cgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* dl,
+                           const lapack_complex_float* d,
+                           const lapack_complex_float* du,
+                           const lapack_complex_float* du2,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zgttrs( int matrix_order, char trans, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* dl,
+                           const lapack_complex_double* d,
+                           const lapack_complex_double* du,
+                           const lapack_complex_double* du2,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_chbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, lapack_complex_float* ab,
+                          lapack_int ldab, float* w, lapack_complex_float* z,
+                          lapack_int ldz );
+lapack_int LAPACKE_zhbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, lapack_complex_double* ab,
+                          lapack_int ldab, double* w, lapack_complex_double* z,
+                          lapack_int ldz );
+
+lapack_int LAPACKE_chbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_float* ab,
+                           lapack_int ldab, float* w, lapack_complex_float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_zhbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_double* ab,
+                           lapack_int ldab, double* w, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_chbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* q, lapack_int ldq, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* bb, lapack_int ldbb,
+                           lapack_complex_float* x, lapack_int ldx );
+lapack_int LAPACKE_zhbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* bb, lapack_int ldbb,
+                           lapack_complex_double* x, lapack_int ldx );
+
+lapack_int LAPACKE_chbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb,
+                          lapack_complex_float* ab, lapack_int ldab,
+                          lapack_complex_float* bb, lapack_int ldbb, float* w,
+                          lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb,
+                          lapack_complex_double* ab, lapack_int ldab,
+                          lapack_complex_double* bb, lapack_int ldbb, double* w,
+                          lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* bb, lapack_int ldbb, float* w,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* bb, lapack_int ldbb,
+                           double* w, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_chbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* bb, lapack_int ldbb,
+                           lapack_complex_float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* bb, lapack_int ldbb,
+                           lapack_complex_double* q, lapack_int ldq, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_float* ab,
+                           lapack_int ldab, float* d, float* e,
+                           lapack_complex_float* q, lapack_int ldq );
+lapack_int LAPACKE_zhbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_double* ab,
+                           lapack_int ldab, double* d, double* e,
+                           lapack_complex_double* q, lapack_int ldq );
+
+lapack_int LAPACKE_checon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zhecon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_cheequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* s, float* scond, float* amax );
+lapack_int LAPACKE_zheequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_cheev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_zheev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_double* a, lapack_int lda, double* w );
+
+lapack_int LAPACKE_cheevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_zheevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           double* w );
+
+lapack_int LAPACKE_cheevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float vl, float vu, lapack_int il,
+                           lapack_int iu, float abstol, lapack_int* m, float* w,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_zheevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double vl, double vu, lapack_int il,
+                           lapack_int iu, double abstol, lapack_int* m,
+                           double* w, lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* isuppz );
+
+lapack_int LAPACKE_cheevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, float vl, float vu, lapack_int il,
+                           lapack_int iu, float abstol, lapack_int* m, float* w,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_zheevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, double vl, double vu, lapack_int il,
+                           lapack_int iu, double abstol, lapack_int* m,
+                           double* w, lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chegst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zhegst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_chegv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_float* a,
+                          lapack_int lda, lapack_complex_float* b,
+                          lapack_int ldb, float* w );
+lapack_int LAPACKE_zhegv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* b,
+                          lapack_int ldb, double* w );
+
+lapack_int LAPACKE_chegvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float* w );
+lapack_int LAPACKE_zhegvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double* w );
+
+lapack_int LAPACKE_chegvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhegvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_cherfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zherfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_cherfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* s,
+                            const lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zherfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* s,
+                            const lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_chesv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zhesv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chesvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zhesvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_chesvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* s,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zhesvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* s,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_chetrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda, float* d,
+                           float* e, lapack_complex_float* tau );
+lapack_int LAPACKE_zhetrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda, double* d,
+                           double* e, lapack_complex_double* tau );
+
+lapack_int LAPACKE_chetrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zhetrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_chetri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_zhetri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv );
+
+lapack_int LAPACKE_chetrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zhetrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, float alpha,
+                          const lapack_complex_float* a, lapack_int lda,
+                          float beta, lapack_complex_float* c );
+lapack_int LAPACKE_zhfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, double alpha,
+                          const lapack_complex_double* a, lapack_int lda,
+                          double beta, lapack_complex_double* c );
+
+lapack_int LAPACKE_shgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           float* h, lapack_int ldh, float* t, lapack_int ldt,
+                           float* alphar, float* alphai, float* beta, float* q,
+                           lapack_int ldq, float* z, lapack_int ldz );
+lapack_int LAPACKE_dhgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           double* h, lapack_int ldh, double* t, lapack_int ldt,
+                           double* alphar, double* alphai, double* beta,
+                           double* q, lapack_int ldq, double* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_chgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_float* h, lapack_int ldh,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta, lapack_complex_float* q,
+                           lapack_int ldq, lapack_complex_float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_zhgeqz( int matrix_order, char job, char compq, char compz,
+                           lapack_int n, lapack_int ilo, lapack_int ihi,
+                           lapack_complex_double* h, lapack_int ldh,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zhpcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_chpev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_float* ap, float* w,
+                          lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_complex_double* ap, double* w,
+                          lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_float* ap, float* w,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_complex_double* ap, double* w,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_float* ap, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_zhpevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_complex_double* ap, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chpgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_float* ap,
+                           const lapack_complex_float* bp );
+lapack_int LAPACKE_zhpgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, lapack_complex_double* ap,
+                           const lapack_complex_double* bp );
+
+lapack_int LAPACKE_chpgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_float* ap,
+                          lapack_complex_float* bp, float* w,
+                          lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, lapack_complex_double* ap,
+                          lapack_complex_double* bp, double* w,
+                          lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_float* ap,
+                           lapack_complex_float* bp, float* w,
+                           lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zhpgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, lapack_complex_double* ap,
+                           lapack_complex_double* bp, double* w,
+                           lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_chpgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_float* ap, lapack_complex_float* bp,
+                           float vl, float vu, lapack_int il, lapack_int iu,
+                           float abstol, lapack_int* m, float* w,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_zhpgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n,
+                           lapack_complex_double* ap, lapack_complex_double* bp,
+                           double vl, double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_chprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_complex_float* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zhprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_complex_double* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_chpsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* ap,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zhpsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* ap,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_chpsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           lapack_complex_float* afp, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zhpsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           lapack_complex_double* afp, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_chptrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, float* d, float* e,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zhptrd( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, double* d, double* e,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_chptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zhptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_chptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, const lapack_int* ipiv );
+lapack_int LAPACKE_zhptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, const lapack_int* ipiv );
+
+lapack_int LAPACKE_chptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zhptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_shsein( int matrix_order, char job, char eigsrc, char initv,
+                           lapack_logical* select, lapack_int n, const float* h,
+                           lapack_int ldh, float* wr, const float* wi,
+                           float* vl, lapack_int ldvl, float* vr,
+                           lapack_int ldvr, lapack_int mm, lapack_int* m,
+                           lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_dhsein( int matrix_order, char job, char eigsrc, char initv,
+                           lapack_logical* select, lapack_int n,
+                           const double* h, lapack_int ldh, double* wr,
+                           const double* wi, double* vl, lapack_int ldvl,
+                           double* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m, lapack_int* ifaill,
+                           lapack_int* ifailr );
+lapack_int LAPACKE_chsein( int matrix_order, char job, char eigsrc, char initv,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* h, lapack_int ldh,
+                           lapack_complex_float* w, lapack_complex_float* vl,
+                           lapack_int ldvl, lapack_complex_float* vr,
+                           lapack_int ldvr, lapack_int mm, lapack_int* m,
+                           lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_zhsein( int matrix_order, char job, char eigsrc, char initv,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* h, lapack_int ldh,
+                           lapack_complex_double* w, lapack_complex_double* vl,
+                           lapack_int ldvl, lapack_complex_double* vr,
+                           lapack_int ldvr, lapack_int mm, lapack_int* m,
+                           lapack_int* ifaill, lapack_int* ifailr );
+
+lapack_int LAPACKE_shseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, float* h,
+                           lapack_int ldh, float* wr, float* wi, float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_dhseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi, double* h,
+                           lapack_int ldh, double* wr, double* wi, double* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_chseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi,
+                           lapack_complex_float* h, lapack_int ldh,
+                           lapack_complex_float* w, lapack_complex_float* z,
+                           lapack_int ldz );
+lapack_int LAPACKE_zhseqr( int matrix_order, char job, char compz, lapack_int n,
+                           lapack_int ilo, lapack_int ihi,
+                           lapack_complex_double* h, lapack_int ldh,
+                           lapack_complex_double* w, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_clacgv( lapack_int n, lapack_complex_float* x,
+                           lapack_int incx );
+lapack_int LAPACKE_zlacgv( lapack_int n, lapack_complex_double* x,
+                           lapack_int incx );
+
+lapack_int LAPACKE_slacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const float* a, lapack_int lda, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dlacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const double* a, lapack_int lda, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_clacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zlacpy( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_zlag2c( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_float* sa, lapack_int ldsa );
+
+lapack_int LAPACKE_slag2d( int matrix_order, lapack_int m, lapack_int n,
+                           const float* sa, lapack_int ldsa, double* a,
+                           lapack_int lda );
+
+lapack_int LAPACKE_dlag2s( int matrix_order, lapack_int m, lapack_int n,
+                           const double* a, lapack_int lda, float* sa,
+                           lapack_int ldsa );
+
+lapack_int LAPACKE_clag2z( int matrix_order, lapack_int m, lapack_int n,
+                           const lapack_complex_float* sa, lapack_int ldsa,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* d,
+                           float* a, lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_dlagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* d,
+                           double* a, lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_clagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const float* d,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* iseed );
+lapack_int LAPACKE_zlagge( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int kl, lapack_int ku, const double* d,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* iseed );
+
+float LAPACKE_slamch( char cmach );
+double LAPACKE_dlamch( char cmach );
+
+float LAPACKE_slange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const float* a, lapack_int lda );
+double LAPACKE_dlange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const double* a, lapack_int lda );
+float LAPACKE_clange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda );
+double LAPACKE_zlange( int matrix_order, char norm, lapack_int m,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda );
+
+float LAPACKE_clanhe( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda );
+double LAPACKE_zlanhe( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda );
+
+float LAPACKE_slansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const float* a, lapack_int lda );
+double LAPACKE_dlansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const double* a, lapack_int lda );
+float LAPACKE_clansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda );
+double LAPACKE_zlansy( int matrix_order, char norm, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda );
+
+float LAPACKE_slantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const float* a,
+                           lapack_int lda );
+double LAPACKE_dlantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const double* a,
+                           lapack_int lda );
+float LAPACKE_clantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda );
+double LAPACKE_zlantr( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int m, lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda );
+
+
+lapack_int LAPACKE_slarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const float* v, lapack_int ldv,
+                           const float* t, lapack_int ldt, float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_dlarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const double* v, lapack_int ldv,
+                           const double* t, lapack_int ldt, double* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_clarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const lapack_complex_float* v,
+                           lapack_int ldv, const lapack_complex_float* t,
+                           lapack_int ldt, lapack_complex_float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_zlarfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, const lapack_complex_double* v,
+                           lapack_int ldv, const lapack_complex_double* t,
+                           lapack_int ldt, lapack_complex_double* c,
+                           lapack_int ldc );
+
+lapack_int LAPACKE_slarfg( lapack_int n, float* alpha, float* x,
+                           lapack_int incx, float* tau );
+lapack_int LAPACKE_dlarfg( lapack_int n, double* alpha, double* x,
+                           lapack_int incx, double* tau );
+lapack_int LAPACKE_clarfg( lapack_int n, lapack_complex_float* alpha,
+                           lapack_complex_float* x, lapack_int incx,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_zlarfg( lapack_int n, lapack_complex_double* alpha,
+                           lapack_complex_double* x, lapack_int incx,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_slarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k, const float* v,
+                           lapack_int ldv, const float* tau, float* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_dlarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k, const double* v,
+                           lapack_int ldv, const double* tau, double* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_clarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k,
+                           const lapack_complex_float* v, lapack_int ldv,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zlarft( int matrix_order, char direct, char storev,
+                           lapack_int n, lapack_int k,
+                           const lapack_complex_double* v, lapack_int ldv,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_slarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const float* v, float tau, float* c,
+                           lapack_int ldc, float* work );
+lapack_int LAPACKE_dlarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const double* v, double tau, double* c,
+                           lapack_int ldc, double* work );
+lapack_int LAPACKE_clarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const lapack_complex_float* v,
+                           lapack_complex_float tau, lapack_complex_float* c,
+                           lapack_int ldc, lapack_complex_float* work );
+lapack_int LAPACKE_zlarfx( int matrix_order, char side, lapack_int m,
+                           lapack_int n, const lapack_complex_double* v,
+                           lapack_complex_double tau, lapack_complex_double* c,
+                           lapack_int ldc, lapack_complex_double* work );
+
+lapack_int LAPACKE_slarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           float* x );
+lapack_int LAPACKE_dlarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           double* x );
+lapack_int LAPACKE_clarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           lapack_complex_float* x );
+lapack_int LAPACKE_zlarnv( lapack_int idist, lapack_int* iseed, lapack_int n,
+                           lapack_complex_double* x );
+
+lapack_int LAPACKE_slaset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, float alpha, float beta, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dlaset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, double alpha, double beta, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_claset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, lapack_complex_float alpha,
+                           lapack_complex_float beta, lapack_complex_float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_zlaset( int matrix_order, char uplo, lapack_int m,
+                           lapack_int n, lapack_complex_double alpha,
+                           lapack_complex_double beta, lapack_complex_double* a,
+                           lapack_int lda );
+
+lapack_int LAPACKE_slasrt( char id, lapack_int n, float* d );
+lapack_int LAPACKE_dlasrt( char id, lapack_int n, double* d );
+
+lapack_int LAPACKE_slaswp( int matrix_order, lapack_int n, float* a,
+                           lapack_int lda, lapack_int k1, lapack_int k2,
+                           const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_dlaswp( int matrix_order, lapack_int n, double* a,
+                           lapack_int lda, lapack_int k1, lapack_int k2,
+                           const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_claswp( int matrix_order, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int k1, lapack_int k2, const lapack_int* ipiv,
+                           lapack_int incx );
+lapack_int LAPACKE_zlaswp( int matrix_order, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int k1, lapack_int k2, const lapack_int* ipiv,
+                           lapack_int incx );
+
+lapack_int LAPACKE_slatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, float* d,
+                           lapack_int mode, float cond, float dmax,
+                           lapack_int kl, lapack_int ku, char pack, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dlatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, double* d,
+                           lapack_int mode, double cond, double dmax,
+                           lapack_int kl, lapack_int ku, char pack, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_clatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, float* d,
+                           lapack_int mode, float cond, float dmax,
+                           lapack_int kl, lapack_int ku, char pack,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlatms( int matrix_order, lapack_int m, lapack_int n,
+                           char dist, lapack_int* iseed, char sym, double* d,
+                           lapack_int mode, double cond, double dmax,
+                           lapack_int kl, lapack_int ku, char pack,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slauum( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dlauum( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_clauum( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlauum( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_sopgtr( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, const float* tau, float* q,
+                           lapack_int ldq );
+lapack_int LAPACKE_dopgtr( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, const double* tau, double* q,
+                           lapack_int ldq );
+
+lapack_int LAPACKE_sopmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const float* ap,
+                           const float* tau, float* c, lapack_int ldc );
+lapack_int LAPACKE_dopmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const double* ap,
+                           const double* tau, double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sorgbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, double* a,
+                           lapack_int lda, const double* tau );
+
+lapack_int LAPACKE_sorghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, float* a, lapack_int lda,
+                           const float* tau );
+lapack_int LAPACKE_dorgrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, double* a, lapack_int lda,
+                           const double* tau );
+
+lapack_int LAPACKE_sorgtr( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, const float* tau );
+lapack_int LAPACKE_dorgtr( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, const double* tau );
+
+lapack_int LAPACKE_sormbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const float* a, lapack_int lda,
+                           const float* tau, float* c, lapack_int ldc );
+lapack_int LAPACKE_dormhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const double* a, lapack_int lda,
+                           const double* tau, double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const float* a, lapack_int lda, const float* tau,
+                           float* c, lapack_int ldc );
+lapack_int LAPACKE_dormrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const double* a, lapack_int lda, const double* tau,
+                           double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const float* a, lapack_int lda,
+                           const float* tau, float* c, lapack_int ldc );
+lapack_int LAPACKE_dormrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const double* a, lapack_int lda,
+                           const double* tau, double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sormtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const float* a,
+                           lapack_int lda, const float* tau, float* c,
+                           lapack_int ldc );
+lapack_int LAPACKE_dormtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n, const double* a,
+                           lapack_int lda, const double* tau, double* c,
+                           lapack_int ldc );
+
+lapack_int LAPACKE_spbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const float* ab, lapack_int ldab,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_dpbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const double* ab, lapack_int ldab,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cpbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_float* ab,
+                           lapack_int ldab, float anorm, float* rcond );
+lapack_int LAPACKE_zpbcon( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_double* ab,
+                           lapack_int ldab, double anorm, double* rcond );
+
+lapack_int LAPACKE_spbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const float* ab, lapack_int ldab,
+                           float* s, float* scond, float* amax );
+lapack_int LAPACKE_dpbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const double* ab, lapack_int ldab,
+                           double* s, double* scond, double* amax );
+lapack_int LAPACKE_cpbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_float* ab,
+                           lapack_int ldab, float* s, float* scond,
+                           float* amax );
+lapack_int LAPACKE_zpbequ( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, const lapack_complex_double* ab,
+                           lapack_int ldab, double* s, double* scond,
+                           double* amax );
+
+lapack_int LAPACKE_spbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const float* ab,
+                           lapack_int ldab, const float* afb, lapack_int ldafb,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dpbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const double* ab,
+                           lapack_int ldab, const double* afb, lapack_int ldafb,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cpbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* afb, lapack_int ldafb,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zpbrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* afb, lapack_int ldafb,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_spbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, float* bb, lapack_int ldbb );
+lapack_int LAPACKE_dpbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, double* bb, lapack_int ldbb );
+lapack_int LAPACKE_cpbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, lapack_complex_float* bb,
+                           lapack_int ldbb );
+lapack_int LAPACKE_zpbstf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kb, lapack_complex_double* bb,
+                           lapack_int ldbb );
+
+lapack_int LAPACKE_spbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs, float* ab,
+                          lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs, double* ab,
+                          lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs,
+                          lapack_complex_float* ab, lapack_int ldab,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int kd, lapack_int nrhs,
+                          lapack_complex_double* ab, lapack_int ldab,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, float* ab,
+                           lapack_int ldab, float* afb, lapack_int ldafb,
+                           char* equed, float* s, float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dpbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, double* ab,
+                           lapack_int ldab, double* afb, lapack_int ldafb,
+                           char* equed, double* s, double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cpbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* afb, lapack_int ldafb,
+                           char* equed, float* s, lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zpbsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* afb, lapack_int ldafb,
+                           char* equed, double* s, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, double* rcond, double* ferr,
+                           double* berr );
+
+lapack_int LAPACKE_spbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, float* ab, lapack_int ldab );
+lapack_int LAPACKE_dpbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, double* ab, lapack_int ldab );
+lapack_int LAPACKE_cpbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_float* ab,
+                           lapack_int ldab );
+lapack_int LAPACKE_zpbtrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_complex_double* ab,
+                           lapack_int ldab );
+
+lapack_int LAPACKE_spbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const float* ab,
+                           lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs, const double* ab,
+                           lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbtrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, float* a );
+lapack_int LAPACKE_dpftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, double* a );
+lapack_int LAPACKE_cpftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftrf( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, float* a );
+lapack_int LAPACKE_dpftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, double* a );
+lapack_int LAPACKE_cpftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftri( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs, const float* a,
+                           float* b, lapack_int ldb );
+lapack_int LAPACKE_dpftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs, const double* a,
+                           double* b, lapack_int ldb );
+lapack_int LAPACKE_cpftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* a,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpftrs( int matrix_order, char transr, char uplo,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* a,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spocon( int matrix_order, char uplo, lapack_int n,
+                           const float* a, lapack_int lda, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dpocon( int matrix_order, char uplo, lapack_int n,
+                           const double* a, lapack_int lda, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_cpocon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_zpocon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double anorm, double* rcond );
+
+lapack_int LAPACKE_spoequ( int matrix_order, lapack_int n, const float* a,
+                           lapack_int lda, float* s, float* scond,
+                           float* amax );
+lapack_int LAPACKE_dpoequ( int matrix_order, lapack_int n, const double* a,
+                           lapack_int lda, double* s, double* scond,
+                           double* amax );
+lapack_int LAPACKE_cpoequ( int matrix_order, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequ( int matrix_order, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_spoequb( int matrix_order, lapack_int n, const float* a,
+                            lapack_int lda, float* s, float* scond,
+                            float* amax );
+lapack_int LAPACKE_dpoequb( int matrix_order, lapack_int n, const double* a,
+                            lapack_int lda, double* s, double* scond,
+                            double* amax );
+lapack_int LAPACKE_cpoequb( int matrix_order, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequb( int matrix_order, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_sporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const float* af, lapack_int ldaf, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const double* af, lapack_int ldaf, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_complex_float* b,
+                           lapack_int ldb, lapack_complex_float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_zporfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+
+lapack_int LAPACKE_sporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const float* a,
+                            lapack_int lda, const float* af, lapack_int ldaf,
+                            const float* s, const float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const double* a,
+                            lapack_int lda, const double* af, lapack_int ldaf,
+                            const double* s, const double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const float* s, const lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zporfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const double* s, const lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_sposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* a, lapack_int lda, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* a, lapack_int lda, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* a,
+                          lapack_int lda, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zposv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* a,
+                          lapack_int lda, lapack_complex_double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dsposv( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           lapack_int* iter );
+lapack_int LAPACKE_zcposv( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, lapack_complex_double* x,
+                           lapack_int ldx, lapack_int* iter );
+
+lapack_int LAPACKE_sposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, float* a, lapack_int lda, float* af,
+                           lapack_int ldaf, char* equed, float* s, float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_dposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, double* a, lapack_int lda,
+                           double* af, lapack_int ldaf, char* equed, double* s,
+                           double* b, lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+lapack_int LAPACKE_cposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* af,
+                           lapack_int ldaf, char* equed, float* s,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zposvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* af,
+                           lapack_int ldaf, char* equed, double* s,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_sposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, float* a,
+                            lapack_int lda, float* af, lapack_int ldaf,
+                            char* equed, float* s, float* b, lapack_int ldb,
+                            float* x, lapack_int ldx, float* rcond,
+                            float* rpvgrw, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_dposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, double* a,
+                            lapack_int lda, double* af, lapack_int ldaf,
+                            char* equed, double* s, double* b, lapack_int ldb,
+                            double* x, lapack_int ldx, double* rcond,
+                            double* rpvgrw, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+lapack_int LAPACKE_cposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            char* equed, float* s, lapack_complex_float* b,
+                            lapack_int ldb, lapack_complex_float* x,
+                            lapack_int ldx, float* rcond, float* rpvgrw,
+                            float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zposvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            char* equed, double* s, lapack_complex_double* b,
+                            lapack_int ldb, lapack_complex_double* x,
+                            lapack_int ldx, double* rcond, double* rpvgrw,
+                            double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_spotrf( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dpotrf( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_cpotrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotri( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dpotri( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_cpotri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           float* b, lapack_int ldb );
+lapack_int LAPACKE_dpotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           double* b, lapack_int ldb );
+lapack_int LAPACKE_cpotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zpotrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_sppcon( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, float anorm, float* rcond );
+lapack_int LAPACKE_dppcon( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, double anorm, double* rcond );
+lapack_int LAPACKE_cppcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_zppcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sppequ( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, float* s, float* scond,
+                           float* amax );
+lapack_int LAPACKE_dppequ( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, double* s, double* scond,
+                           double* amax );
+lapack_int LAPACKE_cppequ( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap, float* s,
+                           float* scond, float* amax );
+lapack_int LAPACKE_zppequ( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap, double* s,
+                           double* scond, double* amax );
+
+lapack_int LAPACKE_spprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, const float* afp,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dpprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, const double* afp,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_cpprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_complex_float* afp,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zpprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_complex_double* afp,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* ap, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* ap, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_cppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* ap,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zppsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* ap,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, float* ap, float* afp, char* equed,
+                           float* s, float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, double* ap, double* afp,
+                           char* equed, double* s, double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_float* ap,
+                           lapack_complex_float* afp, char* equed, float* s,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zppsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, lapack_complex_double* ap,
+                           lapack_complex_double* afp, char* equed, double* s,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_spptrf( int matrix_order, char uplo, lapack_int n,
+                           float* ap );
+lapack_int LAPACKE_dpptrf( int matrix_order, char uplo, lapack_int n,
+                           double* ap );
+lapack_int LAPACKE_cpptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_zpptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptri( int matrix_order, char uplo, lapack_int n,
+                           float* ap );
+lapack_int LAPACKE_dpptri( int matrix_order, char uplo, lapack_int n,
+                           double* ap );
+lapack_int LAPACKE_cpptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_zpptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dpptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_cpptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spstrf( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, lapack_int* piv, lapack_int* rank,
+                           float tol );
+lapack_int LAPACKE_dpstrf( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, lapack_int* piv, lapack_int* rank,
+                           double tol );
+lapack_int LAPACKE_cpstrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* piv, lapack_int* rank, float tol );
+lapack_int LAPACKE_zpstrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* piv, lapack_int* rank, double tol );
+
+lapack_int LAPACKE_sptcon( lapack_int n, const float* d, const float* e,
+                           float anorm, float* rcond );
+lapack_int LAPACKE_dptcon( lapack_int n, const double* d, const double* e,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cptcon( lapack_int n, const float* d,
+                           const lapack_complex_float* e, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_zptcon( lapack_int n, const double* d,
+                           const lapack_complex_double* e, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_spteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dpteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, double* z, lapack_int ldz );
+lapack_int LAPACKE_cpteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zpteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_sptrfs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const float* d, const float* e, const float* df,
+                           const float* ef, const float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dptrfs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const double* d, const double* e, const double* df,
+                           const double* ef, const double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* ferr,
+                           double* berr );
+lapack_int LAPACKE_cptrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* d,
+                           const lapack_complex_float* e, const float* df,
+                           const lapack_complex_float* ef,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zptrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* d,
+                           const lapack_complex_double* e, const double* df,
+                           const lapack_complex_double* ef,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* d, float* e, float* b, lapack_int ldb );
+lapack_int LAPACKE_dptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* d, double* e, double* b, lapack_int ldb );
+lapack_int LAPACKE_cptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          float* d, lapack_complex_float* e,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zptsv( int matrix_order, lapack_int n, lapack_int nrhs,
+                          double* d, lapack_complex_double* e,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const float* d, const float* e,
+                           float* df, float* ef, const float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const double* d, const double* e,
+                           double* df, double* ef, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+lapack_int LAPACKE_cptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const float* d,
+                           const lapack_complex_float* e, float* df,
+                           lapack_complex_float* ef,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zptsvx( int matrix_order, char fact, lapack_int n,
+                           lapack_int nrhs, const double* d,
+                           const lapack_complex_double* e, double* df,
+                           lapack_complex_double* ef,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_spttrf( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dpttrf( lapack_int n, double* d, double* e );
+lapack_int LAPACKE_cpttrf( lapack_int n, float* d, lapack_complex_float* e );
+lapack_int LAPACKE_zpttrf( lapack_int n, double* d, lapack_complex_double* e );
+
+lapack_int LAPACKE_spttrs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const float* d, const float* e, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dpttrs( int matrix_order, lapack_int n, lapack_int nrhs,
+                           const double* d, const double* e, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_cpttrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* d,
+                           const lapack_complex_float* e,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpttrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* d,
+                           const lapack_complex_double* e,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, float* ab, lapack_int ldab, float* w,
+                          float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int kd, double* ab, lapack_int ldab, double* w,
+                          double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, float* ab, lapack_int ldab, float* w,
+                           float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int kd, double* ab, lapack_int ldab,
+                           double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd, float* ab,
+                           lapack_int ldab, float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsbevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int kd, double* ab,
+                           lapack_int ldab, double* q, lapack_int ldq,
+                           double vl, double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, float* ab,
+                           lapack_int ldab, const float* bb, lapack_int ldbb,
+                           float* x, lapack_int ldx );
+lapack_int LAPACKE_dsbgst( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, double* ab,
+                           lapack_int ldab, const double* bb, lapack_int ldbb,
+                           double* x, lapack_int ldx );
+
+lapack_int LAPACKE_ssbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb, float* ab,
+                          lapack_int ldab, float* bb, lapack_int ldbb, float* w,
+                          float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbgv( int matrix_order, char jobz, char uplo, lapack_int n,
+                          lapack_int ka, lapack_int kb, double* ab,
+                          lapack_int ldab, double* bb, lapack_int ldbb,
+                          double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, float* ab,
+                           lapack_int ldab, float* bb, lapack_int ldbb,
+                           float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dsbgvd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           lapack_int ka, lapack_int kb, double* ab,
+                           lapack_int ldab, double* bb, lapack_int ldbb,
+                           double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_ssbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           float* ab, lapack_int ldab, float* bb,
+                           lapack_int ldbb, float* q, lapack_int ldq, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsbgvx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, lapack_int ka, lapack_int kb,
+                           double* ab, lapack_int ldab, double* bb,
+                           lapack_int ldbb, double* q, lapack_int ldq,
+                           double vl, double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, float* ab, lapack_int ldab, float* d,
+                           float* e, float* q, lapack_int ldq );
+lapack_int LAPACKE_dsbtrd( int matrix_order, char vect, char uplo, lapack_int n,
+                           lapack_int kd, double* ab, lapack_int ldab,
+                           double* d, double* e, double* q, lapack_int ldq );
+
+lapack_int LAPACKE_ssfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, float alpha,
+                          const float* a, lapack_int lda, float beta,
+                          float* c );
+lapack_int LAPACKE_dsfrk( int matrix_order, char transr, char uplo, char trans,
+                          lapack_int n, lapack_int k, double alpha,
+                          const double* a, lapack_int lda, double beta,
+                          double* c );
+
+lapack_int LAPACKE_sspcon( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, const lapack_int* ipiv, float anorm,
+                           float* rcond );
+lapack_int LAPACKE_dspcon( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, const lapack_int* ipiv,
+                           double anorm, double* rcond );
+lapack_int LAPACKE_cspcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zspcon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_sspev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          float* ap, float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          double* ap, double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           float* ap, float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           double* ap, double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, float* ap, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dspevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, double* ap, double vl, double vu,
+                           lapack_int il, lapack_int iu, double abstol,
+                           lapack_int* m, double* w, double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_sspgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, float* ap, const float* bp );
+lapack_int LAPACKE_dspgst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, double* ap, const double* bp );
+
+lapack_int LAPACKE_sspgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, float* ap, float* bp,
+                          float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspgv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, double* ap, double* bp,
+                          double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, float* ap, float* bp,
+                           float* w, float* z, lapack_int ldz );
+lapack_int LAPACKE_dspgvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, double* ap, double* bp,
+                           double* w, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sspgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, float* ap,
+                           float* bp, float vl, float vu, lapack_int il,
+                           lapack_int iu, float abstol, lapack_int* m, float* w,
+                           float* z, lapack_int ldz, lapack_int* ifail );
+lapack_int LAPACKE_dspgvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, double* ap,
+                           double* bp, double vl, double vu, lapack_int il,
+                           lapack_int iu, double abstol, lapack_int* m,
+                           double* w, double* z, lapack_int ldz,
+                           lapack_int* ifail );
+
+lapack_int LAPACKE_ssprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, const float* afp,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dsprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, const double* afp,
+                           const lapack_int* ipiv, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_csprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_complex_float* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zsprfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_complex_double* afp,
+                           const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_sspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* ap, lapack_int* ipiv,
+                          float* b, lapack_int ldb );
+lapack_int LAPACKE_dspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* ap, lapack_int* ipiv,
+                          double* b, lapack_int ldb );
+lapack_int LAPACKE_cspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* ap,
+                          lapack_int* ipiv, lapack_complex_float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_zspsv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* ap,
+                          lapack_int* ipiv, lapack_complex_double* b,
+                          lapack_int ldb );
+
+lapack_int LAPACKE_sspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap, float* afp,
+                           lapack_int* ipiv, const float* b, lapack_int ldb,
+                           float* x, lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap, double* afp,
+                           lapack_int* ipiv, const double* b, lapack_int ldb,
+                           double* x, lapack_int ldx, double* rcond,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_cspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           lapack_complex_float* afp, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zspsvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           lapack_complex_double* afp, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_ssptrd( int matrix_order, char uplo, lapack_int n, float* ap,
+                           float* d, float* e, float* tau );
+lapack_int LAPACKE_dsptrd( int matrix_order, char uplo, lapack_int n,
+                           double* ap, double* d, double* e, double* tau );
+
+lapack_int LAPACKE_ssptrf( int matrix_order, char uplo, lapack_int n, float* ap,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_dsptrf( int matrix_order, char uplo, lapack_int n,
+                           double* ap, lapack_int* ipiv );
+lapack_int LAPACKE_csptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zsptrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_ssptri( int matrix_order, char uplo, lapack_int n, float* ap,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_dsptri( int matrix_order, char uplo, lapack_int n,
+                           double* ap, const lapack_int* ipiv );
+lapack_int LAPACKE_csptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* ap, const lapack_int* ipiv );
+lapack_int LAPACKE_zsptri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* ap, const lapack_int* ipiv );
+
+lapack_int LAPACKE_ssptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* ap,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dsptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* ap,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_csptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* ap,
+                           const lapack_int* ipiv, lapack_complex_float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_zsptrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* ap,
+                           const lapack_int* ipiv, lapack_complex_double* b,
+                           lapack_int ldb );
+
+lapack_int LAPACKE_sstebz( char range, char order, lapack_int n, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           const float* d, const float* e, lapack_int* m,
+                           lapack_int* nsplit, float* w, lapack_int* iblock,
+                           lapack_int* isplit );
+lapack_int LAPACKE_dstebz( char range, char order, lapack_int n, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, const double* d, const double* e,
+                           lapack_int* m, lapack_int* nsplit, double* w,
+                           lapack_int* iblock, lapack_int* isplit );
+
+lapack_int LAPACKE_sstedc( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dstedc( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, double* z, lapack_int ldz );
+lapack_int LAPACKE_cstedc( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zstedc( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_sstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_dstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* isuppz );
+lapack_int LAPACKE_cstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* isuppz );
+lapack_int LAPACKE_zstegr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* isuppz );
+
+lapack_int LAPACKE_sstein( int matrix_order, lapack_int n, const float* d,
+                           const float* e, lapack_int m, const float* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           float* z, lapack_int ldz, lapack_int* ifailv );
+lapack_int LAPACKE_dstein( int matrix_order, lapack_int n, const double* d,
+                           const double* e, lapack_int m, const double* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           double* z, lapack_int ldz, lapack_int* ifailv );
+lapack_int LAPACKE_cstein( int matrix_order, lapack_int n, const float* d,
+                           const float* e, lapack_int m, const float* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int* ifailv );
+lapack_int LAPACKE_zstein( int matrix_order, lapack_int n, const double* d,
+                           const double* e, lapack_int m, const double* w,
+                           const lapack_int* iblock, const lapack_int* isplit,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* ifailv );
+
+lapack_int LAPACKE_sstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, lapack_int* m,
+                           float* w, float* z, lapack_int ldz, lapack_int nzc,
+                           lapack_int* isuppz, lapack_logical* tryrac );
+lapack_int LAPACKE_dstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           lapack_int* m, double* w, double* z, lapack_int ldz,
+                           lapack_int nzc, lapack_int* isuppz,
+                           lapack_logical* tryrac );
+lapack_int LAPACKE_cstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, lapack_int* m,
+                           float* w, lapack_complex_float* z, lapack_int ldz,
+                           lapack_int nzc, lapack_int* isuppz,
+                           lapack_logical* tryrac );
+lapack_int LAPACKE_zstemr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           lapack_int* m, double* w, lapack_complex_double* z,
+                           lapack_int ldz, lapack_int nzc, lapack_int* isuppz,
+                           lapack_logical* tryrac );
+
+lapack_int LAPACKE_ssteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dsteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, double* z, lapack_int ldz );
+lapack_int LAPACKE_csteqr( int matrix_order, char compz, lapack_int n, float* d,
+                           float* e, lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zsteqr( int matrix_order, char compz, lapack_int n,
+                           double* d, double* e, lapack_complex_double* z,
+                           lapack_int ldz );
+
+lapack_int LAPACKE_ssterf( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dsterf( lapack_int n, double* d, double* e );
+
+lapack_int LAPACKE_sstev( int matrix_order, char jobz, lapack_int n, float* d,
+                          float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dstev( int matrix_order, char jobz, lapack_int n, double* d,
+                          double* e, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sstevd( int matrix_order, char jobz, lapack_int n, float* d,
+                           float* e, float* z, lapack_int ldz );
+lapack_int LAPACKE_dstevd( int matrix_order, char jobz, lapack_int n, double* d,
+                           double* e, double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sstevr( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_dstevr( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* isuppz );
+
+lapack_int LAPACKE_sstevx( int matrix_order, char jobz, char range,
+                           lapack_int n, float* d, float* e, float vl, float vu,
+                           lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dstevx( int matrix_order, char jobz, char range,
+                           lapack_int n, double* d, double* e, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssycon( int matrix_order, char uplo, lapack_int n,
+                           const float* a, lapack_int lda,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_dsycon( int matrix_order, char uplo, lapack_int n,
+                           const double* a, lapack_int lda,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+lapack_int LAPACKE_csycon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv, float anorm, float* rcond );
+lapack_int LAPACKE_zsycon( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv, double anorm,
+                           double* rcond );
+
+lapack_int LAPACKE_ssyequb( int matrix_order, char uplo, lapack_int n,
+                            const float* a, lapack_int lda, float* s,
+                            float* scond, float* amax );
+lapack_int LAPACKE_dsyequb( int matrix_order, char uplo, lapack_int n,
+                            const double* a, lapack_int lda, double* s,
+                            double* scond, double* amax );
+lapack_int LAPACKE_csyequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_float* a, lapack_int lda,
+                            float* s, float* scond, float* amax );
+lapack_int LAPACKE_zsyequb( int matrix_order, char uplo, lapack_int n,
+                            const lapack_complex_double* a, lapack_int lda,
+                            double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_ssyev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_dsyev( int matrix_order, char jobz, char uplo, lapack_int n,
+                          double* a, lapack_int lda, double* w );
+
+lapack_int LAPACKE_ssyevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           float* a, lapack_int lda, float* w );
+lapack_int LAPACKE_dsyevd( int matrix_order, char jobz, char uplo, lapack_int n,
+                           double* a, lapack_int lda, double* w );
+
+lapack_int LAPACKE_ssyevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, float* a, lapack_int lda, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* isuppz );
+lapack_int LAPACKE_dsyevr( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, double* a, lapack_int lda, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* isuppz );
+
+lapack_int LAPACKE_ssyevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, float* a, lapack_int lda, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsyevx( int matrix_order, char jobz, char range, char uplo,
+                           lapack_int n, double* a, lapack_int lda, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssygst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, float* a, lapack_int lda,
+                           const float* b, lapack_int ldb );
+lapack_int LAPACKE_dsygst( int matrix_order, lapack_int itype, char uplo,
+                           lapack_int n, double* a, lapack_int lda,
+                           const double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssygv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, float* a, lapack_int lda,
+                          float* b, lapack_int ldb, float* w );
+lapack_int LAPACKE_dsygv( int matrix_order, lapack_int itype, char jobz,
+                          char uplo, lapack_int n, double* a, lapack_int lda,
+                          double* b, lapack_int ldb, double* w );
+
+lapack_int LAPACKE_ssygvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, float* a, lapack_int lda,
+                           float* b, lapack_int ldb, float* w );
+lapack_int LAPACKE_dsygvd( int matrix_order, lapack_int itype, char jobz,
+                           char uplo, lapack_int n, double* a, lapack_int lda,
+                           double* b, lapack_int ldb, double* w );
+
+lapack_int LAPACKE_ssygvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb, float vl,
+                           float vu, lapack_int il, lapack_int iu, float abstol,
+                           lapack_int* m, float* w, float* z, lapack_int ldz,
+                           lapack_int* ifail );
+lapack_int LAPACKE_dsygvx( int matrix_order, lapack_int itype, char jobz,
+                           char range, char uplo, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb, double vl,
+                           double vu, lapack_int il, lapack_int iu,
+                           double abstol, lapack_int* m, double* w, double* z,
+                           lapack_int ldz, lapack_int* ifail );
+
+lapack_int LAPACKE_ssyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const float* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const float* b,
+                           lapack_int ldb, float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dsyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const double* af, lapack_int ldaf,
+                           const lapack_int* ipiv, const double* b,
+                           lapack_int ldb, double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_csyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_zsyrfs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* af,
+                           lapack_int ldaf, const lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_ssyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const float* a,
+                            lapack_int lda, const float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* s,
+                            const float* b, lapack_int ldb, float* x,
+                            lapack_int ldx, float* rcond, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dsyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs, const double* a,
+                            lapack_int lda, const double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* s,
+                            const double* b, lapack_int ldb, double* x,
+                            lapack_int ldx, double* rcond, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+lapack_int LAPACKE_csyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_float* a, lapack_int lda,
+                            const lapack_complex_float* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const float* s,
+                            const lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* berr, lapack_int n_err_bnds,
+                            float* err_bnds_norm, float* err_bnds_comp,
+                            lapack_int nparams, float* params );
+lapack_int LAPACKE_zsyrfsx( int matrix_order, char uplo, char equed,
+                            lapack_int n, lapack_int nrhs,
+                            const lapack_complex_double* a, lapack_int lda,
+                            const lapack_complex_double* af, lapack_int ldaf,
+                            const lapack_int* ipiv, const double* s,
+                            const lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* berr, lapack_int n_err_bnds,
+                            double* err_bnds_norm, double* err_bnds_comp,
+                            lapack_int nparams, double* params );
+
+lapack_int LAPACKE_ssysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, float* a, lapack_int lda,
+                          lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dsysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, double* a, lapack_int lda,
+                          lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_csysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_float* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zsysv( int matrix_order, char uplo, lapack_int n,
+                          lapack_int nrhs, lapack_complex_double* a,
+                          lapack_int lda, lapack_int* ipiv,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           float* af, lapack_int ldaf, lapack_int* ipiv,
+                           const float* b, lapack_int ldb, float* x,
+                           lapack_int ldx, float* rcond, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dsysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           double* af, lapack_int ldaf, lapack_int* ipiv,
+                           const double* b, lapack_int ldb, double* x,
+                           lapack_int ldx, double* rcond, double* ferr,
+                           double* berr );
+lapack_int LAPACKE_csysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* x, lapack_int ldx,
+                           float* rcond, float* ferr, float* berr );
+lapack_int LAPACKE_zsysvx( int matrix_order, char fact, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* af,
+                           lapack_int ldaf, lapack_int* ipiv,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* x, lapack_int ldx,
+                           double* rcond, double* ferr, double* berr );
+
+lapack_int LAPACKE_ssysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, float* a,
+                            lapack_int lda, float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* s, float* b,
+                            lapack_int ldb, float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_dsysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs, double* a,
+                            lapack_int lda, double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* s, double* b,
+                            lapack_int ldb, double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+lapack_int LAPACKE_csysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, float* s,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* x, lapack_int ldx,
+                            float* rcond, float* rpvgrw, float* berr,
+                            lapack_int n_err_bnds, float* err_bnds_norm,
+                            float* err_bnds_comp, lapack_int nparams,
+                            float* params );
+lapack_int LAPACKE_zsysvxx( int matrix_order, char fact, char uplo,
+                            lapack_int n, lapack_int nrhs,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* af, lapack_int ldaf,
+                            lapack_int* ipiv, char* equed, double* s,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* x, lapack_int ldx,
+                            double* rcond, double* rpvgrw, double* berr,
+                            lapack_int n_err_bnds, double* err_bnds_norm,
+                            double* err_bnds_comp, lapack_int nparams,
+                            double* params );
+
+lapack_int LAPACKE_ssytrd( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, float* d, float* e, float* tau );
+lapack_int LAPACKE_dsytrd( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, double* d, double* e, double* tau );
+
+lapack_int LAPACKE_ssytrf( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dsytrf( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_csytrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_int* ipiv );
+lapack_int LAPACKE_zsytrf( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_int* ipiv );
+
+lapack_int LAPACKE_ssytri( int matrix_order, char uplo, lapack_int n, float* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dsytri( int matrix_order, char uplo, lapack_int n, double* a,
+                           lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_csytri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_int* ipiv );
+lapack_int LAPACKE_zsytri( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_int* ipiv );
+
+lapack_int LAPACKE_ssytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const float* a, lapack_int lda,
+                           const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_dsytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const double* a, lapack_int lda,
+                           const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_csytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zsytrs( int matrix_order, char uplo, lapack_int n,
+                           lapack_int nrhs, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_int* ipiv,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd, const float* ab,
+                           lapack_int ldab, float* rcond );
+lapack_int LAPACKE_dtbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd, const double* ab,
+                           lapack_int ldab, double* rcond );
+lapack_int LAPACKE_ctbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           float* rcond );
+lapack_int LAPACKE_ztbcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, lapack_int kd,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           double* rcond );
+
+lapack_int LAPACKE_stbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const float* ab, lapack_int ldab, const float* b,
+                           lapack_int ldb, const float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_dtbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const double* ab, lapack_int ldab, const double* b,
+                           lapack_int ldb, const double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+lapack_int LAPACKE_ctbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_ztbrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_stbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const float* ab, lapack_int ldab, float* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_dtbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const double* ab, lapack_int ldab, double* b,
+                           lapack_int ldb );
+lapack_int LAPACKE_ctbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_float* ab, lapack_int ldab,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztbtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int kd, lapack_int nrhs,
+                           const lapack_complex_double* ab, lapack_int ldab,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          float alpha, const float* a, float* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_dtfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          double alpha, const double* a, double* b,
+                          lapack_int ldb );
+lapack_int LAPACKE_ctfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          lapack_complex_float alpha,
+                          const lapack_complex_float* a,
+                          lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztfsm( int matrix_order, char transr, char side, char uplo,
+                          char trans, char diag, lapack_int m, lapack_int n,
+                          lapack_complex_double alpha,
+                          const lapack_complex_double* a,
+                          lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, float* a );
+lapack_int LAPACKE_dtftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, double* a );
+lapack_int LAPACKE_ctftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_ztftri( int matrix_order, char transr, char uplo, char diag,
+                           lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_stfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* arf, float* ap );
+lapack_int LAPACKE_dtfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* arf, double* ap );
+lapack_int LAPACKE_ctfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* arf,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_ztfttp( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* arf,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_stfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* arf, float* a,
+                           lapack_int lda );
+lapack_int LAPACKE_dtfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* arf, double* a,
+                           lapack_int lda );
+lapack_int LAPACKE_ctfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* arf,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztfttr( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* arf,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_stgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const float* s, lapack_int lds, const float* p,
+                           lapack_int ldp, float* vl, lapack_int ldvl,
+                           float* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_dtgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const double* s, lapack_int lds, const double* p,
+                           lapack_int ldp, double* vl, lapack_int ldvl,
+                           double* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_ctgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* s, lapack_int lds,
+                           const lapack_complex_float* p, lapack_int ldp,
+                           lapack_complex_float* vl, lapack_int ldvl,
+                           lapack_complex_float* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztgevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* s, lapack_int lds,
+                           const lapack_complex_double* p, lapack_int ldp,
+                           lapack_complex_double* vl, lapack_int ldvl,
+                           lapack_complex_double* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+
+lapack_int LAPACKE_stgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb, float* q,
+                           lapack_int ldq, float* z, lapack_int ldz,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_dtgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n, double* a,
+                           lapack_int lda, double* b, lapack_int ldb, double* q,
+                           lapack_int ldq, double* z, lapack_int ldz,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_ctgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* z, lapack_int ldz,
+                           lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztgexc( int matrix_order, lapack_logical wantq,
+                           lapack_logical wantz, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_stgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n, float* a,
+                           lapack_int lda, float* b, lapack_int ldb,
+                           float* alphar, float* alphai, float* beta, float* q,
+                           lapack_int ldq, float* z, lapack_int ldz,
+                           lapack_int* m, float* pl, float* pr, float* dif );
+lapack_int LAPACKE_dtgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n,
+                           double* a, lapack_int lda, double* b, lapack_int ldb,
+                           double* alphar, double* alphai, double* beta,
+                           double* q, lapack_int ldq, double* z, lapack_int ldz,
+                           lapack_int* m, double* pl, double* pr, double* dif );
+lapack_int LAPACKE_ctgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* alpha,
+                           lapack_complex_float* beta, lapack_complex_float* q,
+                           lapack_int ldq, lapack_complex_float* z,
+                           lapack_int ldz, lapack_int* m, float* pl, float* pr,
+                           float* dif );
+lapack_int LAPACKE_ztgsen( int matrix_order, lapack_int ijob,
+                           lapack_logical wantq, lapack_logical wantz,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* alpha,
+                           lapack_complex_double* beta,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* z, lapack_int ldz,
+                           lapack_int* m, double* pl, double* pr, double* dif );
+
+lapack_int LAPACKE_stgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, float* a, lapack_int lda,
+                           float* b, lapack_int ldb, float tola, float tolb,
+                           float* alpha, float* beta, float* u, lapack_int ldu,
+                           float* v, lapack_int ldv, float* q, lapack_int ldq,
+                           lapack_int* ncycle );
+lapack_int LAPACKE_dtgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, double* a,
+                           lapack_int lda, double* b, lapack_int ldb,
+                           double tola, double tolb, double* alpha,
+                           double* beta, double* u, lapack_int ldu, double* v,
+                           lapack_int ldv, double* q, lapack_int ldq,
+                           lapack_int* ncycle );
+lapack_int LAPACKE_ctgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* b,
+                           lapack_int ldb, float tola, float tolb, float* alpha,
+                           float* beta, lapack_complex_float* u, lapack_int ldu,
+                           lapack_complex_float* v, lapack_int ldv,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_int* ncycle );
+lapack_int LAPACKE_ztgsja( int matrix_order, char jobu, char jobv, char jobq,
+                           lapack_int m, lapack_int p, lapack_int n,
+                           lapack_int k, lapack_int l, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* b,
+                           lapack_int ldb, double tola, double tolb,
+                           double* alpha, double* beta,
+                           lapack_complex_double* u, lapack_int ldu,
+                           lapack_complex_double* v, lapack_int ldv,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_int* ncycle );
+
+lapack_int LAPACKE_stgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const float* a, lapack_int lda, const float* b,
+                           lapack_int ldb, const float* vl, lapack_int ldvl,
+                           const float* vr, lapack_int ldvr, float* s,
+                           float* dif, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_dtgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const double* a, lapack_int lda, const double* b,
+                           lapack_int ldb, const double* vl, lapack_int ldvl,
+                           const double* vr, lapack_int ldvr, double* s,
+                           double* dif, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ctgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* vl, lapack_int ldvl,
+                           const lapack_complex_float* vr, lapack_int ldvr,
+                           float* s, float* dif, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztgsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* vl, lapack_int ldvl,
+                           const lapack_complex_double* vr, lapack_int ldvr,
+                           double* s, double* dif, lapack_int mm,
+                           lapack_int* m );
+
+lapack_int LAPACKE_stgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n, const float* a,
+                           lapack_int lda, const float* b, lapack_int ldb,
+                           float* c, lapack_int ldc, const float* d,
+                           lapack_int ldd, const float* e, lapack_int lde,
+                           float* f, lapack_int ldf, float* scale, float* dif );
+lapack_int LAPACKE_dtgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n, const double* a,
+                           lapack_int lda, const double* b, lapack_int ldb,
+                           double* c, lapack_int ldc, const double* d,
+                           lapack_int ldd, const double* e, lapack_int lde,
+                           double* f, lapack_int ldf, double* scale,
+                           double* dif );
+lapack_int LAPACKE_ctgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* c, lapack_int ldc,
+                           const lapack_complex_float* d, lapack_int ldd,
+                           const lapack_complex_float* e, lapack_int lde,
+                           lapack_complex_float* f, lapack_int ldf,
+                           float* scale, float* dif );
+lapack_int LAPACKE_ztgsyl( int matrix_order, char trans, lapack_int ijob,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* c, lapack_int ldc,
+                           const lapack_complex_double* d, lapack_int ldd,
+                           const lapack_complex_double* e, lapack_int lde,
+                           lapack_complex_double* f, lapack_int ldf,
+                           double* scale, double* dif );
+
+lapack_int LAPACKE_stpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const float* ap, float* rcond );
+lapack_int LAPACKE_dtpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const double* ap, double* rcond );
+lapack_int LAPACKE_ctpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_float* ap,
+                           float* rcond );
+lapack_int LAPACKE_ztpcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_double* ap,
+                           double* rcond );
+
+lapack_int LAPACKE_stprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* ap,
+                           const float* b, lapack_int ldb, const float* x,
+                           lapack_int ldx, float* ferr, float* berr );
+lapack_int LAPACKE_dtprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* ap,
+                           const double* b, lapack_int ldb, const double* x,
+                           lapack_int ldx, double* ferr, double* berr );
+lapack_int LAPACKE_ctprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* ap,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_ztprfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* ap,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_stptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           float* ap );
+lapack_int LAPACKE_dtptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           double* ap );
+lapack_int LAPACKE_ctptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_ztptri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_stptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* ap,
+                           float* b, lapack_int ldb );
+lapack_int LAPACKE_dtptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* ap,
+                           double* b, lapack_int ldb );
+lapack_int LAPACKE_ctptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* ap,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztptrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* ap,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* ap, float* arf );
+lapack_int LAPACKE_dtpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* ap, double* arf );
+lapack_int LAPACKE_ctpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* ap,
+                           lapack_complex_float* arf );
+lapack_int LAPACKE_ztpttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* ap,
+                           lapack_complex_double* arf );
+
+lapack_int LAPACKE_stpttr( int matrix_order, char uplo, lapack_int n,
+                           const float* ap, float* a, lapack_int lda );
+lapack_int LAPACKE_dtpttr( int matrix_order, char uplo, lapack_int n,
+                           const double* ap, double* a, lapack_int lda );
+lapack_int LAPACKE_ctpttr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztpttr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_strcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const float* a, lapack_int lda,
+                           float* rcond );
+lapack_int LAPACKE_dtrcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const double* a, lapack_int lda,
+                           double* rcond );
+lapack_int LAPACKE_ctrcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda, float* rcond );
+lapack_int LAPACKE_ztrcon( int matrix_order, char norm, char uplo, char diag,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda, double* rcond );
+
+lapack_int LAPACKE_strevc( int matrix_order, char side, char howmny,
+                           lapack_logical* select, lapack_int n, const float* t,
+                           lapack_int ldt, float* vl, lapack_int ldvl,
+                           float* vr, lapack_int ldvr, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_dtrevc( int matrix_order, char side, char howmny,
+                           lapack_logical* select, lapack_int n,
+                           const double* t, lapack_int ldt, double* vl,
+                           lapack_int ldvl, double* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ctrevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* vl, lapack_int ldvl,
+                           lapack_complex_float* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztrevc( int matrix_order, char side, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* vl, lapack_int ldvl,
+                           lapack_complex_double* vr, lapack_int ldvr,
+                           lapack_int mm, lapack_int* m );
+
+lapack_int LAPACKE_strexc( int matrix_order, char compq, lapack_int n, float* t,
+                           lapack_int ldt, float* q, lapack_int ldq,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_dtrexc( int matrix_order, char compq, lapack_int n,
+                           double* t, lapack_int ldt, double* q, lapack_int ldq,
+                           lapack_int* ifst, lapack_int* ilst );
+lapack_int LAPACKE_ctrexc( int matrix_order, char compq, lapack_int n,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztrexc( int matrix_order, char compq, lapack_int n,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_strrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* a,
+                           lapack_int lda, const float* b, lapack_int ldb,
+                           const float* x, lapack_int ldx, float* ferr,
+                           float* berr );
+lapack_int LAPACKE_dtrrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* a,
+                           lapack_int lda, const double* b, lapack_int ldb,
+                           const double* x, lapack_int ldx, double* ferr,
+                           double* berr );
+lapack_int LAPACKE_ctrrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           const lapack_complex_float* x, lapack_int ldx,
+                           float* ferr, float* berr );
+lapack_int LAPACKE_ztrrfs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           const lapack_complex_double* x, lapack_int ldx,
+                           double* ferr, double* berr );
+
+lapack_int LAPACKE_strsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n, float* t,
+                           lapack_int ldt, float* q, lapack_int ldq, float* wr,
+                           float* wi, lapack_int* m, float* s, float* sep );
+lapack_int LAPACKE_dtrsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n,
+                           double* t, lapack_int ldt, double* q, lapack_int ldq,
+                           double* wr, double* wi, lapack_int* m, double* s,
+                           double* sep );
+lapack_int LAPACKE_ctrsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* q, lapack_int ldq,
+                           lapack_complex_float* w, lapack_int* m, float* s,
+                           float* sep );
+lapack_int LAPACKE_ztrsen( int matrix_order, char job, char compq,
+                           const lapack_logical* select, lapack_int n,
+                           lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* q, lapack_int ldq,
+                           lapack_complex_double* w, lapack_int* m, double* s,
+                           double* sep );
+
+lapack_int LAPACKE_strsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const float* t, lapack_int ldt, const float* vl,
+                           lapack_int ldvl, const float* vr, lapack_int ldvr,
+                           float* s, float* sep, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_dtrsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const double* t, lapack_int ldt, const double* vl,
+                           lapack_int ldvl, const double* vr, lapack_int ldvr,
+                           double* s, double* sep, lapack_int mm,
+                           lapack_int* m );
+lapack_int LAPACKE_ctrsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_float* t, lapack_int ldt,
+                           const lapack_complex_float* vl, lapack_int ldvl,
+                           const lapack_complex_float* vr, lapack_int ldvr,
+                           float* s, float* sep, lapack_int mm, lapack_int* m );
+lapack_int LAPACKE_ztrsna( int matrix_order, char job, char howmny,
+                           const lapack_logical* select, lapack_int n,
+                           const lapack_complex_double* t, lapack_int ldt,
+                           const lapack_complex_double* vl, lapack_int ldvl,
+                           const lapack_complex_double* vr, lapack_int ldvr,
+                           double* s, double* sep, lapack_int mm,
+                           lapack_int* m );
+
+lapack_int LAPACKE_strsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const float* a, lapack_int lda, const float* b,
+                           lapack_int ldb, float* c, lapack_int ldc,
+                           float* scale );
+lapack_int LAPACKE_dtrsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const double* a, lapack_int lda, const double* b,
+                           lapack_int ldb, double* c, lapack_int ldc,
+                           double* scale );
+lapack_int LAPACKE_ctrsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* b, lapack_int ldb,
+                           lapack_complex_float* c, lapack_int ldc,
+                           float* scale );
+lapack_int LAPACKE_ztrsyl( int matrix_order, char trana, char tranb,
+                           lapack_int isgn, lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* c, lapack_int ldc,
+                           double* scale );
+
+lapack_int LAPACKE_strtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           float* a, lapack_int lda );
+lapack_int LAPACKE_dtrtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           double* a, lapack_int lda );
+lapack_int LAPACKE_ctrtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztrtri( int matrix_order, char uplo, char diag, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_strtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const float* a,
+                           lapack_int lda, float* b, lapack_int ldb );
+lapack_int LAPACKE_dtrtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs, const double* a,
+                           lapack_int lda, double* b, lapack_int ldb );
+lapack_int LAPACKE_ctrtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztrtrs( int matrix_order, char uplo, char trans, char diag,
+                           lapack_int n, lapack_int nrhs,
+                           const lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_strttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const float* a, lapack_int lda,
+                           float* arf );
+lapack_int LAPACKE_dtrttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const double* a, lapack_int lda,
+                           double* arf );
+lapack_int LAPACKE_ctrttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* arf );
+lapack_int LAPACKE_ztrttf( int matrix_order, char transr, char uplo,
+                           lapack_int n, const lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* arf );
+
+lapack_int LAPACKE_strttp( int matrix_order, char uplo, lapack_int n,
+                           const float* a, lapack_int lda, float* ap );
+lapack_int LAPACKE_dtrttp( int matrix_order, char uplo, lapack_int n,
+                           const double* a, lapack_int lda, double* ap );
+lapack_int LAPACKE_ctrttp( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* ap );
+lapack_int LAPACKE_ztrttp( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* ap );
+
+lapack_int LAPACKE_stzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           float* a, lapack_int lda, float* tau );
+lapack_int LAPACKE_dtzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           double* a, lapack_int lda, double* tau );
+lapack_int LAPACKE_ctzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* tau );
+lapack_int LAPACKE_ztzrzf( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungbr( int matrix_order, char vect, lapack_int m,
+                           lapack_int n, lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cunghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zunghr( int matrix_order, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cunglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zunglq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungql( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungqr( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau );
+lapack_int LAPACKE_zungrq( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cungtr( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau );
+lapack_int LAPACKE_zungtr( int matrix_order, char uplo, lapack_int n,
+                           lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau );
+
+lapack_int LAPACKE_cunmbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmbr( int matrix_order, char vect, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmhr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int ilo,
+                           lapack_int ihi, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmlq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmql( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmqr( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmrq( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const lapack_complex_float* a,
+                           lapack_int lda, const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmrz( int matrix_order, char side, char trans,
+                           lapack_int m, lapack_int n, lapack_int k,
+                           lapack_int l, const lapack_complex_double* a,
+                           lapack_int lda, const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cunmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_float* a, lapack_int lda,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zunmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_double* a, lapack_int lda,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_cupgtr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* q, lapack_int ldq );
+lapack_int LAPACKE_zupgtr( int matrix_order, char uplo, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* q, lapack_int ldq );
+
+lapack_int LAPACKE_cupmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_float* ap,
+                           const lapack_complex_float* tau,
+                           lapack_complex_float* c, lapack_int ldc );
+lapack_int LAPACKE_zupmtr( int matrix_order, char side, char uplo, char trans,
+                           lapack_int m, lapack_int n,
+                           const lapack_complex_double* ap,
+                           const lapack_complex_double* tau,
+                           lapack_complex_double* c, lapack_int ldc );
+
+lapack_int LAPACKE_sbdsdc_work( int matrix_order, char uplo, char compq,
+                                lapack_int n, float* d, float* e, float* u,
+                                lapack_int ldu, float* vt, lapack_int ldvt,
+                                float* q, lapack_int* iq, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dbdsdc_work( int matrix_order, char uplo, char compq,
+                                lapack_int n, double* d, double* e, double* u,
+                                lapack_int ldu, double* vt, lapack_int ldvt,
+                                double* q, lapack_int* iq, double* work,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                float* d, float* e, float* vt, lapack_int ldvt,
+                                float* u, lapack_int ldu, float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_dbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                double* d, double* e, double* vt,
+                                lapack_int ldvt, double* u, lapack_int ldu,
+                                double* c, lapack_int ldc, double* work );
+lapack_int LAPACKE_cbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                float* d, float* e, lapack_complex_float* vt,
+                                lapack_int ldvt, lapack_complex_float* u,
+                                lapack_int ldu, lapack_complex_float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_zbdsqr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int ncvt, lapack_int nru, lapack_int ncc,
+                                double* d, double* e, lapack_complex_double* vt,
+                                lapack_int ldvt, lapack_complex_double* u,
+                                lapack_int ldu, lapack_complex_double* c,
+                                lapack_int ldc, double* work );
+
+lapack_int LAPACKE_sdisna_work( char job, lapack_int m, lapack_int n,
+                                const float* d, float* sep );
+lapack_int LAPACKE_ddisna_work( char job, lapack_int m, lapack_int n,
+                                const double* d, double* sep );
+
+lapack_int LAPACKE_sgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, float* ab, lapack_int ldab,
+                                float* d, float* e, float* q, lapack_int ldq,
+                                float* pt, lapack_int ldpt, float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_dgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, double* ab, lapack_int ldab,
+                                double* d, double* e, double* q, lapack_int ldq,
+                                double* pt, lapack_int ldpt, double* c,
+                                lapack_int ldc, double* work );
+lapack_int LAPACKE_cgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, lapack_complex_float* ab,
+                                lapack_int ldab, float* d, float* e,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* pt, lapack_int ldpt,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgbbrd_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int ncc, lapack_int kl,
+                                lapack_int ku, lapack_complex_double* ab,
+                                lapack_int ldab, double* d, double* e,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* pt, lapack_int ldpt,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                double anorm, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zgbcon_work( int matrix_order, char norm, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* ab,
+                                lapack_int ldab, float* r, float* c,
+                                float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* ab,
+                                lapack_int ldab, double* r, double* c,
+                                double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                float* r, float* c, float* rowcnd,
+                                float* colcnd, float* amax );
+lapack_int LAPACKE_zgbequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, double* r, double* c,
+                                double* rowcnd, double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku, const float* ab,
+                                 lapack_int ldab, float* r, float* c,
+                                 float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_dgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku, const double* ab,
+                                 lapack_int ldab, double* r, double* c,
+                                 double* rowcnd, double* colcnd, double* amax );
+lapack_int LAPACKE_cgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku,
+                                 const lapack_complex_float* ab,
+                                 lapack_int ldab, float* r, float* c,
+                                 float* rowcnd, float* colcnd, float* amax );
+lapack_int LAPACKE_zgbequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_int kl, lapack_int ku,
+                                 const lapack_complex_double* ab,
+                                 lapack_int ldab, double* r, double* c,
+                                 double* rowcnd, double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const float* ab, lapack_int ldab,
+                                const float* afb, lapack_int ldafb,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const double* ab, lapack_int ldab,
+                                const double* afb, lapack_int ldafb,
+                                const lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_complex_float* afb,
+                                lapack_int ldafb, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgbrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab,
+                                const lapack_complex_double* afb,
+                                lapack_int ldafb, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, const float* ab,
+                                 lapack_int ldab, const float* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const float* r, const float* c, const float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, const double* ab,
+                                 lapack_int ldab, const double* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs,
+                                 const lapack_complex_float* ab,
+                                 lapack_int ldab,
+                                 const lapack_complex_float* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const float* r, const float* c,
+                                 const lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zgbrfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs,
+                                 const lapack_complex_double* ab,
+                                 lapack_int ldab,
+                                 const lapack_complex_double* afb,
+                                 lapack_int ldafb, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs, float* ab,
+                               lapack_int ldab, lapack_int* ipiv, float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs, double* ab,
+                               lapack_int ldab, lapack_int* ipiv, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zgbsv_work( int matrix_order, lapack_int n, lapack_int kl,
+                               lapack_int ku, lapack_int nrhs,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+
+lapack_int LAPACKE_sgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, float* ab, lapack_int ldab,
+                                float* afb, lapack_int ldafb, lapack_int* ipiv,
+                                char* equed, float* r, float* c, float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, double* ab, lapack_int ldab,
+                                double* afb, lapack_int ldafb, lapack_int* ipiv,
+                                char* equed, double* r, double* c, double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, lapack_complex_float* ab,
+                                lapack_int ldab, lapack_complex_float* afb,
+                                lapack_int ldafb, lapack_int* ipiv, char* equed,
+                                float* r, float* c, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zgbsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int kl, lapack_int ku,
+                                lapack_int nrhs, lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* afb,
+                                lapack_int ldafb, lapack_int* ipiv, char* equed,
+                                double* r, double* c, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_sgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, float* ab, lapack_int ldab,
+                                 float* afb, lapack_int ldafb, lapack_int* ipiv,
+                                 char* equed, float* r, float* c, float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, double* ab, lapack_int ldab,
+                                 double* afb, lapack_int ldafb,
+                                 lapack_int* ipiv, char* equed, double* r,
+                                 double* c, double* b, lapack_int ldb,
+                                 double* x, lapack_int ldx, double* rcond,
+                                 double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, lapack_complex_float* ab,
+                                 lapack_int ldab, lapack_complex_float* afb,
+                                 lapack_int ldafb, lapack_int* ipiv,
+                                 char* equed, float* r, float* c,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zgbsvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int kl, lapack_int ku,
+                                 lapack_int nrhs, lapack_complex_double* ab,
+                                 lapack_int ldab, lapack_complex_double* afb,
+                                 lapack_int ldafb, lapack_int* ipiv,
+                                 char* equed, double* r, double* c,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, float* ab,
+                                lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_dgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, double* ab,
+                                lapack_int ldab, lapack_int* ipiv );
+lapack_int LAPACKE_cgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_int* ipiv );
+lapack_int LAPACKE_zgbtrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_int* ipiv );
+
+lapack_int LAPACKE_sgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const float* ab, lapack_int ldab,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const double* ab, lapack_int ldab,
+                                const lapack_int* ipiv, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zgbtrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int kl, lapack_int ku, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* scale, lapack_int m, float* v,
+                                lapack_int ldv );
+lapack_int LAPACKE_dgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* scale, lapack_int m, double* v,
+                                lapack_int ldv );
+lapack_int LAPACKE_cgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* scale, lapack_int m,
+                                lapack_complex_float* v, lapack_int ldv );
+lapack_int LAPACKE_zgebak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* scale, lapack_int m,
+                                lapack_complex_double* v, lapack_int ldv );
+
+lapack_int LAPACKE_sgebal_work( int matrix_order, char job, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ilo,
+                                lapack_int* ihi, float* scale );
+lapack_int LAPACKE_dgebal_work( int matrix_order, char job, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ilo,
+                                lapack_int* ihi, double* scale );
+lapack_int LAPACKE_cgebal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ilo, lapack_int* ihi,
+                                float* scale );
+lapack_int LAPACKE_zgebal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ilo, lapack_int* ihi,
+                                double* scale );
+
+lapack_int LAPACKE_sgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* d, float* e,
+                                float* tauq, float* taup, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* d, double* e,
+                                double* tauq, double* taup, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_cgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float* d, float* e, lapack_complex_float* tauq,
+                                lapack_complex_float* taup,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgebrd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double* d, double* e,
+                                lapack_complex_double* tauq,
+                                lapack_complex_double* taup,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgecon_work( int matrix_order, char norm, lapack_int n,
+                                const float* a, lapack_int lda, float anorm,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgecon_work( int matrix_order, char norm, lapack_int n,
+                                const double* a, lapack_int lda, double anorm,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgecon_work( int matrix_order, char norm, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float anorm, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgecon_work( int matrix_order, char norm, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const float* a, lapack_int lda, float* r,
+                                float* c, float* rowcnd, float* colcnd,
+                                float* amax );
+lapack_int LAPACKE_dgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda, double* r,
+                                double* c, double* rowcnd, double* colcnd,
+                                double* amax );
+lapack_int LAPACKE_cgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* r, float* c, float* rowcnd,
+                                float* colcnd, float* amax );
+lapack_int LAPACKE_zgeequ_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* r, double* c, double* rowcnd,
+                                double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const float* a, lapack_int lda, float* r,
+                                 float* c, float* rowcnd, float* colcnd,
+                                 float* amax );
+lapack_int LAPACKE_dgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const double* a, lapack_int lda, double* r,
+                                 double* c, double* rowcnd, double* colcnd,
+                                 double* amax );
+lapack_int LAPACKE_cgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* r, float* c, float* rowcnd,
+                                 float* colcnd, float* amax );
+lapack_int LAPACKE_zgeequb_work( int matrix_order, lapack_int m, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* r, double* c, double* rowcnd,
+                                 double* colcnd, double* amax );
+
+lapack_int LAPACKE_sgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_S_SELECT2 select, lapack_int n, float* a,
+                               lapack_int lda, lapack_int* sdim, float* wr,
+                               float* wi, float* vs, lapack_int ldvs,
+                               float* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_dgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_D_SELECT2 select, lapack_int n, double* a,
+                               lapack_int lda, lapack_int* sdim, double* wr,
+                               double* wi, double* vs, lapack_int ldvs,
+                               double* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_cgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_C_SELECT1 select, lapack_int n,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_int* sdim, lapack_complex_float* w,
+                               lapack_complex_float* vs, lapack_int ldvs,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork, lapack_logical* bwork );
+lapack_int LAPACKE_zgees_work( int matrix_order, char jobvs, char sort,
+                               LAPACK_Z_SELECT1 select, lapack_int n,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_int* sdim, lapack_complex_double* w,
+                               lapack_complex_double* vs, lapack_int ldvs,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_S_SELECT2 select, char sense,
+                                lapack_int n, float* a, lapack_int lda,
+                                lapack_int* sdim, float* wr, float* wi,
+                                float* vs, lapack_int ldvs, float* rconde,
+                                float* rcondv, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_dgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_D_SELECT2 select, char sense,
+                                lapack_int n, double* a, lapack_int lda,
+                                lapack_int* sdim, double* wr, double* wi,
+                                double* vs, lapack_int ldvs, double* rconde,
+                                double* rcondv, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_cgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_C_SELECT1 select, char sense,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_int* sdim,
+                                lapack_complex_float* w,
+                                lapack_complex_float* vs, lapack_int ldvs,
+                                float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_logical* bwork );
+lapack_int LAPACKE_zgeesx_work( int matrix_order, char jobvs, char sort,
+                                LAPACK_Z_SELECT1 select, char sense,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_int* sdim,
+                                lapack_complex_double* w,
+                                lapack_complex_double* vs, lapack_int ldvs,
+                                double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, float* a, lapack_int lda,
+                               float* wr, float* wi, float* vl, lapack_int ldvl,
+                               float* vr, lapack_int ldvr, float* work,
+                               lapack_int lwork );
+lapack_int LAPACKE_dgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, double* a, lapack_int lda,
+                               double* wr, double* wi, double* vl,
+                               lapack_int ldvl, double* vr, lapack_int ldvr,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* w,
+                               lapack_complex_float* vl, lapack_int ldvl,
+                               lapack_complex_float* vr, lapack_int ldvr,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zgeev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_double* a,
+                               lapack_int lda, lapack_complex_double* w,
+                               lapack_complex_double* vl, lapack_int ldvl,
+                               lapack_complex_double* vr, lapack_int ldvr,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork );
+
+lapack_int LAPACKE_sgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, float* a,
+                                lapack_int lda, float* wr, float* wi, float* vl,
+                                lapack_int ldvl, float* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, float* scale,
+                                float* abnrm, float* rconde, float* rcondv,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, double* a,
+                                lapack_int lda, double* wr, double* wi,
+                                double* vl, lapack_int ldvl, double* vr,
+                                lapack_int ldvr, lapack_int* ilo,
+                                lapack_int* ihi, double* scale, double* abnrm,
+                                double* rconde, double* rcondv, double* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_cgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* w,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, float* scale,
+                                float* abnrm, float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork );
+lapack_int LAPACKE_zgeevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* w,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, double* scale,
+                                double* abnrm, double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_sgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, float* a, lapack_int lda,
+                                float* tau, float* work, lapack_int lwork );
+lapack_int LAPACKE_dgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, double* a, lapack_int lda,
+                                double* tau, double* work, lapack_int lwork );
+lapack_int LAPACKE_cgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgehrd_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgejsv_work( int matrix_order, char joba, char jobu,
+                                char jobv, char jobr, char jobt, char jobp,
+                                lapack_int m, lapack_int n, float* a,
+                                lapack_int lda, float* sva, float* u,
+                                lapack_int ldu, float* v, lapack_int ldv,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgejsv_work( int matrix_order, char joba, char jobu,
+                                char jobv, char jobr, char jobt, char jobp,
+                                lapack_int m, lapack_int n, double* a,
+                                lapack_int lda, double* sva, double* u,
+                                lapack_int ldu, double* v, lapack_int ldv,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work );
+lapack_int LAPACKE_dgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work );
+lapack_int LAPACKE_cgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zgelq2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_sgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgelqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs, float* a,
+                               lapack_int lda, float* b, lapack_int ldb,
+                               float* work, lapack_int lwork );
+lapack_int LAPACKE_dgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs, double* a,
+                               lapack_int lda, double* b, lapack_int ldb,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_cgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgels_work( int matrix_order, char trans, lapack_int m,
+                               lapack_int n, lapack_int nrhs,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* s,
+                                double rcond, lapack_int* rank, double* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_cgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_zgelsd_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, double* s, double rcond,
+                                lapack_int* rank, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* s,
+                                double rcond, lapack_int* rank, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_cgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, float* s, float rcond,
+                                lapack_int* rank, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zgelss_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, double* s, double rcond,
+                                lapack_int* rank, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_sgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, lapack_int* jpvt,
+                                float rcond, lapack_int* rank, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, lapack_int* jpvt,
+                                double rcond, lapack_int* rank, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_cgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_int* jpvt, float rcond,
+                                lapack_int* rank, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zgelsy_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_int* jpvt, double rcond,
+                                lapack_int* rank, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_sgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgeqlf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* jpvt,
+                                float* tau, float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* jpvt,
+                                double* tau, double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork );
+lapack_int LAPACKE_zgeqp3_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_sgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* jpvt,
+                                float* tau, float* work );
+lapack_int LAPACKE_dgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* jpvt,
+                                double* tau, double* work );
+lapack_int LAPACKE_cgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_float* tau,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgeqpf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* jpvt, lapack_complex_double* tau,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work );
+lapack_int LAPACKE_dgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work );
+lapack_int LAPACKE_cgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zgeqr2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_sgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgeqrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* tau,
+                                 float* work, lapack_int lwork );
+lapack_int LAPACKE_dgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* tau,
+                                 double* work, lapack_int lwork );
+lapack_int LAPACKE_cgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* tau,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgeqrfp_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* tau,
+                                 lapack_complex_double* work,
+                                 lapack_int lwork );
+
+lapack_int LAPACKE_sgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const float* af, lapack_int ldaf,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgerfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs, const float* a,
+                                 lapack_int lda, const float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* r, const float* c, const float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs, const double* a,
+                                 lapack_int lda, const double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* r, const float* c,
+                                 const lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zgerfsx_work( int matrix_order, char trans, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* r, const double* c,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgerqf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* s, float* u, lapack_int ldu, float* vt,
+                                lapack_int ldvt, float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* s, double* u, lapack_int ldu,
+                                double* vt, lapack_int ldvt, double* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_cgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, float* s,
+                                lapack_complex_float* u, lapack_int ldu,
+                                lapack_complex_float* vt, lapack_int ldvt,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int* iwork );
+lapack_int LAPACKE_zgesdd_work( int matrix_order, char jobz, lapack_int m,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, double* s,
+                                lapack_complex_double* u, lapack_int ldu,
+                                lapack_complex_double* vt, lapack_int ldvt,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int* iwork );
+
+lapack_int LAPACKE_sgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* a, lapack_int lda, lapack_int* ipiv,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* a, lapack_int lda, lapack_int* ipiv,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_cgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dsgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                double* a, lapack_int lda, lapack_int* ipiv,
+                                double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* work, float* swork,
+                                lapack_int* iter );
+lapack_int LAPACKE_zcgesv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, lapack_complex_double* work,
+                                lapack_complex_float* swork, double* rwork,
+                                lapack_int* iter );
+
+lapack_int LAPACKE_sgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n, float* a,
+                                lapack_int lda, float* s, float* u,
+                                lapack_int ldu, float* vt, lapack_int ldvt,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n, double* a,
+                                lapack_int lda, double* s, double* u,
+                                lapack_int ldu, double* vt, lapack_int ldvt,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float* s, lapack_complex_float* u,
+                                lapack_int ldu, lapack_complex_float* vt,
+                                lapack_int ldvt, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zgesvd_work( int matrix_order, char jobu, char jobvt,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double* s, lapack_complex_double* u,
+                                lapack_int ldu, lapack_complex_double* vt,
+                                lapack_int ldvt, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_sgesvj_work( int matrix_order, char joba, char jobu,
+                                char jobv, lapack_int m, lapack_int n, float* a,
+                                lapack_int lda, float* sva, lapack_int mv,
+                                float* v, lapack_int ldv, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dgesvj_work( int matrix_order, char joba, char jobu,
+                                char jobv, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* sva,
+                                lapack_int mv, double* v, lapack_int ldv,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, float* a,
+                                lapack_int lda, float* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, float* r,
+                                float* c, float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, double* a,
+                                lapack_int lda, double* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, double* r,
+                                double* c, double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, float* r,
+                                float* c, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zgesvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                lapack_int* ipiv, char* equed, double* r,
+                                double* c, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_sgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs, float* a,
+                                 lapack_int lda, float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* r,
+                                 float* c, float* b, lapack_int ldb, float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs, double* a,
+                                 lapack_int lda, double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* r,
+                                 double* c, double* b, lapack_int ldb,
+                                 double* x, lapack_int ldx, double* rcond,
+                                 double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* r,
+                                 float* c, lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params,
+                                 lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgesvxx_work( int matrix_order, char fact, char trans,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* r,
+                                 double* c, lapack_complex_double* b,
+                                 lapack_int ldb, lapack_complex_double* x,
+                                 lapack_int ldx, double* rcond, double* rpvgrw,
+                                 double* berr, lapack_int n_err_bnds,
+                                 double* err_bnds_norm, double* err_bnds_comp,
+                                 lapack_int nparams, double* params,
+                                 lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv );
+lapack_int LAPACKE_zgetf2_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_dgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ipiv );
+lapack_int LAPACKE_cgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv );
+lapack_int LAPACKE_zgetrf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv );
+
+lapack_int LAPACKE_sgetri_work( int matrix_order, lapack_int n, float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dgetri_work( int matrix_order, lapack_int n, double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_cgetri_work( int matrix_order, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgetri_work( int matrix_order, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_cgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgetrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* lscale, const float* rscale,
+                                lapack_int m, float* v, lapack_int ldv );
+lapack_int LAPACKE_dggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* lscale, const double* rscale,
+                                lapack_int m, double* v, lapack_int ldv );
+lapack_int LAPACKE_cggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const float* lscale, const float* rscale,
+                                lapack_int m, lapack_complex_float* v,
+                                lapack_int ldv );
+lapack_int LAPACKE_zggbak_work( int matrix_order, char job, char side,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                const double* lscale, const double* rscale,
+                                lapack_int m, lapack_complex_double* v,
+                                lapack_int ldv );
+
+lapack_int LAPACKE_sggbal_work( int matrix_order, char job, lapack_int n,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, lapack_int* ilo,
+                                lapack_int* ihi, float* lscale, float* rscale,
+                                float* work );
+lapack_int LAPACKE_dggbal_work( int matrix_order, char job, lapack_int n,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, lapack_int* ilo,
+                                lapack_int* ihi, double* lscale, double* rscale,
+                                double* work );
+lapack_int LAPACKE_cggbal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_int* ilo, lapack_int* ihi, float* lscale,
+                                float* rscale, float* work );
+lapack_int LAPACKE_zggbal_work( int matrix_order, char job, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_int* ilo, lapack_int* ihi,
+                                double* lscale, double* rscale, double* work );
+
+lapack_int LAPACKE_sgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_S_SELECT3 selctg, lapack_int n,
+                               float* a, lapack_int lda, float* b,
+                               lapack_int ldb, lapack_int* sdim, float* alphar,
+                               float* alphai, float* beta, float* vsl,
+                               lapack_int ldvsl, float* vsr, lapack_int ldvsr,
+                               float* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_dgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_D_SELECT3 selctg, lapack_int n,
+                               double* a, lapack_int lda, double* b,
+                               lapack_int ldb, lapack_int* sdim, double* alphar,
+                               double* alphai, double* beta, double* vsl,
+                               lapack_int ldvsl, double* vsr, lapack_int ldvsr,
+                               double* work, lapack_int lwork,
+                               lapack_logical* bwork );
+lapack_int LAPACKE_cgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_C_SELECT2 selctg, lapack_int n,
+                               lapack_complex_float* a, lapack_int lda,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_int* sdim, lapack_complex_float* alpha,
+                               lapack_complex_float* beta,
+                               lapack_complex_float* vsl, lapack_int ldvsl,
+                               lapack_complex_float* vsr, lapack_int ldvsr,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork, lapack_logical* bwork );
+lapack_int LAPACKE_zgges_work( int matrix_order, char jobvsl, char jobvsr,
+                               char sort, LAPACK_Z_SELECT2 selctg, lapack_int n,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_int* sdim, lapack_complex_double* alpha,
+                               lapack_complex_double* beta,
+                               lapack_complex_double* vsl, lapack_int ldvsl,
+                               lapack_complex_double* vsr, lapack_int ldvsr,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_S_SELECT3 selctg, char sense,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, lapack_int* sdim,
+                                float* alphar, float* alphai, float* beta,
+                                float* vsl, lapack_int ldvsl, float* vsr,
+                                lapack_int ldvsr, float* rconde, float* rcondv,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_dggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_D_SELECT3 selctg, char sense,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, lapack_int* sdim,
+                                double* alphar, double* alphai, double* beta,
+                                double* vsl, lapack_int ldvsl, double* vsr,
+                                lapack_int ldvsr, double* rconde,
+                                double* rcondv, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_cggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_C_SELECT2 selctg, char sense,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_int* sdim,
+                                lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* vsl, lapack_int ldvsl,
+                                lapack_complex_float* vsr, lapack_int ldvsr,
+                                float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int* iwork,
+                                lapack_int liwork, lapack_logical* bwork );
+lapack_int LAPACKE_zggesx_work( int matrix_order, char jobvsl, char jobvsr,
+                                char sort, LAPACK_Z_SELECT2 selctg, char sense,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_int* sdim,
+                                lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* vsl, lapack_int ldvsl,
+                                lapack_complex_double* vsr, lapack_int ldvsr,
+                                double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int* iwork,
+                                lapack_int liwork, lapack_logical* bwork );
+
+lapack_int LAPACKE_sggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, float* a, lapack_int lda, float* b,
+                               lapack_int ldb, float* alphar, float* alphai,
+                               float* beta, float* vl, lapack_int ldvl,
+                               float* vr, lapack_int ldvr, float* work,
+                               lapack_int lwork );
+lapack_int LAPACKE_dggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, double* a, lapack_int lda,
+                               double* b, lapack_int ldb, double* alphar,
+                               double* alphai, double* beta, double* vl,
+                               lapack_int ldvl, double* vr, lapack_int ldvr,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_cggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* b,
+                               lapack_int ldb, lapack_complex_float* alpha,
+                               lapack_complex_float* beta,
+                               lapack_complex_float* vl, lapack_int ldvl,
+                               lapack_complex_float* vr, lapack_int ldvr,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zggev_work( int matrix_order, char jobvl, char jobvr,
+                               lapack_int n, lapack_complex_double* a,
+                               lapack_int lda, lapack_complex_double* b,
+                               lapack_int ldb, lapack_complex_double* alpha,
+                               lapack_complex_double* beta,
+                               lapack_complex_double* vl, lapack_int ldvl,
+                               lapack_complex_double* vr, lapack_int ldvr,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork );
+
+lapack_int LAPACKE_sggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float* alphar, float* alphai, float* beta,
+                                float* vl, lapack_int ldvl, float* vr,
+                                lapack_int ldvr, lapack_int* ilo,
+                                lapack_int* ihi, float* lscale, float* rscale,
+                                float* abnrm, float* bbnrm, float* rconde,
+                                float* rcondv, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_logical* bwork );
+lapack_int LAPACKE_dggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* alphar, double* alphai, double* beta,
+                                double* vl, lapack_int ldvl, double* vr,
+                                lapack_int ldvr, lapack_int* ilo,
+                                lapack_int* ihi, double* lscale, double* rscale,
+                                double* abnrm, double* bbnrm, double* rconde,
+                                double* rcondv, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_logical* bwork );
+lapack_int LAPACKE_cggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi, float* lscale,
+                                float* rscale, float* abnrm, float* bbnrm,
+                                float* rconde, float* rcondv,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int* iwork,
+                                lapack_logical* bwork );
+lapack_int LAPACKE_zggevx_work( int matrix_order, char balanc, char jobvl,
+                                char jobvr, char sense, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int* ilo, lapack_int* ihi,
+                                double* lscale, double* rscale, double* abnrm,
+                                double* bbnrm, double* rconde, double* rcondv,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int* iwork,
+                                lapack_logical* bwork );
+
+lapack_int LAPACKE_sggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* d, float* x,
+                                float* y, float* work, lapack_int lwork );
+lapack_int LAPACKE_dggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* d, double* x,
+                                double* y, double* work, lapack_int lwork );
+lapack_int LAPACKE_cggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* d,
+                                lapack_complex_float* x,
+                                lapack_complex_float* y,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zggglm_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* d,
+                                lapack_complex_double* x,
+                                lapack_complex_double* y,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float* q, lapack_int ldq,
+                                float* z, lapack_int ldz );
+lapack_int LAPACKE_dgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double* q, lapack_int ldq,
+                                double* z, lapack_int ldz );
+lapack_int LAPACKE_cgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz );
+lapack_int LAPACKE_zgghrd_work( int matrix_order, char compq, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz );
+
+lapack_int LAPACKE_sgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float* c, float* d,
+                                float* x, float* work, lapack_int lwork );
+lapack_int LAPACKE_dgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* c, double* d,
+                                double* x, double* work, lapack_int lwork );
+lapack_int LAPACKE_cgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* c,
+                                lapack_complex_float* d,
+                                lapack_complex_float* x,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zgglse_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* c,
+                                lapack_complex_double* d,
+                                lapack_complex_double* x,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, float* a, lapack_int lda,
+                                float* taua, float* b, lapack_int ldb,
+                                float* taub, float* work, lapack_int lwork );
+lapack_int LAPACKE_dggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, double* a, lapack_int lda,
+                                double* taua, double* b, lapack_int ldb,
+                                double* taub, double* work, lapack_int lwork );
+lapack_int LAPACKE_cggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* taua,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* taub,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zggqrf_work( int matrix_order, lapack_int n, lapack_int m,
+                                lapack_int p, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* taua,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* taub,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* taua, float* b, lapack_int ldb,
+                                float* taub, float* work, lapack_int lwork );
+lapack_int LAPACKE_dggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* taua, double* b, lapack_int ldb,
+                                double* taub, double* work, lapack_int lwork );
+lapack_int LAPACKE_cggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* taua,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* taub,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zggrqf_work( int matrix_order, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* taua,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* taub,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float* alpha, float* beta,
+                                float* u, lapack_int ldu, float* v,
+                                lapack_int ldv, float* q, lapack_int ldq,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double* alpha, double* beta,
+                                double* u, lapack_int ldu, double* v,
+                                lapack_int ldv, double* q, lapack_int ldq,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float* alpha, float* beta,
+                                lapack_complex_float* u, lapack_int ldu,
+                                lapack_complex_float* v, lapack_int ldv,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_zggsvd_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int n,
+                                lapack_int p, lapack_int* k, lapack_int* l,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double* alpha, double* beta,
+                                lapack_complex_double* u, lapack_int ldu,
+                                lapack_complex_double* v, lapack_int ldv,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_sggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, float tola,
+                                float tolb, lapack_int* k, lapack_int* l,
+                                float* u, lapack_int ldu, float* v,
+                                lapack_int ldv, float* q, lapack_int ldq,
+                                lapack_int* iwork, float* tau, float* work );
+lapack_int LAPACKE_dggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double tola,
+                                double tolb, lapack_int* k, lapack_int* l,
+                                double* u, lapack_int ldu, double* v,
+                                lapack_int ldv, double* q, lapack_int ldq,
+                                lapack_int* iwork, double* tau, double* work );
+lapack_int LAPACKE_cggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb, float tola, float tolb,
+                                lapack_int* k, lapack_int* l,
+                                lapack_complex_float* u, lapack_int ldu,
+                                lapack_complex_float* v, lapack_int ldv,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_int* iwork, float* rwork,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zggsvp_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, double tola, double tolb,
+                                lapack_int* k, lapack_int* l,
+                                lapack_complex_double* u, lapack_int ldu,
+                                lapack_complex_double* v, lapack_int ldv,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_int* iwork, double* rwork,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_sgtcon_work( char norm, lapack_int n, const float* dl,
+                                const float* d, const float* du,
+                                const float* du2, const lapack_int* ipiv,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgtcon_work( char norm, lapack_int n, const double* dl,
+                                const double* d, const double* du,
+                                const double* du2, const lapack_int* ipiv,
+                                double anorm, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgtcon_work( char norm, lapack_int n,
+                                const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                const lapack_complex_float* du2,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zgtcon_work( char norm, lapack_int n,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                const lapack_complex_double* du2,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_sgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* dl,
+                                const float* d, const float* du,
+                                const float* dlf, const float* df,
+                                const float* duf, const float* du2,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* dl,
+                                const double* d, const double* du,
+                                const double* dlf, const double* df,
+                                const double* duf, const double* du2,
+                                const lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                const lapack_complex_float* dlf,
+                                const lapack_complex_float* df,
+                                const lapack_complex_float* duf,
+                                const lapack_complex_float* du2,
+                                const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgtrfs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                const lapack_complex_double* dlf,
+                                const lapack_complex_double* df,
+                                const lapack_complex_double* duf,
+                                const lapack_complex_double* du2,
+                                const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* dl, float* d, float* du, float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* dl, double* d, double* du, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_float* dl,
+                               lapack_complex_float* d,
+                               lapack_complex_float* du,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zgtsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               lapack_complex_double* dl,
+                               lapack_complex_double* d,
+                               lapack_complex_double* du,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, const float* dl,
+                                const float* d, const float* du, float* dlf,
+                                float* df, float* duf, float* du2,
+                                lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs, const double* dl,
+                                const double* d, const double* du, double* dlf,
+                                double* df, double* duf, double* du2,
+                                lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                lapack_complex_float* dlf,
+                                lapack_complex_float* df,
+                                lapack_complex_float* duf,
+                                lapack_complex_float* du2, lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zgtsvx_work( int matrix_order, char fact, char trans,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                lapack_complex_double* dlf,
+                                lapack_complex_double* df,
+                                lapack_complex_double* duf,
+                                lapack_complex_double* du2, lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sgttrf_work( lapack_int n, float* dl, float* d, float* du,
+                                float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_dgttrf_work( lapack_int n, double* dl, double* d, double* du,
+                                double* du2, lapack_int* ipiv );
+lapack_int LAPACKE_cgttrf_work( lapack_int n, lapack_complex_float* dl,
+                                lapack_complex_float* d,
+                                lapack_complex_float* du,
+                                lapack_complex_float* du2, lapack_int* ipiv );
+lapack_int LAPACKE_zgttrf_work( lapack_int n, lapack_complex_double* dl,
+                                lapack_complex_double* d,
+                                lapack_complex_double* du,
+                                lapack_complex_double* du2, lapack_int* ipiv );
+
+lapack_int LAPACKE_sgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const float* dl,
+                                const float* d, const float* du,
+                                const float* du2, const lapack_int* ipiv,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const double* dl,
+                                const double* d, const double* du,
+                                const double* du2, const lapack_int* ipiv,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_cgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* dl,
+                                const lapack_complex_float* d,
+                                const lapack_complex_float* du,
+                                const lapack_complex_float* du2,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zgttrs_work( int matrix_order, char trans, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* dl,
+                                const lapack_complex_double* d,
+                                const lapack_complex_double* du,
+                                const lapack_complex_double* du2,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               float* w, lapack_complex_float* z,
+                               lapack_int ldz, lapack_complex_float* work,
+                               float* rwork );
+lapack_int LAPACKE_zhbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               double* w, lapack_complex_double* z,
+                               lapack_int ldz, lapack_complex_double* work,
+                               double* rwork );
+
+lapack_int LAPACKE_chbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* q, lapack_int ldq,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                float* rwork, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_zhbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_complex_double* q, lapack_int ldq,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                double* rwork, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_chbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_complex_float* bb, lapack_int ldbb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                const lapack_complex_double* bb,
+                                lapack_int ldbb, lapack_complex_double* x,
+                                lapack_int ldx, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_chbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               lapack_complex_float* bb, lapack_int ldbb,
+                               float* w, lapack_complex_float* z,
+                               lapack_int ldz, lapack_complex_float* work,
+                               float* rwork );
+lapack_int LAPACKE_zhbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               lapack_complex_double* bb, lapack_int ldbb,
+                               double* w, lapack_complex_double* z,
+                               lapack_int ldz, lapack_complex_double* work,
+                               double* rwork );
+
+lapack_int LAPACKE_chbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* bb, lapack_int ldbb,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_complex_double* bb, lapack_int ldbb,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, lapack_complex_float* ab,
+                                lapack_int ldab, lapack_complex_float* bb,
+                                lapack_int ldbb, lapack_complex_float* q,
+                                lapack_int ldq, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* bb,
+                                lapack_int ldbb, lapack_complex_double* q,
+                                lapack_int ldq, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                float* d, float* e, lapack_complex_float* q,
+                                lapack_int ldq, lapack_complex_float* work );
+lapack_int LAPACKE_zhbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                double* d, double* e, lapack_complex_double* q,
+                                lapack_int ldq, lapack_complex_double* work );
+
+lapack_int LAPACKE_checon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zhecon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_cheequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* s, float* scond, float* amax,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_zheequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* s, double* scond, double* amax,
+                                 lapack_complex_double* work );
+
+lapack_int LAPACKE_cheev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, float* w,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zheev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_double* a,
+                               lapack_int lda, double* w,
+                               lapack_complex_double* work, lapack_int lwork,
+                               double* rwork );
+
+lapack_int LAPACKE_cheevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, float* w,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_zheevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, double* w,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_cheevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_int* isuppz,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_zheevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_int* isuppz,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_cheevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zheevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chegst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zhegst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_chegv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* b,
+                               lapack_int ldb, float* w,
+                               lapack_complex_float* work, lapack_int lwork,
+                               float* rwork );
+lapack_int LAPACKE_zhegv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n,
+                               lapack_complex_double* a, lapack_int lda,
+                               lapack_complex_double* b, lapack_int ldb,
+                               double* w, lapack_complex_double* work,
+                               lapack_int lwork, double* rwork );
+
+lapack_int LAPACKE_chegvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float* w, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhegvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double* w, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chegvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhegvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_cherfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zherfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_cherfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* s, const lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zherfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* s,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_chesv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zhesv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_chesvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                lapack_int* ipiv, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zhesvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_chesvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* s,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zhesvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* s,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_chetrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                float* d, float* e, lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zhetrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                double* d, double* e,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_chetrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_zhetrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_chetri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zhetri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_chetrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zhetrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_chfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               float alpha, const lapack_complex_float* a,
+                               lapack_int lda, float beta,
+                               lapack_complex_float* c );
+lapack_int LAPACKE_zhfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               double alpha, const lapack_complex_double* a,
+                               lapack_int lda, double beta,
+                               lapack_complex_double* c );
+
+lapack_int LAPACKE_shgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, float* h, lapack_int ldh,
+                                float* t, lapack_int ldt, float* alphar,
+                                float* alphai, float* beta, float* q,
+                                lapack_int ldq, float* z, lapack_int ldz,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dhgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, double* h, lapack_int ldh,
+                                double* t, lapack_int ldt, double* alphar,
+                                double* alphai, double* beta, double* q,
+                                lapack_int ldq, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_chgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_float* h,
+                                lapack_int ldh, lapack_complex_float* t,
+                                lapack_int ldt, lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork );
+lapack_int LAPACKE_zhgeqz_work( int matrix_order, char job, char compq,
+                                char compz, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_double* h,
+                                lapack_int ldh, lapack_complex_double* t,
+                                lapack_int ldt, lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_chpcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zhpcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_chpev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_float* ap, float* w,
+                               lapack_complex_float* z, lapack_int ldz,
+                               lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhpev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_complex_double* ap,
+                               double* w, lapack_complex_double* z,
+                               lapack_int ldz, lapack_complex_double* work,
+                               double* rwork );
+
+lapack_int LAPACKE_chpevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_float* ap,
+                                float* w, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zhpevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_complex_double* ap,
+                                double* w, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_chpevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* ap, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhpevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* ap, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chpgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_float* ap,
+                                const lapack_complex_float* bp );
+lapack_int LAPACKE_zhpgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, lapack_complex_double* ap,
+                                const lapack_complex_double* bp );
+
+lapack_int LAPACKE_chpgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n,
+                               lapack_complex_float* ap,
+                               lapack_complex_float* bp, float* w,
+                               lapack_complex_float* z, lapack_int ldz,
+                               lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhpgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n,
+                               lapack_complex_double* ap,
+                               lapack_complex_double* bp, double* w,
+                               lapack_complex_double* z, lapack_int ldz,
+                               lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_chpgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                lapack_complex_float* bp, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_zhpgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                lapack_complex_double* bp, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_chpgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                lapack_complex_float* bp, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_zhpgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                lapack_complex_double* bp, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_chprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_complex_float* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_chpsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* ap,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zhpsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* ap,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+
+lapack_int LAPACKE_chpsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* afp, lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zhpsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* afp, lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_chptrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap, float* d, float* e,
+                                lapack_complex_float* tau );
+lapack_int LAPACKE_zhptrd_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap, double* d, double* e,
+                                lapack_complex_double* tau );
+
+lapack_int LAPACKE_chptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zhptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_chptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zhptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_chptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zhptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_shsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, lapack_logical* select,
+                                lapack_int n, const float* h, lapack_int ldh,
+                                float* wr, const float* wi, float* vl,
+                                lapack_int ldvl, float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, float* work,
+                                lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_dhsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, lapack_logical* select,
+                                lapack_int n, const double* h, lapack_int ldh,
+                                double* wr, const double* wi, double* vl,
+                                lapack_int ldvl, double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, double* work,
+                                lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_chsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, const lapack_logical* select,
+                                lapack_int n, const lapack_complex_float* h,
+                                lapack_int ldh, lapack_complex_float* w,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_float* work, float* rwork,
+                                lapack_int* ifaill, lapack_int* ifailr );
+lapack_int LAPACKE_zhsein_work( int matrix_order, char job, char eigsrc,
+                                char initv, const lapack_logical* select,
+                                lapack_int n, const lapack_complex_double* h,
+                                lapack_int ldh, lapack_complex_double* w,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, double* rwork,
+                                lapack_int* ifaill, lapack_int* ifailr );
+
+lapack_int LAPACKE_shseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                float* h, lapack_int ldh, float* wr, float* wi,
+                                float* z, lapack_int ldz, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dhseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                double* h, lapack_int ldh, double* wr,
+                                double* wi, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_chseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_float* h, lapack_int ldh,
+                                lapack_complex_float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zhseqr_work( int matrix_order, char job, char compz,
+                                lapack_int n, lapack_int ilo, lapack_int ihi,
+                                lapack_complex_double* h, lapack_int ldh,
+                                lapack_complex_double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_clacgv_work( lapack_int n, lapack_complex_float* x,
+                                lapack_int incx );
+lapack_int LAPACKE_zlacgv_work( lapack_int n, lapack_complex_double* x,
+                                lapack_int incx );
+
+lapack_int LAPACKE_slacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dlacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_clacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zlacpy_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_zlag2c_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_float* sa, lapack_int ldsa );
+
+lapack_int LAPACKE_slag2d_work( int matrix_order, lapack_int m, lapack_int n,
+                                const float* sa, lapack_int ldsa, double* a,
+                                lapack_int lda );
+
+lapack_int LAPACKE_dlag2s_work( int matrix_order, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda, float* sa,
+                                lapack_int ldsa );
+
+lapack_int LAPACKE_clag2z_work( int matrix_order, lapack_int m, lapack_int n,
+                                const lapack_complex_float* sa, lapack_int ldsa,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* d,
+                                float* a, lapack_int lda, lapack_int* iseed,
+                                float* work );
+lapack_int LAPACKE_dlagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* d,
+                                double* a, lapack_int lda, lapack_int* iseed,
+                                double* work );
+lapack_int LAPACKE_clagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const float* d,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* iseed, lapack_complex_float* work );
+lapack_int LAPACKE_zlagge_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int kl, lapack_int ku, const double* d,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* iseed,
+                                lapack_complex_double* work );
+                                
+lapack_int LAPACKE_claghe_work( int matrix_order, lapack_int n, lapack_int k,
+                                const float* d, lapack_complex_float* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zlaghe_work( int matrix_order, lapack_int n, lapack_int k,
+                                const double* d, lapack_complex_double* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_slagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const float* d, float* a, lapack_int lda,
+                                lapack_int* iseed, float* work );
+lapack_int LAPACKE_dlagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const double* d, double* a, lapack_int lda,
+                                lapack_int* iseed, double* work );
+lapack_int LAPACKE_clagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const float* d, lapack_complex_float* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zlagsy_work( int matrix_order, lapack_int n, lapack_int k,
+                                const double* d, lapack_complex_double* a,
+                                lapack_int lda, lapack_int* iseed,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_slapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n, float* x,
+                                lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_dlapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n, double* x,
+                                lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_clapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_float* x, lapack_int ldx,
+                                lapack_int* k );
+lapack_int LAPACKE_zlapmr_work( int matrix_order, lapack_logical forwrd,
+                                lapack_int m, lapack_int n,
+                                lapack_complex_double* x, lapack_int ldx,
+                                lapack_int* k );
+
+lapack_int LAPACKE_slartgp_work( float f, float g, float* cs, float* sn,
+                                 float* r );
+lapack_int LAPACKE_dlartgp_work( double f, double g, double* cs, double* sn,
+                                 double* r );
+
+lapack_int LAPACKE_slartgs_work( float x, float y, float sigma, float* cs,
+                                 float* sn );
+lapack_int LAPACKE_dlartgs_work( double x, double y, double sigma, double* cs,
+                                 double* sn );
+                                
+float LAPACKE_slapy2_work( float x, float y );
+double LAPACKE_dlapy2_work( double x, double y );
+
+float LAPACKE_slapy3_work( float x, float y, float z );
+double LAPACKE_dlapy3_work( double x, double y, double z );
+
+float LAPACKE_slamch_work( char cmach );
+double LAPACKE_dlamch_work( char cmach );
+
+float LAPACKE_slange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* work );
+double LAPACKE_dlange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* work );
+float LAPACKE_clange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_zlange_work( int matrix_order, char norm, lapack_int m,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, double* work );
+
+float LAPACKE_clanhe_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_zlanhe_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, double* work );
+
+float LAPACKE_slansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* work );
+double LAPACKE_dlansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* work );
+float LAPACKE_clansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_zlansy_work( int matrix_order, char norm, char uplo,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, double* work );
+
+float LAPACKE_slantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n, const float* a,
+                                lapack_int lda, float* work );
+double LAPACKE_dlantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda, double* work );
+float LAPACKE_clantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* work );
+double LAPACKE_zlantr_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* work );
+
+lapack_int LAPACKE_slarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, const float* v,
+                                lapack_int ldv, const float* t, lapack_int ldt,
+                                float* c, lapack_int ldc, float* work,
+                                lapack_int ldwork );
+lapack_int LAPACKE_dlarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, const double* v,
+                                lapack_int ldv, const double* t, lapack_int ldt,
+                                double* c, lapack_int ldc, double* work,
+                                lapack_int ldwork );
+lapack_int LAPACKE_clarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_float* v, lapack_int ldv,
+                                const lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int ldwork );
+lapack_int LAPACKE_zlarfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_double* v, lapack_int ldv,
+                                const lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work,
+                                lapack_int ldwork );
+
+lapack_int LAPACKE_slarfg_work( lapack_int n, float* alpha, float* x,
+                                lapack_int incx, float* tau );
+lapack_int LAPACKE_dlarfg_work( lapack_int n, double* alpha, double* x,
+                                lapack_int incx, double* tau );
+lapack_int LAPACKE_clarfg_work( lapack_int n, lapack_complex_float* alpha,
+                                lapack_complex_float* x, lapack_int incx,
+                                lapack_complex_float* tau );
+lapack_int LAPACKE_zlarfg_work( lapack_int n, lapack_complex_double* alpha,
+                                lapack_complex_double* x, lapack_int incx,
+                                lapack_complex_double* tau );
+
+lapack_int LAPACKE_slarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k, const float* v,
+                                lapack_int ldv, const float* tau, float* t,
+                                lapack_int ldt );
+lapack_int LAPACKE_dlarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k, const double* v,
+                                lapack_int ldv, const double* tau, double* t,
+                                lapack_int ldt );
+lapack_int LAPACKE_clarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_float* v, lapack_int ldv,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zlarft_work( int matrix_order, char direct, char storev,
+                                lapack_int n, lapack_int k,
+                                const lapack_complex_double* v, lapack_int ldv,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_slarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const float* v, float tau,
+                                float* c, lapack_int ldc, float* work );
+lapack_int LAPACKE_dlarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const double* v, double tau,
+                                double* c, lapack_int ldc, double* work );
+lapack_int LAPACKE_clarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const lapack_complex_float* v,
+                                lapack_complex_float tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zlarfx_work( int matrix_order, char side, lapack_int m,
+                                lapack_int n, const lapack_complex_double* v,
+                                lapack_complex_double tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_slarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, float* x );
+lapack_int LAPACKE_dlarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, double* x );
+lapack_int LAPACKE_clarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, lapack_complex_float* x );
+lapack_int LAPACKE_zlarnv_work( lapack_int idist, lapack_int* iseed,
+                                lapack_int n, lapack_complex_double* x );
+
+lapack_int LAPACKE_slaset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, float alpha, float beta, float* a,
+                                lapack_int lda );
+lapack_int LAPACKE_dlaset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, double alpha, double beta,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_claset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, lapack_complex_float alpha,
+                                lapack_complex_float beta,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlaset_work( int matrix_order, char uplo, lapack_int m,
+                                lapack_int n, lapack_complex_double alpha,
+                                lapack_complex_double beta,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_slasrt_work( char id, lapack_int n, float* d );
+lapack_int LAPACKE_dlasrt_work( char id, lapack_int n, double* d );
+
+lapack_int LAPACKE_slaswp_work( int matrix_order, lapack_int n, float* a,
+                                lapack_int lda, lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_dlaswp_work( int matrix_order, lapack_int n, double* a,
+                                lapack_int lda, lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_claswp_work( int matrix_order, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+lapack_int LAPACKE_zlaswp_work( int matrix_order, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int k1, lapack_int k2,
+                                const lapack_int* ipiv, lapack_int incx );
+
+lapack_int LAPACKE_slatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                float* d, lapack_int mode, float cond,
+                                float dmax, lapack_int kl, lapack_int ku,
+                                char pack, float* a, lapack_int lda,
+                                float* work );
+lapack_int LAPACKE_dlatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                double* d, lapack_int mode, double cond,
+                                double dmax, lapack_int kl, lapack_int ku,
+                                char pack, double* a, lapack_int lda,
+                                double* work );
+lapack_int LAPACKE_clatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                float* d, lapack_int mode, float cond,
+                                float dmax, lapack_int kl, lapack_int ku,
+                                char pack, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* work );
+lapack_int LAPACKE_zlatms_work( int matrix_order, lapack_int m, lapack_int n,
+                                char dist, lapack_int* iseed, char sym,
+                                double* d, lapack_int mode, double cond,
+                                double dmax, lapack_int kl, lapack_int ku,
+                                char pack, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* work );
+
+lapack_int LAPACKE_slauum_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda );
+lapack_int LAPACKE_dlauum_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_clauum_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zlauum_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_sopgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, const float* tau, float* q,
+                                lapack_int ldq, float* work );
+lapack_int LAPACKE_dopgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, const double* tau, double* q,
+                                lapack_int ldq, double* work );
+
+lapack_int LAPACKE_sopmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const float* ap, const float* tau, float* c,
+                                lapack_int ldc, float* work );
+lapack_int LAPACKE_dopmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const double* ap, const double* tau, double* c,
+                                lapack_int ldc, double* work );
+
+lapack_int LAPACKE_sorgbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k, float* a,
+                                lapack_int lda, const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k, double* a,
+                                lapack_int lda, const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, float* a, lapack_int lda,
+                                const float* tau, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorgrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, double* a, lapack_int lda,
+                                const double* tau, double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_sorgtr_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, const float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dorgtr_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, const double* tau,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_sormtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const float* a, lapack_int lda,
+                                const float* tau, float* c, lapack_int ldc,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dormtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const double* tau, double* c, lapack_int ldc,
+                                double* work, lapack_int lwork );
+
+lapack_int LAPACKE_spbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const float* ab, lapack_int ldab,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dpbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const double* ab,
+                                lapack_int ldab, double anorm, double* rcond,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cpbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_float* ab,
+                                lapack_int ldab, float anorm, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpbcon_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_double* ab,
+                                lapack_int ldab, double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const float* ab, lapack_int ldab,
+                                float* s, float* scond, float* amax );
+lapack_int LAPACKE_dpbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const double* ab,
+                                lapack_int ldab, double* s, double* scond,
+                                double* amax );
+lapack_int LAPACKE_cpbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_float* ab,
+                                lapack_int ldab, float* s, float* scond,
+                                float* amax );
+lapack_int LAPACKE_zpbequ_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, const lapack_complex_double* ab,
+                                lapack_int ldab, double* s, double* scond,
+                                double* amax );
+
+lapack_int LAPACKE_spbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs, const float* ab,
+                                lapack_int ldab, const float* afb,
+                                lapack_int ldafb, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dpbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const double* ab, lapack_int ldab,
+                                const double* afb, lapack_int ldafb,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cpbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                const lapack_complex_float* afb,
+                                lapack_int ldafb, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpbrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab,
+                                const lapack_complex_double* afb,
+                                lapack_int ldafb,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, float* bb, lapack_int ldbb );
+lapack_int LAPACKE_dpbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, double* bb, lapack_int ldbb );
+lapack_int LAPACKE_cpbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, lapack_complex_float* bb,
+                                lapack_int ldbb );
+lapack_int LAPACKE_zpbstf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kb, lapack_complex_double* bb,
+                                lapack_int ldbb );
+
+lapack_int LAPACKE_spbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs, float* ab,
+                               lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs, double* ab,
+                               lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs,
+                               lapack_complex_float* ab, lapack_int ldab,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int kd, lapack_int nrhs,
+                               lapack_complex_double* ab, lapack_int ldab,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                float* ab, lapack_int ldab, float* afb,
+                                lapack_int ldafb, char* equed, float* s,
+                                float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dpbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                double* ab, lapack_int ldab, double* afb,
+                                lapack_int ldafb, char* equed, double* s,
+                                double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, double* work, lapack_int* iwork );
+lapack_int LAPACKE_cpbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* afb, lapack_int ldafb,
+                                char* equed, float* s, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zpbsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int kd, lapack_int nrhs,
+                                lapack_complex_double* ab, lapack_int ldab,
+                                lapack_complex_double* afb, lapack_int ldafb,
+                                char* equed, double* s,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, float* ab, lapack_int ldab );
+lapack_int LAPACKE_dpbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, double* ab, lapack_int ldab );
+lapack_int LAPACKE_cpbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_complex_float* ab,
+                                lapack_int ldab );
+lapack_int LAPACKE_zpbtrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_complex_double* ab,
+                                lapack_int ldab );
+
+lapack_int LAPACKE_spbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs, const float* ab,
+                                lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dpbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const double* ab, lapack_int ldab, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cpbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpbtrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int kd, lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_spftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, float* a );
+lapack_int LAPACKE_dpftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, double* a );
+lapack_int LAPACKE_cpftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftrf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, float* a );
+lapack_int LAPACKE_dpftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, double* a );
+lapack_int LAPACKE_cpftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_float* a );
+lapack_int LAPACKE_zpftri_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_complex_double* a );
+
+lapack_int LAPACKE_spftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs, const float* a,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dpftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs, const double* a,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_cpftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpftrs_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spocon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* a, lapack_int lda, float anorm,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dpocon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* a, lapack_int lda, double anorm,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cpocon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float anorm, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpocon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double anorm, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spoequ_work( int matrix_order, lapack_int n, const float* a,
+                                lapack_int lda, float* s, float* scond,
+                                float* amax );
+lapack_int LAPACKE_dpoequ_work( int matrix_order, lapack_int n, const double* a,
+                                lapack_int lda, double* s, double* scond,
+                                double* amax );
+lapack_int LAPACKE_cpoequ_work( int matrix_order, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequ_work( int matrix_order, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_spoequb_work( int matrix_order, lapack_int n, const float* a,
+                                 lapack_int lda, float* s, float* scond,
+                                 float* amax );
+lapack_int LAPACKE_dpoequb_work( int matrix_order, lapack_int n,
+                                 const double* a, lapack_int lda, double* s,
+                                 double* scond, double* amax );
+lapack_int LAPACKE_cpoequb_work( int matrix_order, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* s, float* scond, float* amax );
+lapack_int LAPACKE_zpoequb_work( int matrix_order, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* s, double* scond, double* amax );
+
+lapack_int LAPACKE_sporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const float* af, lapack_int ldaf,
+                                const float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const double* af,
+                                lapack_int ldaf, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zporfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const float* a,
+                                 lapack_int lda, const float* af,
+                                 lapack_int ldaf, const float* s,
+                                 const float* b, lapack_int ldb, float* x,
+                                 lapack_int ldx, float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const double* a,
+                                 lapack_int lda, const double* af,
+                                 lapack_int ldaf, const double* s,
+                                 const double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const float* s,
+                                 const lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zporfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const double* s,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_sposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* a, lapack_int lda,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* a, lapack_int lda,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_cposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* a,
+                               lapack_int lda, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zposv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* a,
+                               lapack_int lda, lapack_complex_double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dsposv_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, double* a, lapack_int lda,
+                                double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* work, float* swork,
+                                lapack_int* iter );
+lapack_int LAPACKE_zcposv_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, lapack_complex_double* work,
+                                lapack_complex_float* swork, double* rwork,
+                                lapack_int* iter );
+
+lapack_int LAPACKE_sposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, float* a,
+                                lapack_int lda, float* af, lapack_int ldaf,
+                                char* equed, float* s, float* b, lapack_int ldb,
+                                float* x, lapack_int ldx, float* rcond,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, double* a,
+                                lapack_int lda, double* af, lapack_int ldaf,
+                                char* equed, double* s, double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                char* equed, float* s, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zposvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                char* equed, double* s,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, float* a,
+                                 lapack_int lda, float* af, lapack_int ldaf,
+                                 char* equed, float* s, float* b,
+                                 lapack_int ldb, float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, double* a,
+                                 lapack_int lda, double* af, lapack_int ldaf,
+                                 char* equed, double* s, double* b,
+                                 lapack_int ldb, double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_cposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 char* equed, float* s, lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params,
+                                 lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zposvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 char* equed, double* s,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_spotrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda );
+lapack_int LAPACKE_dpotrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_cpotrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotri_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda );
+lapack_int LAPACKE_dpotri_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda );
+lapack_int LAPACKE_cpotri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zpotri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_spotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                float* b, lapack_int ldb );
+lapack_int LAPACKE_dpotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, double* b, lapack_int ldb );
+lapack_int LAPACKE_cpotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zpotrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_sppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, float anorm, float* rcond,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, double anorm, double* rcond,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap, float anorm,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zppcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap, double anorm,
+                                double* rcond, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_sppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, float* s, float* scond,
+                                float* amax );
+lapack_int LAPACKE_dppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, double* s, double* scond,
+                                double* amax );
+lapack_int LAPACKE_cppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap, float* s,
+                                float* scond, float* amax );
+lapack_int LAPACKE_zppequ_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap, double* s,
+                                double* scond, double* amax );
+
+lapack_int LAPACKE_spprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap,
+                                const float* afp, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dpprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap,
+                                const double* afp, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cpprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_complex_float* afp,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zpprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* afp,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* ap, float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_dppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* ap, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* ap,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zppsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* ap,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, float* ap,
+                                float* afp, char* equed, float* s, float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, double* ap,
+                                double* afp, char* equed, double* s, double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_float* ap,
+                                lapack_complex_float* afp, char* equed,
+                                float* s, lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_zppsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                lapack_complex_double* ap,
+                                lapack_complex_double* afp, char* equed,
+                                double* s, lapack_complex_double* b,
+                                lapack_int ldb, lapack_complex_double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_spptrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap );
+lapack_int LAPACKE_dpptrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap );
+lapack_int LAPACKE_cpptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_zpptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptri_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap );
+lapack_int LAPACKE_dpptri_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap );
+lapack_int LAPACKE_cpptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_zpptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_spptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dpptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cpptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_spstrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* piv,
+                                lapack_int* rank, float tol, float* work );
+lapack_int LAPACKE_dpstrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* piv,
+                                lapack_int* rank, double tol, double* work );
+lapack_int LAPACKE_cpstrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* piv, lapack_int* rank, float tol,
+                                float* work );
+lapack_int LAPACKE_zpstrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* piv, lapack_int* rank, double tol,
+                                double* work );
+
+lapack_int LAPACKE_sptcon_work( lapack_int n, const float* d, const float* e,
+                                float anorm, float* rcond, float* work );
+lapack_int LAPACKE_dptcon_work( lapack_int n, const double* d, const double* e,
+                                double anorm, double* rcond, double* work );
+lapack_int LAPACKE_cptcon_work( lapack_int n, const float* d,
+                                const lapack_complex_float* e, float anorm,
+                                float* rcond, float* work );
+lapack_int LAPACKE_zptcon_work( lapack_int n, const double* d,
+                                const lapack_complex_double* e, double anorm,
+                                double* rcond, double* work );
+
+lapack_int LAPACKE_spteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work );
+lapack_int LAPACKE_dpteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work );
+lapack_int LAPACKE_cpteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, lapack_complex_float* z,
+                                lapack_int ldz, float* work );
+lapack_int LAPACKE_zpteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, lapack_complex_double* z,
+                                lapack_int ldz, double* work );
+
+lapack_int LAPACKE_sptrfs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const float* d, const float* e, const float* df,
+                                const float* ef, const float* b, lapack_int ldb,
+                                float* x, lapack_int ldx, float* ferr,
+                                float* berr, float* work );
+lapack_int LAPACKE_dptrfs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const double* d, const double* e,
+                                const double* df, const double* ef,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work );
+lapack_int LAPACKE_cptrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* d,
+                                const lapack_complex_float* e, const float* df,
+                                const lapack_complex_float* ef,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zptrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const lapack_complex_double* e,
+                                const double* df,
+                                const lapack_complex_double* ef,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* d, float* e, float* b, lapack_int ldb );
+lapack_int LAPACKE_dptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* d, double* e, double* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_cptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               float* d, lapack_complex_float* e,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zptsv_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                               double* d, lapack_complex_double* e,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const float* d, const float* e,
+                                float* df, float* ef, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work );
+lapack_int LAPACKE_dptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const double* e, double* df, double* ef,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* rcond, double* ferr,
+                                double* berr, double* work );
+lapack_int LAPACKE_cptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const float* d,
+                                const lapack_complex_float* e, float* df,
+                                lapack_complex_float* ef,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zptsvx_work( int matrix_order, char fact, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const lapack_complex_double* e, double* df,
+                                lapack_complex_double* ef,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_spttrf_work( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dpttrf_work( lapack_int n, double* d, double* e );
+lapack_int LAPACKE_cpttrf_work( lapack_int n, float* d,
+                                lapack_complex_float* e );
+lapack_int LAPACKE_zpttrf_work( lapack_int n, double* d,
+                                lapack_complex_double* e );
+
+lapack_int LAPACKE_spttrs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const float* d, const float* e, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dpttrs_work( int matrix_order, lapack_int n, lapack_int nrhs,
+                                const double* d, const double* e, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_cpttrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* d,
+                                const lapack_complex_float* e,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zpttrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* d,
+                                const lapack_complex_double* e,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd, float* ab,
+                               lapack_int ldab, float* w, float* z,
+                               lapack_int ldz, float* work );
+lapack_int LAPACKE_dsbev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int kd, double* ab,
+                               lapack_int ldab, double* w, double* z,
+                               lapack_int ldz, double* work );
+
+lapack_int LAPACKE_ssbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd, float* ab,
+                                lapack_int ldab, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsbevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int kd, double* ab,
+                                lapack_int ldab, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                float* ab, lapack_int ldab, float* q,
+                                lapack_int ldq, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dsbevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int kd,
+                                double* ab, lapack_int ldab, double* q,
+                                lapack_int ldq, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_ssbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                float* ab, lapack_int ldab, const float* bb,
+                                lapack_int ldbb, float* x, lapack_int ldx,
+                                float* work );
+lapack_int LAPACKE_dsbgst_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                double* ab, lapack_int ldab, const double* bb,
+                                lapack_int ldbb, double* x, lapack_int ldx,
+                                double* work );
+
+lapack_int LAPACKE_ssbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               float* ab, lapack_int ldab, float* bb,
+                               lapack_int ldbb, float* w, float* z,
+                               lapack_int ldz, float* work );
+lapack_int LAPACKE_dsbgv_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, lapack_int ka, lapack_int kb,
+                               double* ab, lapack_int ldab, double* bb,
+                               lapack_int ldbb, double* w, double* z,
+                               lapack_int ldz, double* work );
+
+lapack_int LAPACKE_ssbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                float* ab, lapack_int ldab, float* bb,
+                                lapack_int ldbb, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsbgvd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, lapack_int ka, lapack_int kb,
+                                double* ab, lapack_int ldab, double* bb,
+                                lapack_int ldbb, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, float* ab, lapack_int ldab,
+                                float* bb, lapack_int ldbb, float* q,
+                                lapack_int ldq, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dsbgvx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, lapack_int ka,
+                                lapack_int kb, double* ab, lapack_int ldab,
+                                double* bb, lapack_int ldbb, double* q,
+                                lapack_int ldq, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_ssbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd, float* ab,
+                                lapack_int ldab, float* d, float* e, float* q,
+                                lapack_int ldq, float* work );
+lapack_int LAPACKE_dsbtrd_work( int matrix_order, char vect, char uplo,
+                                lapack_int n, lapack_int kd, double* ab,
+                                lapack_int ldab, double* d, double* e,
+                                double* q, lapack_int ldq, double* work );
+
+lapack_int LAPACKE_ssfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               float alpha, const float* a, lapack_int lda,
+                               float beta, float* c );
+lapack_int LAPACKE_dsfrk_work( int matrix_order, char transr, char uplo,
+                               char trans, lapack_int n, lapack_int k,
+                               double alpha, const double* a, lapack_int lda,
+                               double beta, double* c );
+
+lapack_int LAPACKE_sspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, const lapack_int* ipiv,
+                                float anorm, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, const lapack_int* ipiv,
+                                double anorm, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_cspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zspcon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_sspev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, float* ap, float* w, float* z,
+                               lapack_int ldz, float* work );
+lapack_int LAPACKE_dspev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, double* ap, double* w, double* z,
+                               lapack_int ldz, double* work );
+
+lapack_int LAPACKE_sspevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, float* ap, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dspevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, double* ap, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sspevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, float* ap, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dspevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, double* ap, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, double* work,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_sspgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, float* ap, const float* bp );
+lapack_int LAPACKE_dspgst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, double* ap, const double* bp );
+
+lapack_int LAPACKE_sspgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, float* ap, float* bp,
+                               float* w, float* z, lapack_int ldz,
+                               float* work );
+lapack_int LAPACKE_dspgv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, double* ap, double* bp,
+                               double* w, double* z, lapack_int ldz,
+                               double* work );
+
+lapack_int LAPACKE_sspgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, float* ap, float* bp,
+                                float* w, float* z, lapack_int ldz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dspgvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, double* ap, double* bp,
+                                double* w, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sspgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, float* ap,
+                                float* bp, float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, float* z, lapack_int ldz, float* work,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_dspgvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, double* ap,
+                                double* bp, double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, double* z, lapack_int ldz,
+                                double* work, lapack_int* iwork,
+                                lapack_int* ifail );
+
+lapack_int LAPACKE_ssprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap,
+                                const float* afp, const lapack_int* ipiv,
+                                const float* b, lapack_int ldb, float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dsprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap,
+                                const double* afp, const lapack_int* ipiv,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_csprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_complex_float* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zsprfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* afp,
+                                const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_sspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* ap, lapack_int* ipiv,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* ap, lapack_int* ipiv,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_cspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* ap,
+                               lapack_int* ipiv, lapack_complex_float* b,
+                               lapack_int ldb );
+lapack_int LAPACKE_zspsv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* ap,
+                               lapack_int* ipiv, lapack_complex_double* b,
+                               lapack_int ldb );
+
+lapack_int LAPACKE_sspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const float* ap,
+                                float* afp, lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const double* ap,
+                                double* afp, lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_cspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* afp, lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zspsvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* afp, lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_ssptrd_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap, float* d, float* e, float* tau );
+lapack_int LAPACKE_dsptrd_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap, double* d, double* e, double* tau );
+
+lapack_int LAPACKE_ssptrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_dsptrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap, lapack_int* ipiv );
+lapack_int LAPACKE_csptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap, lapack_int* ipiv );
+lapack_int LAPACKE_zsptrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap, lapack_int* ipiv );
+
+lapack_int LAPACKE_ssptri_work( int matrix_order, char uplo, lapack_int n,
+                                float* ap, const lapack_int* ipiv,
+                                float* work );
+lapack_int LAPACKE_dsptri_work( int matrix_order, char uplo, lapack_int n,
+                                double* ap, const lapack_int* ipiv,
+                                double* work );
+lapack_int LAPACKE_csptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zsptri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_ssptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* ap,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dsptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* ap,
+                                const lapack_int* ipiv, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_csptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* ap,
+                                const lapack_int* ipiv, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_zsptrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_sstebz_work( char range, char order, lapack_int n, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, const float* d, const float* e,
+                                lapack_int* m, lapack_int* nsplit, float* w,
+                                lapack_int* iblock, lapack_int* isplit,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dstebz_work( char range, char order, lapack_int n, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, const double* d, const double* e,
+                                lapack_int* m, lapack_int* nsplit, double* w,
+                                lapack_int* iblock, lapack_int* isplit,
+                                double* work, lapack_int* iwork );
+
+lapack_int LAPACKE_sstedc_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dstedc_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_cstedc_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, lapack_complex_float* z,
+                                lapack_int ldz, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zstedc_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, lapack_complex_double* z,
+                                lapack_int ldz, lapack_complex_double* work,
+                                lapack_int lwork, double* rwork,
+                                lapack_int lrwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_sstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, lapack_int* isuppz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_cstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zstegr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int* isuppz, double* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_sstein_work( int matrix_order, lapack_int n, const float* d,
+                                const float* e, lapack_int m, const float* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+lapack_int LAPACKE_dstein_work( int matrix_order, lapack_int n, const double* d,
+                                const double* e, lapack_int m, const double* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit, double* z,
+                                lapack_int ldz, double* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+lapack_int LAPACKE_cstein_work( int matrix_order, lapack_int n, const float* d,
+                                const float* e, lapack_int m, const float* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit,
+                                lapack_complex_float* z, lapack_int ldz,
+                                float* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+lapack_int LAPACKE_zstein_work( int matrix_order, lapack_int n, const double* d,
+                                const double* e, lapack_int m, const double* w,
+                                const lapack_int* iblock,
+                                const lapack_int* isplit,
+                                lapack_complex_double* z, lapack_int ldz,
+                                double* work, lapack_int* iwork,
+                                lapack_int* ifailv );
+
+lapack_int LAPACKE_sstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int nzc,
+                                lapack_int* isuppz, lapack_logical* tryrac,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, lapack_int nzc,
+                                lapack_int* isuppz, lapack_logical* tryrac,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_cstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, float* w,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int nzc, lapack_int* isuppz,
+                                lapack_logical* tryrac, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_zstemr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                lapack_int* m, double* w,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int nzc, lapack_int* isuppz,
+                                lapack_logical* tryrac, double* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_ssteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work );
+lapack_int LAPACKE_dsteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work );
+lapack_int LAPACKE_csteqr_work( int matrix_order, char compz, lapack_int n,
+                                float* d, float* e, lapack_complex_float* z,
+                                lapack_int ldz, float* work );
+lapack_int LAPACKE_zsteqr_work( int matrix_order, char compz, lapack_int n,
+                                double* d, double* e, lapack_complex_double* z,
+                                lapack_int ldz, double* work );
+
+lapack_int LAPACKE_ssterf_work( lapack_int n, float* d, float* e );
+lapack_int LAPACKE_dsterf_work( lapack_int n, double* d, double* e );
+
+lapack_int LAPACKE_sstev_work( int matrix_order, char jobz, lapack_int n,
+                               float* d, float* e, float* z, lapack_int ldz,
+                               float* work );
+lapack_int LAPACKE_dstev_work( int matrix_order, char jobz, lapack_int n,
+                               double* d, double* e, double* z, lapack_int ldz,
+                               double* work );
+
+lapack_int LAPACKE_sstevd_work( int matrix_order, char jobz, lapack_int n,
+                                float* d, float* e, float* z, lapack_int ldz,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dstevd_work( int matrix_order, char jobz, lapack_int n,
+                                double* d, double* e, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sstevr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dstevr_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, lapack_int* isuppz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_sstevx_work( int matrix_order, char jobz, char range,
+                                lapack_int n, float* d, float* e, float vl,
+                                float vu, lapack_int il, lapack_int iu,
+                                float abstol, lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dstevx_work( int matrix_order, char jobz, char range,
+                                lapack_int n, double* d, double* e, double vl,
+                                double vu, lapack_int il, lapack_int iu,
+                                double abstol, lapack_int* m, double* w,
+                                double* z, lapack_int ldz, double* work,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_ssycon_work( int matrix_order, char uplo, lapack_int n,
+                                const float* a, lapack_int lda,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dsycon_work( int matrix_order, char uplo, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_csycon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv, float anorm,
+                                float* rcond, lapack_complex_float* work );
+lapack_int LAPACKE_zsycon_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv, double anorm,
+                                double* rcond, lapack_complex_double* work );
+
+lapack_int LAPACKE_ssyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const float* a, lapack_int lda, float* s,
+                                 float* scond, float* amax, float* work );
+lapack_int LAPACKE_dsyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const double* a, lapack_int lda, double* s,
+                                 double* scond, double* amax, double* work );
+lapack_int LAPACKE_csyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 float* s, float* scond, float* amax,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_zsyequb_work( int matrix_order, char uplo, lapack_int n,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 double* s, double* scond, double* amax,
+                                 lapack_complex_double* work );
+
+lapack_int LAPACKE_ssyev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, float* a, lapack_int lda, float* w,
+                               float* work, lapack_int lwork );
+lapack_int LAPACKE_dsyev_work( int matrix_order, char jobz, char uplo,
+                               lapack_int n, double* a, lapack_int lda,
+                               double* w, double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssyevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, float* a, lapack_int lda,
+                                float* w, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsyevd_work( int matrix_order, char jobz, char uplo,
+                                lapack_int n, double* a, lapack_int lda,
+                                double* w, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssyevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, float* a,
+                                lapack_int lda, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, lapack_int* isuppz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dsyevr_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, double* a,
+                                lapack_int lda, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, lapack_int* isuppz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssyevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, float* a,
+                                lapack_int lda, float vl, float vu,
+                                lapack_int il, lapack_int iu, float abstol,
+                                lapack_int* m, float* w, float* z,
+                                lapack_int ldz, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int* ifail );
+lapack_int LAPACKE_dsyevx_work( int matrix_order, char jobz, char range,
+                                char uplo, lapack_int n, double* a,
+                                lapack_int lda, double vl, double vu,
+                                lapack_int il, lapack_int iu, double abstol,
+                                lapack_int* m, double* w, double* z,
+                                lapack_int ldz, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_ssygst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, float* a, lapack_int lda,
+                                const float* b, lapack_int ldb );
+lapack_int LAPACKE_dsygst_work( int matrix_order, lapack_int itype, char uplo,
+                                lapack_int n, double* a, lapack_int lda,
+                                const double* b, lapack_int ldb );
+
+lapack_int LAPACKE_ssygv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, float* a,
+                               lapack_int lda, float* b, lapack_int ldb,
+                               float* w, float* work, lapack_int lwork );
+lapack_int LAPACKE_dsygv_work( int matrix_order, lapack_int itype, char jobz,
+                               char uplo, lapack_int n, double* a,
+                               lapack_int lda, double* b, lapack_int ldb,
+                               double* w, double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssygvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float* w, float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dsygvd_work( int matrix_order, lapack_int itype, char jobz,
+                                char uplo, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* w, double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+
+lapack_int LAPACKE_ssygvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float vl, float vu, lapack_int il,
+                                lapack_int iu, float abstol, lapack_int* m,
+                                float* w, float* z, lapack_int ldz, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int* ifail );
+lapack_int LAPACKE_dsygvx_work( int matrix_order, lapack_int itype, char jobz,
+                                char range, char uplo, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double vl, double vu, lapack_int il,
+                                lapack_int iu, double abstol, lapack_int* m,
+                                double* w, double* z, lapack_int ldz,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int* ifail );
+
+lapack_int LAPACKE_ssyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const float* af, lapack_int ldaf,
+                                const lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dsyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const double* b, lapack_int ldb, double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                double* work, lapack_int* iwork );
+lapack_int LAPACKE_csyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_zsyrfs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_complex_double* af,
+                                lapack_int ldaf, const lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_ssyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const float* a,
+                                 lapack_int lda, const float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* s, const float* b, lapack_int ldb,
+                                 float* x, lapack_int ldx, float* rcond,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dsyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs, const double* a,
+                                 lapack_int lda, const double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* s, const double* b,
+                                 lapack_int ldb, double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, double* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_csyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_float* a, lapack_int lda,
+                                 const lapack_complex_float* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const float* s, const lapack_complex_float* b,
+                                 lapack_int ldb, lapack_complex_float* x,
+                                 lapack_int ldx, float* rcond, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zsyrfsx_work( int matrix_order, char uplo, char equed,
+                                 lapack_int n, lapack_int nrhs,
+                                 const lapack_complex_double* a, lapack_int lda,
+                                 const lapack_complex_double* af,
+                                 lapack_int ldaf, const lapack_int* ipiv,
+                                 const double* s,
+                                 const lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_ssysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, float* a, lapack_int lda,
+                               lapack_int* ipiv, float* b, lapack_int ldb,
+                               float* work, lapack_int lwork );
+lapack_int LAPACKE_dsysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, double* a, lapack_int lda,
+                               lapack_int* ipiv, double* b, lapack_int ldb,
+                               double* work, lapack_int lwork );
+lapack_int LAPACKE_csysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_float* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_float* b, lapack_int ldb,
+                               lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zsysv_work( int matrix_order, char uplo, lapack_int n,
+                               lapack_int nrhs, lapack_complex_double* a,
+                               lapack_int lda, lapack_int* ipiv,
+                               lapack_complex_double* b, lapack_int ldb,
+                               lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const float* a,
+                                lapack_int lda, float* af, lapack_int ldaf,
+                                lapack_int* ipiv, const float* b,
+                                lapack_int ldb, float* x, lapack_int ldx,
+                                float* rcond, float* ferr, float* berr,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dsysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs, const double* a,
+                                lapack_int lda, double* af, lapack_int ldaf,
+                                lapack_int* ipiv, const double* b,
+                                lapack_int ldb, double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_csysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* af, lapack_int ldaf,
+                                lapack_int* ipiv, const lapack_complex_float* b,
+                                lapack_int ldb, lapack_complex_float* x,
+                                lapack_int ldx, float* rcond, float* ferr,
+                                float* berr, lapack_complex_float* work,
+                                lapack_int lwork, float* rwork );
+lapack_int LAPACKE_zsysvx_work( int matrix_order, char fact, char uplo,
+                                lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* af, lapack_int ldaf,
+                                lapack_int* ipiv,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* x, lapack_int ldx,
+                                double* rcond, double* ferr, double* berr,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork );
+
+lapack_int LAPACKE_ssysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, float* a,
+                                 lapack_int lda, float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* s,
+                                 float* b, lapack_int ldb, float* x,
+                                 lapack_int ldx, float* rcond, float* rpvgrw,
+                                 float* berr, lapack_int n_err_bnds,
+                                 float* err_bnds_norm, float* err_bnds_comp,
+                                 lapack_int nparams, float* params, float* work,
+                                 lapack_int* iwork );
+lapack_int LAPACKE_dsysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs, double* a,
+                                 lapack_int lda, double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* s,
+                                 double* b, lapack_int ldb, double* x,
+                                 lapack_int ldx, double* rcond, double* rpvgrw,
+                                 double* berr, lapack_int n_err_bnds,
+                                 double* err_bnds_norm, double* err_bnds_comp,
+                                 lapack_int nparams, double* params,
+                                 double* work, lapack_int* iwork );
+lapack_int LAPACKE_csysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, float* s,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* x, lapack_int ldx,
+                                 float* rcond, float* rpvgrw, float* berr,
+                                 lapack_int n_err_bnds, float* err_bnds_norm,
+                                 float* err_bnds_comp, lapack_int nparams,
+                                 float* params, lapack_complex_float* work,
+                                 float* rwork );
+lapack_int LAPACKE_zsysvxx_work( int matrix_order, char fact, char uplo,
+                                 lapack_int n, lapack_int nrhs,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* af, lapack_int ldaf,
+                                 lapack_int* ipiv, char* equed, double* s,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* x, lapack_int ldx,
+                                 double* rcond, double* rpvgrw, double* berr,
+                                 lapack_int n_err_bnds, double* err_bnds_norm,
+                                 double* err_bnds_comp, lapack_int nparams,
+                                 double* params, lapack_complex_double* work,
+                                 double* rwork );
+
+lapack_int LAPACKE_ssytrd_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, float* d, float* e,
+                                float* tau, float* work, lapack_int lwork );
+lapack_int LAPACKE_dsytrd_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, double* d, double* e,
+                                double* tau, double* work, lapack_int lwork );
+
+lapack_int LAPACKE_ssytrf_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda, lapack_int* ipiv,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dsytrf_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda, lapack_int* ipiv,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_csytrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_zsytrf_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_int* ipiv, lapack_complex_double* work,
+                                lapack_int lwork );
+
+lapack_int LAPACKE_ssytri_work( int matrix_order, char uplo, lapack_int n,
+                                float* a, lapack_int lda,
+                                const lapack_int* ipiv, float* work );
+lapack_int LAPACKE_dsytri_work( int matrix_order, char uplo, lapack_int n,
+                                double* a, lapack_int lda,
+                                const lapack_int* ipiv, double* work );
+lapack_int LAPACKE_csytri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zsytri_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_int* ipiv,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_ssytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const float* a, lapack_int lda,
+                                const lapack_int* ipiv, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dsytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                double* b, lapack_int ldb );
+lapack_int LAPACKE_csytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_zsytrs_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_int nrhs, const lapack_complex_double* a,
+                                lapack_int lda, const lapack_int* ipiv,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const float* ab, lapack_int ldab, float* rcond,
+                                float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const double* ab, lapack_int ldab,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const lapack_complex_float* ab, lapack_int ldab,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_ztbcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, lapack_int kd,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const float* ab,
+                                lapack_int ldab, const float* b, lapack_int ldb,
+                                const float* x, lapack_int ldx, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const double* ab,
+                                lapack_int ldab, const double* b,
+                                lapack_int ldb, const double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const lapack_complex_float* ab,
+                                lapack_int ldab, const lapack_complex_float* b,
+                                lapack_int ldb, const lapack_complex_float* x,
+                                lapack_int ldx, float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztbrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, const lapack_complex_double* b,
+                                lapack_int ldb, const lapack_complex_double* x,
+                                lapack_int ldx, double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const float* ab,
+                                lapack_int ldab, float* b, lapack_int ldb );
+lapack_int LAPACKE_dtbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const double* ab,
+                                lapack_int ldab, double* b, lapack_int ldb );
+lapack_int LAPACKE_ctbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs, const lapack_complex_float* ab,
+                                lapack_int ldab, lapack_complex_float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_ztbtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int kd,
+                                lapack_int nrhs,
+                                const lapack_complex_double* ab,
+                                lapack_int ldab, lapack_complex_double* b,
+                                lapack_int ldb );
+
+lapack_int LAPACKE_stfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, float alpha, const float* a,
+                               float* b, lapack_int ldb );
+lapack_int LAPACKE_dtfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, double alpha, const double* a,
+                               double* b, lapack_int ldb );
+lapack_int LAPACKE_ctfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, lapack_complex_float alpha,
+                               const lapack_complex_float* a,
+                               lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztfsm_work( int matrix_order, char transr, char side,
+                               char uplo, char trans, char diag, lapack_int m,
+                               lapack_int n, lapack_complex_double alpha,
+                               const lapack_complex_double* a,
+                               lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n, float* a );
+lapack_int LAPACKE_dtftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n, double* a );
+lapack_int LAPACKE_ctftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n,
+                                lapack_complex_float* a );
+lapack_int LAPACKE_ztftri_work( int matrix_order, char transr, char uplo,
+                                char diag, lapack_int n,
+                                lapack_complex_double* a );
+
+lapack_int LAPACKE_stfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* arf, float* ap );
+lapack_int LAPACKE_dtfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* arf, double* ap );
+lapack_int LAPACKE_ctfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* arf,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_ztfttp_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* arf,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_stfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* arf, float* a,
+                                lapack_int lda );
+lapack_int LAPACKE_dtfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* arf, double* a,
+                                lapack_int lda );
+lapack_int LAPACKE_ctfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* arf,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztfttr_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* arf,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_stgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const float* s, lapack_int lds, const float* p,
+                                lapack_int ldp, float* vl, lapack_int ldvl,
+                                float* vr, lapack_int ldvr, lapack_int mm,
+                                lapack_int* m, float* work );
+lapack_int LAPACKE_dtgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const double* s, lapack_int lds,
+                                const double* p, lapack_int ldp, double* vl,
+                                lapack_int ldvl, double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, double* work );
+lapack_int LAPACKE_ctgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_float* s, lapack_int lds,
+                                const lapack_complex_float* p, lapack_int ldp,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztgevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_double* s, lapack_int lds,
+                                const lapack_complex_double* p, lapack_int ldp,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n, float* a,
+                                lapack_int lda, float* b, lapack_int ldb,
+                                float* q, lapack_int ldq, float* z,
+                                lapack_int ldz, lapack_int* ifst,
+                                lapack_int* ilst, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dtgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* q, lapack_int ldq, double* z,
+                                lapack_int ldz, lapack_int* ifst,
+                                lapack_int* ilst, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_ctgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztgexc_work( int matrix_order, lapack_logical wantq,
+                                lapack_logical wantz, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_stgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float* alphar, float* alphai,
+                                float* beta, float* q, lapack_int ldq, float* z,
+                                lapack_int ldz, lapack_int* m, float* pl,
+                                float* pr, float* dif, float* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+lapack_int LAPACKE_dtgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double* alphar, double* alphai,
+                                double* beta, double* q, lapack_int ldq,
+                                double* z, lapack_int ldz, lapack_int* m,
+                                double* pl, double* pr, double* dif,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_ctgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* alpha,
+                                lapack_complex_float* beta,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* z, lapack_int ldz,
+                                lapack_int* m, float* pl, float* pr, float* dif,
+                                lapack_complex_float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_ztgsen_work( int matrix_order, lapack_int ijob,
+                                lapack_logical wantq, lapack_logical wantz,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* alpha,
+                                lapack_complex_double* beta,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* z, lapack_int ldz,
+                                lapack_int* m, double* pl, double* pr,
+                                double* dif, lapack_complex_double* work,
+                                lapack_int lwork, lapack_int* iwork,
+                                lapack_int liwork );
+
+lapack_int LAPACKE_stgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                float* a, lapack_int lda, float* b,
+                                lapack_int ldb, float tola, float tolb,
+                                float* alpha, float* beta, float* u,
+                                lapack_int ldu, float* v, lapack_int ldv,
+                                float* q, lapack_int ldq, float* work,
+                                lapack_int* ncycle );
+lapack_int LAPACKE_dtgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                double* a, lapack_int lda, double* b,
+                                lapack_int ldb, double tola, double tolb,
+                                double* alpha, double* beta, double* u,
+                                lapack_int ldu, double* v, lapack_int ldv,
+                                double* q, lapack_int ldq, double* work,
+                                lapack_int* ncycle );
+lapack_int LAPACKE_ctgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                float tola, float tolb, float* alpha,
+                                float* beta, lapack_complex_float* u,
+                                lapack_int ldu, lapack_complex_float* v,
+                                lapack_int ldv, lapack_complex_float* q,
+                                lapack_int ldq, lapack_complex_float* work,
+                                lapack_int* ncycle );
+lapack_int LAPACKE_ztgsja_work( int matrix_order, char jobu, char jobv,
+                                char jobq, lapack_int m, lapack_int p,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                double tola, double tolb, double* alpha,
+                                double* beta, lapack_complex_double* u,
+                                lapack_int ldu, lapack_complex_double* v,
+                                lapack_int ldv, lapack_complex_double* q,
+                                lapack_int ldq, lapack_complex_double* work,
+                                lapack_int* ncycle );
+
+lapack_int LAPACKE_stgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const float* a, lapack_int lda, const float* b,
+                                lapack_int ldb, const float* vl,
+                                lapack_int ldvl, const float* vr,
+                                lapack_int ldvr, float* s, float* dif,
+                                lapack_int mm, lapack_int* m, float* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_dtgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const double* b, lapack_int ldb,
+                                const double* vl, lapack_int ldvl,
+                                const double* vr, lapack_int ldvr, double* s,
+                                double* dif, lapack_int mm, lapack_int* m,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                const lapack_complex_float* vl, lapack_int ldvl,
+                                const lapack_complex_float* vr, lapack_int ldvr,
+                                float* s, float* dif, lapack_int mm,
+                                lapack_int* m, lapack_complex_float* work,
+                                lapack_int lwork, lapack_int* iwork );
+lapack_int LAPACKE_ztgsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                const lapack_complex_double* vl,
+                                lapack_int ldvl,
+                                const lapack_complex_double* vr,
+                                lapack_int ldvr, double* s, double* dif,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, lapack_int lwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_stgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n, const float* a,
+                                lapack_int lda, const float* b, lapack_int ldb,
+                                float* c, lapack_int ldc, const float* d,
+                                lapack_int ldd, const float* e, lapack_int lde,
+                                float* f, lapack_int ldf, float* scale,
+                                float* dif, float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dtgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n, const double* a,
+                                lapack_int lda, const double* b, lapack_int ldb,
+                                double* c, lapack_int ldc, const double* d,
+                                lapack_int ldd, const double* e, lapack_int lde,
+                                double* f, lapack_int ldf, double* scale,
+                                double* dif, double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* c, lapack_int ldc,
+                                const lapack_complex_float* d, lapack_int ldd,
+                                const lapack_complex_float* e, lapack_int lde,
+                                lapack_complex_float* f, lapack_int ldf,
+                                float* scale, float* dif,
+                                lapack_complex_float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ztgsyl_work( int matrix_order, char trans, lapack_int ijob,
+                                lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* c, lapack_int ldc,
+                                const lapack_complex_double* d, lapack_int ldd,
+                                const lapack_complex_double* e, lapack_int lde,
+                                lapack_complex_double* f, lapack_int ldf,
+                                double* scale, double* dif,
+                                lapack_complex_double* work, lapack_int lwork,
+                                lapack_int* iwork );
+
+lapack_int LAPACKE_stpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const float* ap,
+                                float* rcond, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const double* ap,
+                                double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_float* ap, float* rcond,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztpcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_double* ap, double* rcond,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* ap, const float* b, lapack_int ldb,
+                                const float* x, lapack_int ldx, float* ferr,
+                                float* berr, float* work, lapack_int* iwork );
+lapack_int LAPACKE_dtprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* ap, const double* b,
+                                lapack_int ldb, const double* x, lapack_int ldx,
+                                double* ferr, double* berr, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                const lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztprfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                const lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_stptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, float* ap );
+lapack_int LAPACKE_dtptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, double* ap );
+lapack_int LAPACKE_ctptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_float* ap );
+lapack_int LAPACKE_ztptri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_double* ap );
+
+lapack_int LAPACKE_stptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* ap, float* b, lapack_int ldb );
+lapack_int LAPACKE_dtptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* ap, double* b, lapack_int ldb );
+lapack_int LAPACKE_ctptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztptrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_stpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* ap, float* arf );
+lapack_int LAPACKE_dtpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* ap, double* arf );
+lapack_int LAPACKE_ctpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* ap,
+                                lapack_complex_float* arf );
+lapack_int LAPACKE_ztpttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* ap,
+                                lapack_complex_double* arf );
+
+lapack_int LAPACKE_stpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const float* ap, float* a, lapack_int lda );
+lapack_int LAPACKE_dtpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const double* ap, double* a, lapack_int lda );
+lapack_int LAPACKE_ctpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_ztpttr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_strcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const float* a,
+                                lapack_int lda, float* rcond, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dtrcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n, const double* a,
+                                lapack_int lda, double* rcond, double* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctrcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                float* rcond, lapack_complex_float* work,
+                                float* rwork );
+lapack_int LAPACKE_ztrcon_work( int matrix_order, char norm, char uplo,
+                                char diag, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                double* rcond, lapack_complex_double* work,
+                                double* rwork );
+
+lapack_int LAPACKE_strevc_work( int matrix_order, char side, char howmny,
+                                lapack_logical* select, lapack_int n,
+                                const float* t, lapack_int ldt, float* vl,
+                                lapack_int ldvl, float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, float* work );
+lapack_int LAPACKE_dtrevc_work( int matrix_order, char side, char howmny,
+                                lapack_logical* select, lapack_int n,
+                                const double* t, lapack_int ldt, double* vl,
+                                lapack_int ldvl, double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m, double* work );
+lapack_int LAPACKE_ctrevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* vl, lapack_int ldvl,
+                                lapack_complex_float* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztrevc_work( int matrix_order, char side, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* vl, lapack_int ldvl,
+                                lapack_complex_double* vr, lapack_int ldvr,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_strexc_work( int matrix_order, char compq, lapack_int n,
+                                float* t, lapack_int ldt, float* q,
+                                lapack_int ldq, lapack_int* ifst,
+                                lapack_int* ilst, float* work );
+lapack_int LAPACKE_dtrexc_work( int matrix_order, char compq, lapack_int n,
+                                double* t, lapack_int ldt, double* q,
+                                lapack_int ldq, lapack_int* ifst,
+                                lapack_int* ilst, double* work );
+lapack_int LAPACKE_ctrexc_work( int matrix_order, char compq, lapack_int n,
+                                lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_int ifst, lapack_int ilst );
+lapack_int LAPACKE_ztrexc_work( int matrix_order, char compq, lapack_int n,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_int ifst, lapack_int ilst );
+
+lapack_int LAPACKE_strrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* a, lapack_int lda, const float* b,
+                                lapack_int ldb, const float* x, lapack_int ldx,
+                                float* ferr, float* berr, float* work,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dtrrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* a, lapack_int lda,
+                                const double* b, lapack_int ldb,
+                                const double* x, lapack_int ldx, double* ferr,
+                                double* berr, double* work, lapack_int* iwork );
+lapack_int LAPACKE_ctrrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                const lapack_complex_float* x, lapack_int ldx,
+                                float* ferr, float* berr,
+                                lapack_complex_float* work, float* rwork );
+lapack_int LAPACKE_ztrrfs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                const lapack_complex_double* x, lapack_int ldx,
+                                double* ferr, double* berr,
+                                lapack_complex_double* work, double* rwork );
+
+lapack_int LAPACKE_strsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                float* t, lapack_int ldt, float* q,
+                                lapack_int ldq, float* wr, float* wi,
+                                lapack_int* m, float* s, float* sep,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_dtrsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                double* t, lapack_int ldt, double* q,
+                                lapack_int ldq, double* wr, double* wi,
+                                lapack_int* m, double* s, double* sep,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork, lapack_int liwork );
+lapack_int LAPACKE_ctrsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* w, lapack_int* m,
+                                float* s, float* sep,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_ztrsen_work( int matrix_order, char job, char compq,
+                                const lapack_logical* select, lapack_int n,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* w, lapack_int* m,
+                                double* s, double* sep,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_strsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const float* t, lapack_int ldt, const float* vl,
+                                lapack_int ldvl, const float* vr,
+                                lapack_int ldvr, float* s, float* sep,
+                                lapack_int mm, lapack_int* m, float* work,
+                                lapack_int ldwork, lapack_int* iwork );
+lapack_int LAPACKE_dtrsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const double* t, lapack_int ldt,
+                                const double* vl, lapack_int ldvl,
+                                const double* vr, lapack_int ldvr, double* s,
+                                double* sep, lapack_int mm, lapack_int* m,
+                                double* work, lapack_int ldwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ctrsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_float* t, lapack_int ldt,
+                                const lapack_complex_float* vl, lapack_int ldvl,
+                                const lapack_complex_float* vr, lapack_int ldvr,
+                                float* s, float* sep, lapack_int mm,
+                                lapack_int* m, lapack_complex_float* work,
+                                lapack_int ldwork, float* rwork );
+lapack_int LAPACKE_ztrsna_work( int matrix_order, char job, char howmny,
+                                const lapack_logical* select, lapack_int n,
+                                const lapack_complex_double* t, lapack_int ldt,
+                                const lapack_complex_double* vl,
+                                lapack_int ldvl,
+                                const lapack_complex_double* vr,
+                                lapack_int ldvr, double* s, double* sep,
+                                lapack_int mm, lapack_int* m,
+                                lapack_complex_double* work, lapack_int ldwork,
+                                double* rwork );
+
+lapack_int LAPACKE_strsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const float* a, lapack_int lda, const float* b,
+                                lapack_int ldb, float* c, lapack_int ldc,
+                                float* scale );
+lapack_int LAPACKE_dtrsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const double* a, lapack_int lda,
+                                const double* b, lapack_int ldb, double* c,
+                                lapack_int ldc, double* scale );
+lapack_int LAPACKE_ctrsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* b, lapack_int ldb,
+                                lapack_complex_float* c, lapack_int ldc,
+                                float* scale );
+lapack_int LAPACKE_ztrsyl_work( int matrix_order, char trana, char tranb,
+                                lapack_int isgn, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* c, lapack_int ldc,
+                                double* scale );
+
+lapack_int LAPACKE_strtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, float* a, lapack_int lda );
+lapack_int LAPACKE_dtrtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, double* a, lapack_int lda );
+lapack_int LAPACKE_ctrtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_float* a,
+                                lapack_int lda );
+lapack_int LAPACKE_ztrtri_work( int matrix_order, char uplo, char diag,
+                                lapack_int n, lapack_complex_double* a,
+                                lapack_int lda );
+
+lapack_int LAPACKE_strtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const float* a, lapack_int lda, float* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_dtrtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const double* a, lapack_int lda, double* b,
+                                lapack_int ldb );
+lapack_int LAPACKE_ctrtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztrtrs_work( int matrix_order, char uplo, char trans,
+                                char diag, lapack_int n, lapack_int nrhs,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_strttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const float* a, lapack_int lda,
+                                float* arf );
+lapack_int LAPACKE_dtrttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const double* a, lapack_int lda,
+                                double* arf );
+lapack_int LAPACKE_ctrttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* arf );
+lapack_int LAPACKE_ztrttf_work( int matrix_order, char transr, char uplo,
+                                lapack_int n, const lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* arf );
+
+lapack_int LAPACKE_strttp_work( int matrix_order, char uplo, lapack_int n,
+                                const float* a, lapack_int lda, float* ap );
+lapack_int LAPACKE_dtrttp_work( int matrix_order, char uplo, lapack_int n,
+                                const double* a, lapack_int lda, double* ap );
+lapack_int LAPACKE_ctrttp_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* ap );
+lapack_int LAPACKE_ztrttp_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* ap );
+
+lapack_int LAPACKE_stzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                float* a, lapack_int lda, float* tau,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_dtzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                double* a, lapack_int lda, double* tau,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_ctzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_ztzrzf_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungbr_work( int matrix_order, char vect, lapack_int m,
+                                lapack_int n, lapack_int k,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunghr_work( int matrix_order, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunglq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungql_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungqr_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungrq_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int k, lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cungtr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zungtr_work( int matrix_order, char uplo, lapack_int n,
+                                lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmbr_work( int matrix_order, char vect, char side,
+                                char trans, lapack_int m, lapack_int n,
+                                lapack_int k, const lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmhr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int ilo,
+                                lapack_int ihi, const lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmlq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmql_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmqr_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmrq_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const lapack_complex_float* a,
+                                lapack_int lda, const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmrz_work( int matrix_order, char side, char trans,
+                                lapack_int m, lapack_int n, lapack_int k,
+                                lapack_int l, const lapack_complex_double* a,
+                                lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cunmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_float* a, lapack_int lda,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_zunmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_double* a, lapack_int lda,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work, lapack_int lwork );
+
+lapack_int LAPACKE_cupgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* q, lapack_int ldq,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zupgtr_work( int matrix_order, char uplo, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* q, lapack_int ldq,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_cupmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_float* ap,
+                                const lapack_complex_float* tau,
+                                lapack_complex_float* c, lapack_int ldc,
+                                lapack_complex_float* work );
+lapack_int LAPACKE_zupmtr_work( int matrix_order, char side, char uplo,
+                                char trans, lapack_int m, lapack_int n,
+                                const lapack_complex_double* ap,
+                                const lapack_complex_double* tau,
+                                lapack_complex_double* c, lapack_int ldc,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_claghe( int matrix_order, lapack_int n, lapack_int k,
+                           const float* d, lapack_complex_float* a,
+                           lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_zlaghe( int matrix_order, lapack_int n, lapack_int k,
+                           const double* d, lapack_complex_double* a,
+                           lapack_int lda, lapack_int* iseed );
+
+lapack_int LAPACKE_slagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const float* d, float* a, lapack_int lda,
+                           lapack_int* iseed );
+lapack_int LAPACKE_dlagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const double* d, double* a, lapack_int lda,
+                           lapack_int* iseed );
+lapack_int LAPACKE_clagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const float* d, lapack_complex_float* a,
+                           lapack_int lda, lapack_int* iseed );
+lapack_int LAPACKE_zlagsy( int matrix_order, lapack_int n, lapack_int k,
+                           const double* d, lapack_complex_double* a,
+                           lapack_int lda, lapack_int* iseed );
+
+lapack_int LAPACKE_slapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, float* x, lapack_int ldx,
+                           lapack_int* k );
+lapack_int LAPACKE_dlapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, double* x,
+                           lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_clapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, lapack_complex_float* x,
+                           lapack_int ldx, lapack_int* k );
+lapack_int LAPACKE_zlapmr( int matrix_order, lapack_logical forwrd,
+                           lapack_int m, lapack_int n, lapack_complex_double* x,
+                           lapack_int ldx, lapack_int* k );
+
+
+float LAPACKE_slapy2( float x, float y );
+double LAPACKE_dlapy2( double x, double y );
+
+float LAPACKE_slapy3( float x, float y, float z );
+double LAPACKE_dlapy3( double x, double y, double z );
+
+lapack_int LAPACKE_slartgp( float f, float g, float* cs, float* sn, float* r );
+lapack_int LAPACKE_dlartgp( double f, double g, double* cs, double* sn,
+                            double* r );
+
+lapack_int LAPACKE_slartgs( float x, float y, float sigma, float* cs,
+                            float* sn );
+lapack_int LAPACKE_dlartgs( double x, double y, double sigma, double* cs,
+                            double* sn );
+
+
+//LAPACK 3.3.0
+lapack_int LAPACKE_cbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, float* theta, float* phi,
+                           lapack_complex_float* u1, lapack_int ldu1,
+                           lapack_complex_float* u2, lapack_int ldu2,
+                           lapack_complex_float* v1t, lapack_int ldv1t,
+                           lapack_complex_float* v2t, lapack_int ldv2t,
+                           float* b11d, float* b11e, float* b12d, float* b12e,
+                           float* b21d, float* b21e, float* b22d, float* b22e );
+lapack_int LAPACKE_cbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                float* theta, float* phi,
+                                lapack_complex_float* u1, lapack_int ldu1,
+                                lapack_complex_float* u2, lapack_int ldu2,
+                                lapack_complex_float* v1t, lapack_int ldv1t,
+                                lapack_complex_float* v2t, lapack_int ldv2t,
+                                float* b11d, float* b11e, float* b12d,
+                                float* b12e, float* b21d, float* b21e,
+                                float* b22d, float* b22e, float* rwork,
+                                lapack_int lrwork );
+lapack_int LAPACKE_cheswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_cheswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_chetri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_chetri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_chetri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_chetri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_float* work, lapack_int nb );
+lapack_int LAPACKE_chetrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_float* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_chetrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_csyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_csyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, lapack_complex_float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_csyswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_csyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_csytri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_csytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_csytri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_csytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_float* work, lapack_int nb );
+lapack_int LAPACKE_csytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_float* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_csytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_cunbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_float* x11, lapack_int ldx11,
+                           lapack_complex_float* x12, lapack_int ldx12,
+                           lapack_complex_float* x21, lapack_int ldx21,
+                           lapack_complex_float* x22, lapack_int ldx22,
+                           float* theta, float* phi,
+                           lapack_complex_float* taup1,
+                           lapack_complex_float* taup2,
+                           lapack_complex_float* tauq1,
+                           lapack_complex_float* tauq2 );
+lapack_int LAPACKE_cunbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                lapack_complex_float* x11, lapack_int ldx11,
+                                lapack_complex_float* x12, lapack_int ldx12,
+                                lapack_complex_float* x21, lapack_int ldx21,
+                                lapack_complex_float* x22, lapack_int ldx22,
+                                float* theta, float* phi,
+                                lapack_complex_float* taup1,
+                                lapack_complex_float* taup2,
+                                lapack_complex_float* tauq1,
+                                lapack_complex_float* tauq2,
+                                lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_cuncsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_float* x11, lapack_int ldx11,
+                           lapack_complex_float* x12, lapack_int ldx12,
+                           lapack_complex_float* x21, lapack_int ldx21,
+                           lapack_complex_float* x22, lapack_int ldx22,
+                           float* theta, lapack_complex_float* u1,
+                           lapack_int ldu1, lapack_complex_float* u2,
+                           lapack_int ldu2, lapack_complex_float* v1t,
+                           lapack_int ldv1t, lapack_complex_float* v2t,
+                           lapack_int ldv2t );
+lapack_int LAPACKE_cuncsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, lapack_complex_float* x11,
+                                lapack_int ldx11, lapack_complex_float* x12,
+                                lapack_int ldx12, lapack_complex_float* x21,
+                                lapack_int ldx21, lapack_complex_float* x22,
+                                lapack_int ldx22, float* theta,
+                                lapack_complex_float* u1, lapack_int ldu1,
+                                lapack_complex_float* u2, lapack_int ldu2,
+                                lapack_complex_float* v1t, lapack_int ldv1t,
+                                lapack_complex_float* v2t, lapack_int ldv2t,
+                                lapack_complex_float* work, lapack_int lwork,
+                                float* rwork, lapack_int lrwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, double* theta,
+                           double* phi, double* u1, lapack_int ldu1, double* u2,
+                           lapack_int ldu2, double* v1t, lapack_int ldv1t,
+                           double* v2t, lapack_int ldv2t, double* b11d,
+                           double* b11e, double* b12d, double* b12e,
+                           double* b21d, double* b21e, double* b22d,
+                           double* b22e );
+lapack_int LAPACKE_dbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                double* theta, double* phi, double* u1,
+                                lapack_int ldu1, double* u2, lapack_int ldu2,
+                                double* v1t, lapack_int ldv1t, double* v2t,
+                                lapack_int ldv2t, double* b11d, double* b11e,
+                                double* b12d, double* b12e, double* b21d,
+                                double* b21e, double* b22d, double* b22e,
+                                double* work, lapack_int lwork );
+lapack_int LAPACKE_dorbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           double* x11, lapack_int ldx11, double* x12,
+                           lapack_int ldx12, double* x21, lapack_int ldx21,
+                           double* x22, lapack_int ldx22, double* theta,
+                           double* phi, double* taup1, double* taup2,
+                           double* tauq1, double* tauq2 );
+lapack_int LAPACKE_dorbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                double* x11, lapack_int ldx11, double* x12,
+                                lapack_int ldx12, double* x21, lapack_int ldx21,
+                                double* x22, lapack_int ldx22, double* theta,
+                                double* phi, double* taup1, double* taup2,
+                                double* tauq1, double* tauq2, double* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_dorcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           double* x11, lapack_int ldx11, double* x12,
+                           lapack_int ldx12, double* x21, lapack_int ldx21,
+                           double* x22, lapack_int ldx22, double* theta,
+                           double* u1, lapack_int ldu1, double* u2,
+                           lapack_int ldu2, double* v1t, lapack_int ldv1t,
+                           double* v2t, lapack_int ldv2t );
+lapack_int LAPACKE_dorcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, double* x11, lapack_int ldx11,
+                                double* x12, lapack_int ldx12, double* x21,
+                                lapack_int ldx21, double* x22, lapack_int ldx22,
+                                double* theta, double* u1, lapack_int ldu1,
+                                double* u2, lapack_int ldu2, double* v1t,
+                                lapack_int ldv1t, double* v2t, lapack_int ldv2t,
+                                double* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_dsyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            double* a, lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dsyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, double* a, lapack_int lda,
+                                 const lapack_int* ipiv, double* work );
+lapack_int LAPACKE_dsyswapr( int matrix_order, char uplo, lapack_int n,
+                             double* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_dsyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  double* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_dsytri2( int matrix_order, char uplo, lapack_int n,
+                            double* a, lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_dsytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 double* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_dsytri2x( int matrix_order, char uplo, lapack_int n,
+                             double* a, lapack_int lda, const lapack_int* ipiv,
+                             lapack_int nb );
+lapack_int LAPACKE_dsytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  double* a, lapack_int lda,
+                                  const lapack_int* ipiv, double* work,
+                                  lapack_int nb );
+lapack_int LAPACKE_dsytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const double* a, lapack_int lda,
+                            const lapack_int* ipiv, double* b, lapack_int ldb );
+lapack_int LAPACKE_dsytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 double* b, lapack_int ldb, double* work );
+lapack_int LAPACKE_sbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, float* theta, float* phi,
+                           float* u1, lapack_int ldu1, float* u2,
+                           lapack_int ldu2, float* v1t, lapack_int ldv1t,
+                           float* v2t, lapack_int ldv2t, float* b11d,
+                           float* b11e, float* b12d, float* b12e, float* b21d,
+                           float* b21e, float* b22d, float* b22e );
+lapack_int LAPACKE_sbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                float* theta, float* phi, float* u1,
+                                lapack_int ldu1, float* u2, lapack_int ldu2,
+                                float* v1t, lapack_int ldv1t, float* v2t,
+                                lapack_int ldv2t, float* b11d, float* b11e,
+                                float* b12d, float* b12e, float* b21d,
+                                float* b21e, float* b22d, float* b22e,
+                                float* work, lapack_int lwork );
+lapack_int LAPACKE_sorbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q, float* x11,
+                           lapack_int ldx11, float* x12, lapack_int ldx12,
+                           float* x21, lapack_int ldx21, float* x22,
+                           lapack_int ldx22, float* theta, float* phi,
+                           float* taup1, float* taup2, float* tauq1,
+                           float* tauq2 );
+lapack_int LAPACKE_sorbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                float* x11, lapack_int ldx11, float* x12,
+                                lapack_int ldx12, float* x21, lapack_int ldx21,
+                                float* x22, lapack_int ldx22, float* theta,
+                                float* phi, float* taup1, float* taup2,
+                                float* tauq1, float* tauq2, float* work,
+                                lapack_int lwork );
+lapack_int LAPACKE_sorcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q, float* x11,
+                           lapack_int ldx11, float* x12, lapack_int ldx12,
+                           float* x21, lapack_int ldx21, float* x22,
+                           lapack_int ldx22, float* theta, float* u1,
+                           lapack_int ldu1, float* u2, lapack_int ldu2,
+                           float* v1t, lapack_int ldv1t, float* v2t,
+                           lapack_int ldv2t );
+lapack_int LAPACKE_sorcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, float* x11, lapack_int ldx11,
+                                float* x12, lapack_int ldx12, float* x21,
+                                lapack_int ldx21, float* x22, lapack_int ldx22,
+                                float* theta, float* u1, lapack_int ldu1,
+                                float* u2, lapack_int ldu2, float* v1t,
+                                lapack_int ldv1t, float* v2t, lapack_int ldv2t,
+                                float* work, lapack_int lwork,
+                                lapack_int* iwork );
+lapack_int LAPACKE_ssyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            float* a, lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_ssyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, float* a, lapack_int lda,
+                                 const lapack_int* ipiv, float* work );
+lapack_int LAPACKE_ssyswapr( int matrix_order, char uplo, lapack_int n,
+                             float* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_ssyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  float* a, lapack_int i1, lapack_int i2 );
+lapack_int LAPACKE_ssytri2( int matrix_order, char uplo, lapack_int n, float* a,
+                            lapack_int lda, const lapack_int* ipiv );
+lapack_int LAPACKE_ssytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 float* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_float* work, lapack_int lwork );
+lapack_int LAPACKE_ssytri2x( int matrix_order, char uplo, lapack_int n,
+                             float* a, lapack_int lda, const lapack_int* ipiv,
+                             lapack_int nb );
+lapack_int LAPACKE_ssytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  float* a, lapack_int lda,
+                                  const lapack_int* ipiv, float* work,
+                                  lapack_int nb );
+lapack_int LAPACKE_ssytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const float* a, lapack_int lda,
+                            const lapack_int* ipiv, float* b, lapack_int ldb );
+lapack_int LAPACKE_ssytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const float* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 float* b, lapack_int ldb, float* work );
+lapack_int LAPACKE_zbbcsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, lapack_int m,
+                           lapack_int p, lapack_int q, double* theta,
+                           double* phi, lapack_complex_double* u1,
+                           lapack_int ldu1, lapack_complex_double* u2,
+                           lapack_int ldu2, lapack_complex_double* v1t,
+                           lapack_int ldv1t, lapack_complex_double* v2t,
+                           lapack_int ldv2t, double* b11d, double* b11e,
+                           double* b12d, double* b12e, double* b21d,
+                           double* b21e, double* b22d, double* b22e );
+lapack_int LAPACKE_zbbcsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                double* theta, double* phi,
+                                lapack_complex_double* u1, lapack_int ldu1,
+                                lapack_complex_double* u2, lapack_int ldu2,
+                                lapack_complex_double* v1t, lapack_int ldv1t,
+                                lapack_complex_double* v2t, lapack_int ldv2t,
+                                double* b11d, double* b11e, double* b12d,
+                                double* b12e, double* b21d, double* b21e,
+                                double* b22d, double* b22e, double* rwork,
+                                lapack_int lrwork );
+lapack_int LAPACKE_zheswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_zheswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_zhetri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_zhetri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_zhetri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_zhetri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_double* work, lapack_int nb );
+lapack_int LAPACKE_zhetrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_double* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_double* b, lapack_int ldb );
+lapack_int LAPACKE_zhetrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* work );
+lapack_int LAPACKE_zsyconv( int matrix_order, char uplo, char way, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_zsyconv_work( int matrix_order, char uplo, char way,
+                                 lapack_int n, lapack_complex_double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_double* work );
+lapack_int LAPACKE_zsyswapr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int i1,
+                             lapack_int i2 );
+lapack_int LAPACKE_zsyswapr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int i1,
+                                  lapack_int i2 );
+lapack_int LAPACKE_zsytri2( int matrix_order, char uplo, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            const lapack_int* ipiv );
+lapack_int LAPACKE_zsytri2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 const lapack_int* ipiv,
+                                 lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_zsytri2x( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double* a, lapack_int lda,
+                             const lapack_int* ipiv, lapack_int nb );
+lapack_int LAPACKE_zsytri2x_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double* a, lapack_int lda,
+                                  const lapack_int* ipiv,
+                                  lapack_complex_double* work, lapack_int nb );
+lapack_int LAPACKE_zsytrs2( int matrix_order, char uplo, lapack_int n,
+                            lapack_int nrhs, const lapack_complex_double* a,
+                            lapack_int lda, const lapack_int* ipiv,
+                            lapack_complex_double* b, lapack_int ldb );
+lapack_int LAPACKE_zsytrs2_work( int matrix_order, char uplo, lapack_int n,
+                                 lapack_int nrhs, const lapack_complex_double* a,
+                                 lapack_int lda, const lapack_int* ipiv,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* work );
+lapack_int LAPACKE_zunbdb( int matrix_order, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_double* x11, lapack_int ldx11,
+                           lapack_complex_double* x12, lapack_int ldx12,
+                           lapack_complex_double* x21, lapack_int ldx21,
+                           lapack_complex_double* x22, lapack_int ldx22,
+                           double* theta, double* phi,
+                           lapack_complex_double* taup1,
+                           lapack_complex_double* taup2,
+                           lapack_complex_double* tauq1,
+                           lapack_complex_double* tauq2 );
+lapack_int LAPACKE_zunbdb_work( int matrix_order, char trans, char signs,
+                                lapack_int m, lapack_int p, lapack_int q,
+                                lapack_complex_double* x11, lapack_int ldx11,
+                                lapack_complex_double* x12, lapack_int ldx12,
+                                lapack_complex_double* x21, lapack_int ldx21,
+                                lapack_complex_double* x22, lapack_int ldx22,
+                                double* theta, double* phi,
+                                lapack_complex_double* taup1,
+                                lapack_complex_double* taup2,
+                                lapack_complex_double* tauq1,
+                                lapack_complex_double* tauq2,
+                                lapack_complex_double* work, lapack_int lwork );
+lapack_int LAPACKE_zuncsd( int matrix_order, char jobu1, char jobu2,
+                           char jobv1t, char jobv2t, char trans, char signs,
+                           lapack_int m, lapack_int p, lapack_int q,
+                           lapack_complex_double* x11, lapack_int ldx11,
+                           lapack_complex_double* x12, lapack_int ldx12,
+                           lapack_complex_double* x21, lapack_int ldx21,
+                           lapack_complex_double* x22, lapack_int ldx22,
+                           double* theta, lapack_complex_double* u1,
+                           lapack_int ldu1, lapack_complex_double* u2,
+                           lapack_int ldu2, lapack_complex_double* v1t,
+                           lapack_int ldv1t, lapack_complex_double* v2t,
+                           lapack_int ldv2t );
+lapack_int LAPACKE_zuncsd_work( int matrix_order, char jobu1, char jobu2,
+                                char jobv1t, char jobv2t, char trans,
+                                char signs, lapack_int m, lapack_int p,
+                                lapack_int q, lapack_complex_double* x11,
+                                lapack_int ldx11, lapack_complex_double* x12,
+                                lapack_int ldx12, lapack_complex_double* x21,
+                                lapack_int ldx21, lapack_complex_double* x22,
+                                lapack_int ldx22, double* theta,
+                                lapack_complex_double* u1, lapack_int ldu1,
+                                lapack_complex_double* u2, lapack_int ldu2,
+                                lapack_complex_double* v1t, lapack_int ldv1t,
+                                lapack_complex_double* v2t, lapack_int ldv2t,
+                                lapack_complex_double* work, lapack_int lwork,
+                                double* rwork, lapack_int lrwork,
+                                lapack_int* iwork );
+//LAPACK 3.4.0
+lapack_int LAPACKE_sgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const float* v, lapack_int ldv,
+                            const float* t, lapack_int ldt, float* c,
+                            lapack_int ldc );
+lapack_int LAPACKE_dgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const double* v, lapack_int ldv,
+                            const double* t, lapack_int ldt, double* c,
+                            lapack_int ldc );
+lapack_int LAPACKE_cgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const lapack_complex_float* v,
+                            lapack_int ldv, const lapack_complex_float* t,
+                            lapack_int ldt, lapack_complex_float* c,
+                            lapack_int ldc );
+lapack_int LAPACKE_zgemqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int nb, const lapack_complex_double* v,
+                            lapack_int ldv, const lapack_complex_double* t,
+                            lapack_int ldt, lapack_complex_double* c,
+                            lapack_int ldc );
+
+lapack_int LAPACKE_sgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, float* a, lapack_int lda, float* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_dgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, double* a, lapack_int lda, double* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_cgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_zgeqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int nb, lapack_complex_double* a,
+                           lapack_int lda, lapack_complex_double* t,
+                           lapack_int ldt );
+
+lapack_int LAPACKE_sgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_dgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_cgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_sgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_dgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* t,
+                            lapack_int ldt );
+lapack_int LAPACKE_cgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt3( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb, const float* v,
+                            lapack_int ldv, const float* t, lapack_int ldt,
+                            float* a, lapack_int lda, float* b,
+                            lapack_int ldb );
+lapack_int LAPACKE_dtpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb, const double* v,
+                            lapack_int ldv, const double* t, lapack_int ldt,
+                            double* a, lapack_int lda, double* b,
+                            lapack_int ldb );
+lapack_int LAPACKE_ctpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb,
+                            const lapack_complex_float* v, lapack_int ldv,
+                            const lapack_complex_float* t, lapack_int ldt,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* b, lapack_int ldb );
+lapack_int LAPACKE_ztpmqrt( int matrix_order, char side, char trans,
+                            lapack_int m, lapack_int n, lapack_int k,
+                            lapack_int l, lapack_int nb,
+                            const lapack_complex_double* v, lapack_int ldv,
+                            const lapack_complex_double* t, lapack_int ldt,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* b, lapack_int ldb );
+
+lapack_int LAPACKE_dtpqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int l, lapack_int nb, double* a,
+                           lapack_int lda, double* b, lapack_int ldb, double* t,
+                           lapack_int ldt );
+lapack_int LAPACKE_ctpqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int l, lapack_int nb, lapack_complex_float* a,
+                           lapack_int lda, lapack_complex_float* t,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_int ldt );
+lapack_int LAPACKE_ztpqrt( int matrix_order, lapack_int m, lapack_int n,
+                           lapack_int l, lapack_int nb,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            float* a, lapack_int lda, float* b, lapack_int ldb,
+                            float* t, lapack_int ldt );
+lapack_int LAPACKE_dtpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            double* a, lapack_int lda, double* b,
+                            lapack_int ldb, double* t, lapack_int ldt );
+lapack_int LAPACKE_ctpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_float* a, lapack_int lda,
+                            lapack_complex_float* b, lapack_int ldb,
+                            lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_ztpqrt2( int matrix_order, lapack_int m, lapack_int n,
+                            lapack_complex_double* a, lapack_int lda,
+                            lapack_complex_double* b, lapack_int ldb,
+                            lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l, const float* v,
+                           lapack_int ldv, const float* t, lapack_int ldt,
+                           float* a, lapack_int lda, float* b, lapack_int ldb,
+                           lapack_int myldwork );
+lapack_int LAPACKE_dtprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l, const double* v,
+                           lapack_int ldv, const double* t, lapack_int ldt,
+                           double* a, lapack_int lda, double* b, lapack_int ldb,
+                           lapack_int myldwork );
+lapack_int LAPACKE_ctprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l,
+                           const lapack_complex_float* v, lapack_int ldv,
+                           const lapack_complex_float* t, lapack_int ldt,
+                           lapack_complex_float* a, lapack_int lda,
+                           lapack_complex_float* b, lapack_int ldb,
+                           lapack_int myldwork );
+lapack_int LAPACKE_ztprfb( int matrix_order, char side, char trans, char direct,
+                           char storev, lapack_int m, lapack_int n,
+                           lapack_int k, lapack_int l,
+                           const lapack_complex_double* v, lapack_int ldv,
+                           const lapack_complex_double* t, lapack_int ldt,
+                           lapack_complex_double* a, lapack_int lda,
+                           lapack_complex_double* b, lapack_int ldb,
+                           lapack_int myldwork );
+
+lapack_int LAPACKE_sgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const float* v, lapack_int ldv,
+                                 const float* t, lapack_int ldt, float* c,
+                                 lapack_int ldc, float* work );
+lapack_int LAPACKE_dgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const double* v, lapack_int ldv,
+                                 const double* t, lapack_int ldt, double* c,
+                                 lapack_int ldc, double* work );
+lapack_int LAPACKE_cgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const lapack_complex_float* v,
+                                 lapack_int ldv, const lapack_complex_float* t,
+                                 lapack_int ldt, lapack_complex_float* c,
+                                 lapack_int ldc, lapack_complex_float* work );
+lapack_int LAPACKE_zgemqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int nb, const lapack_complex_double* v,
+                                 lapack_int ldv, const lapack_complex_double* t,
+                                 lapack_int ldt, lapack_complex_double* c,
+                                 lapack_int ldc, lapack_complex_double* work );
+
+lapack_int LAPACKE_sgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, float* a, lapack_int lda,
+                                float* t, lapack_int ldt, float* work );
+lapack_int LAPACKE_dgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, double* a, lapack_int lda,
+                                double* t, lapack_int ldt, double* work );
+lapack_int LAPACKE_cgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, lapack_complex_float* a,
+                                lapack_int lda, lapack_complex_float* t,
+                                lapack_int ldt, lapack_complex_float* work );
+lapack_int LAPACKE_zgeqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int nb, lapack_complex_double* a,
+                                lapack_int lda, lapack_complex_double* t,
+                                lapack_int ldt, lapack_complex_double* work );
+
+lapack_int LAPACKE_sgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_dgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_cgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_sgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_dgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* t,
+                                 lapack_int ldt );
+lapack_int LAPACKE_cgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_zgeqrt3_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb, const float* v,
+                                 lapack_int ldv, const float* t, lapack_int ldt,
+                                 float* a, lapack_int lda, float* b,
+                                 lapack_int ldb, float* work );
+lapack_int LAPACKE_dtpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb, const double* v,
+                                 lapack_int ldv, const double* t,
+                                 lapack_int ldt, double* a, lapack_int lda,
+                                 double* b, lapack_int ldb, double* work );
+lapack_int LAPACKE_ctpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb,
+                                 const lapack_complex_float* v, lapack_int ldv,
+                                 const lapack_complex_float* t, lapack_int ldt,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* work );
+lapack_int LAPACKE_ztpmqrt_work( int matrix_order, char side, char trans,
+                                 lapack_int m, lapack_int n, lapack_int k,
+                                 lapack_int l, lapack_int nb,
+                                 const lapack_complex_double* v, lapack_int ldv,
+                                 const lapack_complex_double* t, lapack_int ldt,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* work );
+
+lapack_int LAPACKE_dtpqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int l, lapack_int nb, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                double* t, lapack_int ldt, double* work );
+lapack_int LAPACKE_ctpqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int l, lapack_int nb,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* t,
+                                lapack_complex_float* b, lapack_int ldb,
+                                lapack_int ldt, lapack_complex_float* work );
+lapack_int LAPACKE_ztpqrt_work( int matrix_order, lapack_int m, lapack_int n,
+                                lapack_int l, lapack_int nb,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* work );
+
+lapack_int LAPACKE_stpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 float* a, lapack_int lda, float* b,
+                                 lapack_int ldb, float* t, lapack_int ldt );
+lapack_int LAPACKE_dtpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 double* a, lapack_int lda, double* b,
+                                 lapack_int ldb, double* t, lapack_int ldt );
+lapack_int LAPACKE_ctpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_float* a, lapack_int lda,
+                                 lapack_complex_float* b, lapack_int ldb,
+                                 lapack_complex_float* t, lapack_int ldt );
+lapack_int LAPACKE_ztpqrt2_work( int matrix_order, lapack_int m, lapack_int n,
+                                 lapack_complex_double* a, lapack_int lda,
+                                 lapack_complex_double* b, lapack_int ldb,
+                                 lapack_complex_double* t, lapack_int ldt );
+
+lapack_int LAPACKE_stprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const float* v, lapack_int ldv, const float* t,
+                                lapack_int ldt, float* a, lapack_int lda,
+                                float* b, lapack_int ldb, const float* mywork,
+                                lapack_int myldwork );
+lapack_int LAPACKE_dtprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const double* v, lapack_int ldv,
+                                const double* t, lapack_int ldt, double* a,
+                                lapack_int lda, double* b, lapack_int ldb,
+                                const double* mywork, lapack_int myldwork );
+lapack_int LAPACKE_ctprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const lapack_complex_float* v, lapack_int ldv,
+                                const lapack_complex_float* t, lapack_int ldt,
+                                lapack_complex_float* a, lapack_int lda,
+                                lapack_complex_float* b, lapack_int ldb,
+                                const float* mywork, lapack_int myldwork );
+lapack_int LAPACKE_ztprfb_work( int matrix_order, char side, char trans,
+                                char direct, char storev, lapack_int m,
+                                lapack_int n, lapack_int k, lapack_int l,
+                                const lapack_complex_double* v, lapack_int ldv,
+                                const lapack_complex_double* t, lapack_int ldt,
+                                lapack_complex_double* a, lapack_int lda,
+                                lapack_complex_double* b, lapack_int ldb,
+                                const double* mywork, lapack_int myldwork );
+//LAPACK 3.X.X
+lapack_int LAPACKE_csyr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_float alpha,
+                             const lapack_complex_float* x, lapack_int incx,
+                             lapack_complex_float* a, lapack_int lda );
+lapack_int LAPACKE_zsyr( int matrix_order, char uplo, lapack_int n,
+                             lapack_complex_double alpha,
+                             const lapack_complex_double* x, lapack_int incx,
+                             lapack_complex_double* a, lapack_int lda );
+
+lapack_int LAPACKE_csyr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_float alpha,
+                                  const lapack_complex_float* x,
+                                  lapack_int incx, lapack_complex_float* a,
+                                  lapack_int lda );
+lapack_int LAPACKE_zsyr_work( int matrix_order, char uplo, lapack_int n,
+                                  lapack_complex_double alpha,
+                                  const lapack_complex_double* x,
+                                  lapack_int incx, lapack_complex_double* a,
+                                  lapack_int lda );
+
+
+
+#define LAPACK_sgetrf LAPACK_GLOBAL(sgetrf,SGETRF)
+#define LAPACK_dgetrf LAPACK_GLOBAL(dgetrf,DGETRF)
+#define LAPACK_cgetrf LAPACK_GLOBAL(cgetrf,CGETRF)
+#define LAPACK_zgetrf LAPACK_GLOBAL(zgetrf,ZGETRF)
+#define LAPACK_sgbtrf LAPACK_GLOBAL(sgbtrf,SGBTRF)
+#define LAPACK_dgbtrf LAPACK_GLOBAL(dgbtrf,DGBTRF)
+#define LAPACK_cgbtrf LAPACK_GLOBAL(cgbtrf,CGBTRF)
+#define LAPACK_zgbtrf LAPACK_GLOBAL(zgbtrf,ZGBTRF)
+#define LAPACK_sgttrf LAPACK_GLOBAL(sgttrf,SGTTRF)
+#define LAPACK_dgttrf LAPACK_GLOBAL(dgttrf,DGTTRF)
+#define LAPACK_cgttrf LAPACK_GLOBAL(cgttrf,CGTTRF)
+#define LAPACK_zgttrf LAPACK_GLOBAL(zgttrf,ZGTTRF)
+#define LAPACK_spotrf LAPACK_GLOBAL(spotrf,SPOTRF)
+#define LAPACK_dpotrf LAPACK_GLOBAL(dpotrf,DPOTRF)
+#define LAPACK_cpotrf LAPACK_GLOBAL(cpotrf,CPOTRF)
+#define LAPACK_zpotrf LAPACK_GLOBAL(zpotrf,ZPOTRF)
+#define LAPACK_dpstrf LAPACK_GLOBAL(dpstrf,DPSTRF)
+#define LAPACK_spstrf LAPACK_GLOBAL(spstrf,SPSTRF)
+#define LAPACK_zpstrf LAPACK_GLOBAL(zpstrf,ZPSTRF)
+#define LAPACK_cpstrf LAPACK_GLOBAL(cpstrf,CPSTRF)
+#define LAPACK_dpftrf LAPACK_GLOBAL(dpftrf,DPFTRF)
+#define LAPACK_spftrf LAPACK_GLOBAL(spftrf,SPFTRF)
+#define LAPACK_zpftrf LAPACK_GLOBAL(zpftrf,ZPFTRF)
+#define LAPACK_cpftrf LAPACK_GLOBAL(cpftrf,CPFTRF)
+#define LAPACK_spptrf LAPACK_GLOBAL(spptrf,SPPTRF)
+#define LAPACK_dpptrf LAPACK_GLOBAL(dpptrf,DPPTRF)
+#define LAPACK_cpptrf LAPACK_GLOBAL(cpptrf,CPPTRF)
+#define LAPACK_zpptrf LAPACK_GLOBAL(zpptrf,ZPPTRF)
+#define LAPACK_spbtrf LAPACK_GLOBAL(spbtrf,SPBTRF)
+#define LAPACK_dpbtrf LAPACK_GLOBAL(dpbtrf,DPBTRF)
+#define LAPACK_cpbtrf LAPACK_GLOBAL(cpbtrf,CPBTRF)
+#define LAPACK_zpbtrf LAPACK_GLOBAL(zpbtrf,ZPBTRF)
+#define LAPACK_spttrf LAPACK_GLOBAL(spttrf,SPTTRF)
+#define LAPACK_dpttrf LAPACK_GLOBAL(dpttrf,DPTTRF)
+#define LAPACK_cpttrf LAPACK_GLOBAL(cpttrf,CPTTRF)
+#define LAPACK_zpttrf LAPACK_GLOBAL(zpttrf,ZPTTRF)
+#define LAPACK_ssytrf LAPACK_GLOBAL(ssytrf,SSYTRF)
+#define LAPACK_dsytrf LAPACK_GLOBAL(dsytrf,DSYTRF)
+#define LAPACK_csytrf LAPACK_GLOBAL(csytrf,CSYTRF)
+#define LAPACK_zsytrf LAPACK_GLOBAL(zsytrf,ZSYTRF)
+#define LAPACK_chetrf LAPACK_GLOBAL(chetrf,CHETRF)
+#define LAPACK_zhetrf LAPACK_GLOBAL(zhetrf,ZHETRF)
+#define LAPACK_ssptrf LAPACK_GLOBAL(ssptrf,SSPTRF)
+#define LAPACK_dsptrf LAPACK_GLOBAL(dsptrf,DSPTRF)
+#define LAPACK_csptrf LAPACK_GLOBAL(csptrf,CSPTRF)
+#define LAPACK_zsptrf LAPACK_GLOBAL(zsptrf,ZSPTRF)
+#define LAPACK_chptrf LAPACK_GLOBAL(chptrf,CHPTRF)
+#define LAPACK_zhptrf LAPACK_GLOBAL(zhptrf,ZHPTRF)
+#define LAPACK_sgetrs LAPACK_GLOBAL(sgetrs,SGETRS)
+#define LAPACK_dgetrs LAPACK_GLOBAL(dgetrs,DGETRS)
+#define LAPACK_cgetrs LAPACK_GLOBAL(cgetrs,CGETRS)
+#define LAPACK_zgetrs LAPACK_GLOBAL(zgetrs,ZGETRS)
+#define LAPACK_sgbtrs LAPACK_GLOBAL(sgbtrs,SGBTRS)
+#define LAPACK_dgbtrs LAPACK_GLOBAL(dgbtrs,DGBTRS)
+#define LAPACK_cgbtrs LAPACK_GLOBAL(cgbtrs,CGBTRS)
+#define LAPACK_zgbtrs LAPACK_GLOBAL(zgbtrs,ZGBTRS)
+#define LAPACK_sgttrs LAPACK_GLOBAL(sgttrs,SGTTRS)
+#define LAPACK_dgttrs LAPACK_GLOBAL(dgttrs,DGTTRS)
+#define LAPACK_cgttrs LAPACK_GLOBAL(cgttrs,CGTTRS)
+#define LAPACK_zgttrs LAPACK_GLOBAL(zgttrs,ZGTTRS)
+#define LAPACK_spotrs LAPACK_GLOBAL(spotrs,SPOTRS)
+#define LAPACK_dpotrs LAPACK_GLOBAL(dpotrs,DPOTRS)
+#define LAPACK_cpotrs LAPACK_GLOBAL(cpotrs,CPOTRS)
+#define LAPACK_zpotrs LAPACK_GLOBAL(zpotrs,ZPOTRS)
+#define LAPACK_dpftrs LAPACK_GLOBAL(dpftrs,DPFTRS)
+#define LAPACK_spftrs LAPACK_GLOBAL(spftrs,SPFTRS)
+#define LAPACK_zpftrs LAPACK_GLOBAL(zpftrs,ZPFTRS)
+#define LAPACK_cpftrs LAPACK_GLOBAL(cpftrs,CPFTRS)
+#define LAPACK_spptrs LAPACK_GLOBAL(spptrs,SPPTRS)
+#define LAPACK_dpptrs LAPACK_GLOBAL(dpptrs,DPPTRS)
+#define LAPACK_cpptrs LAPACK_GLOBAL(cpptrs,CPPTRS)
+#define LAPACK_zpptrs LAPACK_GLOBAL(zpptrs,ZPPTRS)
+#define LAPACK_spbtrs LAPACK_GLOBAL(spbtrs,SPBTRS)
+#define LAPACK_dpbtrs LAPACK_GLOBAL(dpbtrs,DPBTRS)
+#define LAPACK_cpbtrs LAPACK_GLOBAL(cpbtrs,CPBTRS)
+#define LAPACK_zpbtrs LAPACK_GLOBAL(zpbtrs,ZPBTRS)
+#define LAPACK_spttrs LAPACK_GLOBAL(spttrs,SPTTRS)
+#define LAPACK_dpttrs LAPACK_GLOBAL(dpttrs,DPTTRS)
+#define LAPACK_cpttrs LAPACK_GLOBAL(cpttrs,CPTTRS)
+#define LAPACK_zpttrs LAPACK_GLOBAL(zpttrs,ZPTTRS)
+#define LAPACK_ssytrs LAPACK_GLOBAL(ssytrs,SSYTRS)
+#define LAPACK_dsytrs LAPACK_GLOBAL(dsytrs,DSYTRS)
+#define LAPACK_csytrs LAPACK_GLOBAL(csytrs,CSYTRS)
+#define LAPACK_zsytrs LAPACK_GLOBAL(zsytrs,ZSYTRS)
+#define LAPACK_chetrs LAPACK_GLOBAL(chetrs,CHETRS)
+#define LAPACK_zhetrs LAPACK_GLOBAL(zhetrs,ZHETRS)
+#define LAPACK_ssptrs LAPACK_GLOBAL(ssptrs,SSPTRS)
+#define LAPACK_dsptrs LAPACK_GLOBAL(dsptrs,DSPTRS)
+#define LAPACK_csptrs LAPACK_GLOBAL(csptrs,CSPTRS)
+#define LAPACK_zsptrs LAPACK_GLOBAL(zsptrs,ZSPTRS)
+#define LAPACK_chptrs LAPACK_GLOBAL(chptrs,CHPTRS)
+#define LAPACK_zhptrs LAPACK_GLOBAL(zhptrs,ZHPTRS)
+#define LAPACK_strtrs LAPACK_GLOBAL(strtrs,STRTRS)
+#define LAPACK_dtrtrs LAPACK_GLOBAL(dtrtrs,DTRTRS)
+#define LAPACK_ctrtrs LAPACK_GLOBAL(ctrtrs,CTRTRS)
+#define LAPACK_ztrtrs LAPACK_GLOBAL(ztrtrs,ZTRTRS)
+#define LAPACK_stptrs LAPACK_GLOBAL(stptrs,STPTRS)
+#define LAPACK_dtptrs LAPACK_GLOBAL(dtptrs,DTPTRS)
+#define LAPACK_ctptrs LAPACK_GLOBAL(ctptrs,CTPTRS)
+#define LAPACK_ztptrs LAPACK_GLOBAL(ztptrs,ZTPTRS)
+#define LAPACK_stbtrs LAPACK_GLOBAL(stbtrs,STBTRS)
+#define LAPACK_dtbtrs LAPACK_GLOBAL(dtbtrs,DTBTRS)
+#define LAPACK_ctbtrs LAPACK_GLOBAL(ctbtrs,CTBTRS)
+#define LAPACK_ztbtrs LAPACK_GLOBAL(ztbtrs,ZTBTRS)
+#define LAPACK_sgecon LAPACK_GLOBAL(sgecon,SGECON)
+#define LAPACK_dgecon LAPACK_GLOBAL(dgecon,DGECON)
+#define LAPACK_cgecon LAPACK_GLOBAL(cgecon,CGECON)
+#define LAPACK_zgecon LAPACK_GLOBAL(zgecon,ZGECON)
+#define LAPACK_sgbcon LAPACK_GLOBAL(sgbcon,SGBCON)
+#define LAPACK_dgbcon LAPACK_GLOBAL(dgbcon,DGBCON)
+#define LAPACK_cgbcon LAPACK_GLOBAL(cgbcon,CGBCON)
+#define LAPACK_zgbcon LAPACK_GLOBAL(zgbcon,ZGBCON)
+#define LAPACK_sgtcon LAPACK_GLOBAL(sgtcon,SGTCON)
+#define LAPACK_dgtcon LAPACK_GLOBAL(dgtcon,DGTCON)
+#define LAPACK_cgtcon LAPACK_GLOBAL(cgtcon,CGTCON)
+#define LAPACK_zgtcon LAPACK_GLOBAL(zgtcon,ZGTCON)
+#define LAPACK_spocon LAPACK_GLOBAL(spocon,SPOCON)
+#define LAPACK_dpocon LAPACK_GLOBAL(dpocon,DPOCON)
+#define LAPACK_cpocon LAPACK_GLOBAL(cpocon,CPOCON)
+#define LAPACK_zpocon LAPACK_GLOBAL(zpocon,ZPOCON)
+#define LAPACK_sppcon LAPACK_GLOBAL(sppcon,SPPCON)
+#define LAPACK_dppcon LAPACK_GLOBAL(dppcon,DPPCON)
+#define LAPACK_cppcon LAPACK_GLOBAL(cppcon,CPPCON)
+#define LAPACK_zppcon LAPACK_GLOBAL(zppcon,ZPPCON)
+#define LAPACK_spbcon LAPACK_GLOBAL(spbcon,SPBCON)
+#define LAPACK_dpbcon LAPACK_GLOBAL(dpbcon,DPBCON)
+#define LAPACK_cpbcon LAPACK_GLOBAL(cpbcon,CPBCON)
+#define LAPACK_zpbcon LAPACK_GLOBAL(zpbcon,ZPBCON)
+#define LAPACK_sptcon LAPACK_GLOBAL(sptcon,SPTCON)
+#define LAPACK_dptcon LAPACK_GLOBAL(dptcon,DPTCON)
+#define LAPACK_cptcon LAPACK_GLOBAL(cptcon,CPTCON)
+#define LAPACK_zptcon LAPACK_GLOBAL(zptcon,ZPTCON)
+#define LAPACK_ssycon LAPACK_GLOBAL(ssycon,SSYCON)
+#define LAPACK_dsycon LAPACK_GLOBAL(dsycon,DSYCON)
+#define LAPACK_csycon LAPACK_GLOBAL(csycon,CSYCON)
+#define LAPACK_zsycon LAPACK_GLOBAL(zsycon,ZSYCON)
+#define LAPACK_checon LAPACK_GLOBAL(checon,CHECON)
+#define LAPACK_zhecon LAPACK_GLOBAL(zhecon,ZHECON)
+#define LAPACK_sspcon LAPACK_GLOBAL(sspcon,SSPCON)
+#define LAPACK_dspcon LAPACK_GLOBAL(dspcon,DSPCON)
+#define LAPACK_cspcon LAPACK_GLOBAL(cspcon,CSPCON)
+#define LAPACK_zspcon LAPACK_GLOBAL(zspcon,ZSPCON)
+#define LAPACK_chpcon LAPACK_GLOBAL(chpcon,CHPCON)
+#define LAPACK_zhpcon LAPACK_GLOBAL(zhpcon,ZHPCON)
+#define LAPACK_strcon LAPACK_GLOBAL(strcon,STRCON)
+#define LAPACK_dtrcon LAPACK_GLOBAL(dtrcon,DTRCON)
+#define LAPACK_ctrcon LAPACK_GLOBAL(ctrcon,CTRCON)
+#define LAPACK_ztrcon LAPACK_GLOBAL(ztrcon,ZTRCON)
+#define LAPACK_stpcon LAPACK_GLOBAL(stpcon,STPCON)
+#define LAPACK_dtpcon LAPACK_GLOBAL(dtpcon,DTPCON)
+#define LAPACK_ctpcon LAPACK_GLOBAL(ctpcon,CTPCON)
+#define LAPACK_ztpcon LAPACK_GLOBAL(ztpcon,ZTPCON)
+#define LAPACK_stbcon LAPACK_GLOBAL(stbcon,STBCON)
+#define LAPACK_dtbcon LAPACK_GLOBAL(dtbcon,DTBCON)
+#define LAPACK_ctbcon LAPACK_GLOBAL(ctbcon,CTBCON)
+#define LAPACK_ztbcon LAPACK_GLOBAL(ztbcon,ZTBCON)
+#define LAPACK_sgerfs LAPACK_GLOBAL(sgerfs,SGERFS)
+#define LAPACK_dgerfs LAPACK_GLOBAL(dgerfs,DGERFS)
+#define LAPACK_cgerfs LAPACK_GLOBAL(cgerfs,CGERFS)
+#define LAPACK_zgerfs LAPACK_GLOBAL(zgerfs,ZGERFS)
+#define LAPACK_dgerfsx LAPACK_GLOBAL(dgerfsx,DGERFSX)
+#define LAPACK_sgerfsx LAPACK_GLOBAL(sgerfsx,SGERFSX)
+#define LAPACK_zgerfsx LAPACK_GLOBAL(zgerfsx,ZGERFSX)
+#define LAPACK_cgerfsx LAPACK_GLOBAL(cgerfsx,CGERFSX)
+#define LAPACK_sgbrfs LAPACK_GLOBAL(sgbrfs,SGBRFS)
+#define LAPACK_dgbrfs LAPACK_GLOBAL(dgbrfs,DGBRFS)
+#define LAPACK_cgbrfs LAPACK_GLOBAL(cgbrfs,CGBRFS)
+#define LAPACK_zgbrfs LAPACK_GLOBAL(zgbrfs,ZGBRFS)
+#define LAPACK_dgbrfsx LAPACK_GLOBAL(dgbrfsx,DGBRFSX)
+#define LAPACK_sgbrfsx LAPACK_GLOBAL(sgbrfsx,SGBRFSX)
+#define LAPACK_zgbrfsx LAPACK_GLOBAL(zgbrfsx,ZGBRFSX)
+#define LAPACK_cgbrfsx LAPACK_GLOBAL(cgbrfsx,CGBRFSX)
+#define LAPACK_sgtrfs LAPACK_GLOBAL(sgtrfs,SGTRFS)
+#define LAPACK_dgtrfs LAPACK_GLOBAL(dgtrfs,DGTRFS)
+#define LAPACK_cgtrfs LAPACK_GLOBAL(cgtrfs,CGTRFS)
+#define LAPACK_zgtrfs LAPACK_GLOBAL(zgtrfs,ZGTRFS)
+#define LAPACK_sporfs LAPACK_GLOBAL(sporfs,SPORFS)
+#define LAPACK_dporfs LAPACK_GLOBAL(dporfs,DPORFS)
+#define LAPACK_cporfs LAPACK_GLOBAL(cporfs,CPORFS)
+#define LAPACK_zporfs LAPACK_GLOBAL(zporfs,ZPORFS)
+#define LAPACK_dporfsx LAPACK_GLOBAL(dporfsx,DPORFSX)
+#define LAPACK_sporfsx LAPACK_GLOBAL(sporfsx,SPORFSX)
+#define LAPACK_zporfsx LAPACK_GLOBAL(zporfsx,ZPORFSX)
+#define LAPACK_cporfsx LAPACK_GLOBAL(cporfsx,CPORFSX)
+#define LAPACK_spprfs LAPACK_GLOBAL(spprfs,SPPRFS)
+#define LAPACK_dpprfs LAPACK_GLOBAL(dpprfs,DPPRFS)
+#define LAPACK_cpprfs LAPACK_GLOBAL(cpprfs,CPPRFS)
+#define LAPACK_zpprfs LAPACK_GLOBAL(zpprfs,ZPPRFS)
+#define LAPACK_spbrfs LAPACK_GLOBAL(spbrfs,SPBRFS)
+#define LAPACK_dpbrfs LAPACK_GLOBAL(dpbrfs,DPBRFS)
+#define LAPACK_cpbrfs LAPACK_GLOBAL(cpbrfs,CPBRFS)
+#define LAPACK_zpbrfs LAPACK_GLOBAL(zpbrfs,ZPBRFS)
+#define LAPACK_sptrfs LAPACK_GLOBAL(sptrfs,SPTRFS)
+#define LAPACK_dptrfs LAPACK_GLOBAL(dptrfs,DPTRFS)
+#define LAPACK_cptrfs LAPACK_GLOBAL(cptrfs,CPTRFS)
+#define LAPACK_zptrfs LAPACK_GLOBAL(zptrfs,ZPTRFS)
+#define LAPACK_ssyrfs LAPACK_GLOBAL(ssyrfs,SSYRFS)
+#define LAPACK_dsyrfs LAPACK_GLOBAL(dsyrfs,DSYRFS)
+#define LAPACK_csyrfs LAPACK_GLOBAL(csyrfs,CSYRFS)
+#define LAPACK_zsyrfs LAPACK_GLOBAL(zsyrfs,ZSYRFS)
+#define LAPACK_dsyrfsx LAPACK_GLOBAL(dsyrfsx,DSYRFSX)
+#define LAPACK_ssyrfsx LAPACK_GLOBAL(ssyrfsx,SSYRFSX)
+#define LAPACK_zsyrfsx LAPACK_GLOBAL(zsyrfsx,ZSYRFSX)
+#define LAPACK_csyrfsx LAPACK_GLOBAL(csyrfsx,CSYRFSX)
+#define LAPACK_cherfs LAPACK_GLOBAL(cherfs,CHERFS)
+#define LAPACK_zherfs LAPACK_GLOBAL(zherfs,ZHERFS)
+#define LAPACK_zherfsx LAPACK_GLOBAL(zherfsx,ZHERFSX)
+#define LAPACK_cherfsx LAPACK_GLOBAL(cherfsx,CHERFSX)
+#define LAPACK_ssprfs LAPACK_GLOBAL(ssprfs,SSPRFS)
+#define LAPACK_dsprfs LAPACK_GLOBAL(dsprfs,DSPRFS)
+#define LAPACK_csprfs LAPACK_GLOBAL(csprfs,CSPRFS)
+#define LAPACK_zsprfs LAPACK_GLOBAL(zsprfs,ZSPRFS)
+#define LAPACK_chprfs LAPACK_GLOBAL(chprfs,CHPRFS)
+#define LAPACK_zhprfs LAPACK_GLOBAL(zhprfs,ZHPRFS)
+#define LAPACK_strrfs LAPACK_GLOBAL(strrfs,STRRFS)
+#define LAPACK_dtrrfs LAPACK_GLOBAL(dtrrfs,DTRRFS)
+#define LAPACK_ctrrfs LAPACK_GLOBAL(ctrrfs,CTRRFS)
+#define LAPACK_ztrrfs LAPACK_GLOBAL(ztrrfs,ZTRRFS)
+#define LAPACK_stprfs LAPACK_GLOBAL(stprfs,STPRFS)
+#define LAPACK_dtprfs LAPACK_GLOBAL(dtprfs,DTPRFS)
+#define LAPACK_ctprfs LAPACK_GLOBAL(ctprfs,CTPRFS)
+#define LAPACK_ztprfs LAPACK_GLOBAL(ztprfs,ZTPRFS)
+#define LAPACK_stbrfs LAPACK_GLOBAL(stbrfs,STBRFS)
+#define LAPACK_dtbrfs LAPACK_GLOBAL(dtbrfs,DTBRFS)
+#define LAPACK_ctbrfs LAPACK_GLOBAL(ctbrfs,CTBRFS)
+#define LAPACK_ztbrfs LAPACK_GLOBAL(ztbrfs,ZTBRFS)
+#define LAPACK_sgetri LAPACK_GLOBAL(sgetri,SGETRI)
+#define LAPACK_dgetri LAPACK_GLOBAL(dgetri,DGETRI)
+#define LAPACK_cgetri LAPACK_GLOBAL(cgetri,CGETRI)
+#define LAPACK_zgetri LAPACK_GLOBAL(zgetri,ZGETRI)
+#define LAPACK_spotri LAPACK_GLOBAL(spotri,SPOTRI)
+#define LAPACK_dpotri LAPACK_GLOBAL(dpotri,DPOTRI)
+#define LAPACK_cpotri LAPACK_GLOBAL(cpotri,CPOTRI)
+#define LAPACK_zpotri LAPACK_GLOBAL(zpotri,ZPOTRI)
+#define LAPACK_dpftri LAPACK_GLOBAL(dpftri,DPFTRI)
+#define LAPACK_spftri LAPACK_GLOBAL(spftri,SPFTRI)
+#define LAPACK_zpftri LAPACK_GLOBAL(zpftri,ZPFTRI)
+#define LAPACK_cpftri LAPACK_GLOBAL(cpftri,CPFTRI)
+#define LAPACK_spptri LAPACK_GLOBAL(spptri,SPPTRI)
+#define LAPACK_dpptri LAPACK_GLOBAL(dpptri,DPPTRI)
+#define LAPACK_cpptri LAPACK_GLOBAL(cpptri,CPPTRI)
+#define LAPACK_zpptri LAPACK_GLOBAL(zpptri,ZPPTRI)
+#define LAPACK_ssytri LAPACK_GLOBAL(ssytri,SSYTRI)
+#define LAPACK_dsytri LAPACK_GLOBAL(dsytri,DSYTRI)
+#define LAPACK_csytri LAPACK_GLOBAL(csytri,CSYTRI)
+#define LAPACK_zsytri LAPACK_GLOBAL(zsytri,ZSYTRI)
+#define LAPACK_chetri LAPACK_GLOBAL(chetri,CHETRI)
+#define LAPACK_zhetri LAPACK_GLOBAL(zhetri,ZHETRI)
+#define LAPACK_ssptri LAPACK_GLOBAL(ssptri,SSPTRI)
+#define LAPACK_dsptri LAPACK_GLOBAL(dsptri,DSPTRI)
+#define LAPACK_csptri LAPACK_GLOBAL(csptri,CSPTRI)
+#define LAPACK_zsptri LAPACK_GLOBAL(zsptri,ZSPTRI)
+#define LAPACK_chptri LAPACK_GLOBAL(chptri,CHPTRI)
+#define LAPACK_zhptri LAPACK_GLOBAL(zhptri,ZHPTRI)
+#define LAPACK_strtri LAPACK_GLOBAL(strtri,STRTRI)
+#define LAPACK_dtrtri LAPACK_GLOBAL(dtrtri,DTRTRI)
+#define LAPACK_ctrtri LAPACK_GLOBAL(ctrtri,CTRTRI)
+#define LAPACK_ztrtri LAPACK_GLOBAL(ztrtri,ZTRTRI)
+#define LAPACK_dtftri LAPACK_GLOBAL(dtftri,DTFTRI)
+#define LAPACK_stftri LAPACK_GLOBAL(stftri,STFTRI)
+#define LAPACK_ztftri LAPACK_GLOBAL(ztftri,ZTFTRI)
+#define LAPACK_ctftri LAPACK_GLOBAL(ctftri,CTFTRI)
+#define LAPACK_stptri LAPACK_GLOBAL(stptri,STPTRI)
+#define LAPACK_dtptri LAPACK_GLOBAL(dtptri,DTPTRI)
+#define LAPACK_ctptri LAPACK_GLOBAL(ctptri,CTPTRI)
+#define LAPACK_ztptri LAPACK_GLOBAL(ztptri,ZTPTRI)
+#define LAPACK_sgeequ LAPACK_GLOBAL(sgeequ,SGEEQU)
+#define LAPACK_dgeequ LAPACK_GLOBAL(dgeequ,DGEEQU)
+#define LAPACK_cgeequ LAPACK_GLOBAL(cgeequ,CGEEQU)
+#define LAPACK_zgeequ LAPACK_GLOBAL(zgeequ,ZGEEQU)
+#define LAPACK_dgeequb LAPACK_GLOBAL(dgeequb,DGEEQUB)
+#define LAPACK_sgeequb LAPACK_GLOBAL(sgeequb,SGEEQUB)
+#define LAPACK_zgeequb LAPACK_GLOBAL(zgeequb,ZGEEQUB)
+#define LAPACK_cgeequb LAPACK_GLOBAL(cgeequb,CGEEQUB)
+#define LAPACK_sgbequ LAPACK_GLOBAL(sgbequ,SGBEQU)
+#define LAPACK_dgbequ LAPACK_GLOBAL(dgbequ,DGBEQU)
+#define LAPACK_cgbequ LAPACK_GLOBAL(cgbequ,CGBEQU)
+#define LAPACK_zgbequ LAPACK_GLOBAL(zgbequ,ZGBEQU)
+#define LAPACK_dgbequb LAPACK_GLOBAL(dgbequb,DGBEQUB)
+#define LAPACK_sgbequb LAPACK_GLOBAL(sgbequb,SGBEQUB)
+#define LAPACK_zgbequb LAPACK_GLOBAL(zgbequb,ZGBEQUB)
+#define LAPACK_cgbequb LAPACK_GLOBAL(cgbequb,CGBEQUB)
+#define LAPACK_spoequ LAPACK_GLOBAL(spoequ,SPOEQU)
+#define LAPACK_dpoequ LAPACK_GLOBAL(dpoequ,DPOEQU)
+#define LAPACK_cpoequ LAPACK_GLOBAL(cpoequ,CPOEQU)
+#define LAPACK_zpoequ LAPACK_GLOBAL(zpoequ,ZPOEQU)
+#define LAPACK_dpoequb LAPACK_GLOBAL(dpoequb,DPOEQUB)
+#define LAPACK_spoequb LAPACK_GLOBAL(spoequb,SPOEQUB)
+#define LAPACK_zpoequb LAPACK_GLOBAL(zpoequb,ZPOEQUB)
+#define LAPACK_cpoequb LAPACK_GLOBAL(cpoequb,CPOEQUB)
+#define LAPACK_sppequ LAPACK_GLOBAL(sppequ,SPPEQU)
+#define LAPACK_dppequ LAPACK_GLOBAL(dppequ,DPPEQU)
+#define LAPACK_cppequ LAPACK_GLOBAL(cppequ,CPPEQU)
+#define LAPACK_zppequ LAPACK_GLOBAL(zppequ,ZPPEQU)
+#define LAPACK_spbequ LAPACK_GLOBAL(spbequ,SPBEQU)
+#define LAPACK_dpbequ LAPACK_GLOBAL(dpbequ,DPBEQU)
+#define LAPACK_cpbequ LAPACK_GLOBAL(cpbequ,CPBEQU)
+#define LAPACK_zpbequ LAPACK_GLOBAL(zpbequ,ZPBEQU)
+#define LAPACK_dsyequb LAPACK_GLOBAL(dsyequb,DSYEQUB)
+#define LAPACK_ssyequb LAPACK_GLOBAL(ssyequb,SSYEQUB)
+#define LAPACK_zsyequb LAPACK_GLOBAL(zsyequb,ZSYEQUB)
+#define LAPACK_csyequb LAPACK_GLOBAL(csyequb,CSYEQUB)
+#define LAPACK_zheequb LAPACK_GLOBAL(zheequb,ZHEEQUB)
+#define LAPACK_cheequb LAPACK_GLOBAL(cheequb,CHEEQUB)
+#define LAPACK_sgesv LAPACK_GLOBAL(sgesv,SGESV)
+#define LAPACK_dgesv LAPACK_GLOBAL(dgesv,DGESV)
+#define LAPACK_cgesv LAPACK_GLOBAL(cgesv,CGESV)
+#define LAPACK_zgesv LAPACK_GLOBAL(zgesv,ZGESV)
+#define LAPACK_dsgesv LAPACK_GLOBAL(dsgesv,DSGESV)
+#define LAPACK_zcgesv LAPACK_GLOBAL(zcgesv,ZCGESV)
+#define LAPACK_sgesvx LAPACK_GLOBAL(sgesvx,SGESVX)
+#define LAPACK_dgesvx LAPACK_GLOBAL(dgesvx,DGESVX)
+#define LAPACK_cgesvx LAPACK_GLOBAL(cgesvx,CGESVX)
+#define LAPACK_zgesvx LAPACK_GLOBAL(zgesvx,ZGESVX)
+#define LAPACK_dgesvxx LAPACK_GLOBAL(dgesvxx,DGESVXX)
+#define LAPACK_sgesvxx LAPACK_GLOBAL(sgesvxx,SGESVXX)
+#define LAPACK_zgesvxx LAPACK_GLOBAL(zgesvxx,ZGESVXX)
+#define LAPACK_cgesvxx LAPACK_GLOBAL(cgesvxx,CGESVXX)
+#define LAPACK_sgbsv LAPACK_GLOBAL(sgbsv,SGBSV)
+#define LAPACK_dgbsv LAPACK_GLOBAL(dgbsv,DGBSV)
+#define LAPACK_cgbsv LAPACK_GLOBAL(cgbsv,CGBSV)
+#define LAPACK_zgbsv LAPACK_GLOBAL(zgbsv,ZGBSV)
+#define LAPACK_sgbsvx LAPACK_GLOBAL(sgbsvx,SGBSVX)
+#define LAPACK_dgbsvx LAPACK_GLOBAL(dgbsvx,DGBSVX)
+#define LAPACK_cgbsvx LAPACK_GLOBAL(cgbsvx,CGBSVX)
+#define LAPACK_zgbsvx LAPACK_GLOBAL(zgbsvx,ZGBSVX)
+#define LAPACK_dgbsvxx LAPACK_GLOBAL(dgbsvxx,DGBSVXX)
+#define LAPACK_sgbsvxx LAPACK_GLOBAL(sgbsvxx,SGBSVXX)
+#define LAPACK_zgbsvxx LAPACK_GLOBAL(zgbsvxx,ZGBSVXX)
+#define LAPACK_cgbsvxx LAPACK_GLOBAL(cgbsvxx,CGBSVXX)
+#define LAPACK_sgtsv LAPACK_GLOBAL(sgtsv,SGTSV)
+#define LAPACK_dgtsv LAPACK_GLOBAL(dgtsv,DGTSV)
+#define LAPACK_cgtsv LAPACK_GLOBAL(cgtsv,CGTSV)
+#define LAPACK_zgtsv LAPACK_GLOBAL(zgtsv,ZGTSV)
+#define LAPACK_sgtsvx LAPACK_GLOBAL(sgtsvx,SGTSVX)
+#define LAPACK_dgtsvx LAPACK_GLOBAL(dgtsvx,DGTSVX)
+#define LAPACK_cgtsvx LAPACK_GLOBAL(cgtsvx,CGTSVX)
+#define LAPACK_zgtsvx LAPACK_GLOBAL(zgtsvx,ZGTSVX)
+#define LAPACK_sposv LAPACK_GLOBAL(sposv,SPOSV)
+#define LAPACK_dposv LAPACK_GLOBAL(dposv,DPOSV)
+#define LAPACK_cposv LAPACK_GLOBAL(cposv,CPOSV)
+#define LAPACK_zposv LAPACK_GLOBAL(zposv,ZPOSV)
+#define LAPACK_dsposv LAPACK_GLOBAL(dsposv,DSPOSV)
+#define LAPACK_zcposv LAPACK_GLOBAL(zcposv,ZCPOSV)
+#define LAPACK_sposvx LAPACK_GLOBAL(sposvx,SPOSVX)
+#define LAPACK_dposvx LAPACK_GLOBAL(dposvx,DPOSVX)
+#define LAPACK_cposvx LAPACK_GLOBAL(cposvx,CPOSVX)
+#define LAPACK_zposvx LAPACK_GLOBAL(zposvx,ZPOSVX)
+#define LAPACK_dposvxx LAPACK_GLOBAL(dposvxx,DPOSVXX)
+#define LAPACK_sposvxx LAPACK_GLOBAL(sposvxx,SPOSVXX)
+#define LAPACK_zposvxx LAPACK_GLOBAL(zposvxx,ZPOSVXX)
+#define LAPACK_cposvxx LAPACK_GLOBAL(cposvxx,CPOSVXX)
+#define LAPACK_sppsv LAPACK_GLOBAL(sppsv,SPPSV)
+#define LAPACK_dppsv LAPACK_GLOBAL(dppsv,DPPSV)
+#define LAPACK_cppsv LAPACK_GLOBAL(cppsv,CPPSV)
+#define LAPACK_zppsv LAPACK_GLOBAL(zppsv,ZPPSV)
+#define LAPACK_sppsvx LAPACK_GLOBAL(sppsvx,SPPSVX)
+#define LAPACK_dppsvx LAPACK_GLOBAL(dppsvx,DPPSVX)
+#define LAPACK_cppsvx LAPACK_GLOBAL(cppsvx,CPPSVX)
+#define LAPACK_zppsvx LAPACK_GLOBAL(zppsvx,ZPPSVX)
+#define LAPACK_spbsv LAPACK_GLOBAL(spbsv,SPBSV)
+#define LAPACK_dpbsv LAPACK_GLOBAL(dpbsv,DPBSV)
+#define LAPACK_cpbsv LAPACK_GLOBAL(cpbsv,CPBSV)
+#define LAPACK_zpbsv LAPACK_GLOBAL(zpbsv,ZPBSV)
+#define LAPACK_spbsvx LAPACK_GLOBAL(spbsvx,SPBSVX)
+#define LAPACK_dpbsvx LAPACK_GLOBAL(dpbsvx,DPBSVX)
+#define LAPACK_cpbsvx LAPACK_GLOBAL(cpbsvx,CPBSVX)
+#define LAPACK_zpbsvx LAPACK_GLOBAL(zpbsvx,ZPBSVX)
+#define LAPACK_sptsv LAPACK_GLOBAL(sptsv,SPTSV)
+#define LAPACK_dptsv LAPACK_GLOBAL(dptsv,DPTSV)
+#define LAPACK_cptsv LAPACK_GLOBAL(cptsv,CPTSV)
+#define LAPACK_zptsv LAPACK_GLOBAL(zptsv,ZPTSV)
+#define LAPACK_sptsvx LAPACK_GLOBAL(sptsvx,SPTSVX)
+#define LAPACK_dptsvx LAPACK_GLOBAL(dptsvx,DPTSVX)
+#define LAPACK_cptsvx LAPACK_GLOBAL(cptsvx,CPTSVX)
+#define LAPACK_zptsvx LAPACK_GLOBAL(zptsvx,ZPTSVX)
+#define LAPACK_ssysv LAPACK_GLOBAL(ssysv,SSYSV)
+#define LAPACK_dsysv LAPACK_GLOBAL(dsysv,DSYSV)
+#define LAPACK_csysv LAPACK_GLOBAL(csysv,CSYSV)
+#define LAPACK_zsysv LAPACK_GLOBAL(zsysv,ZSYSV)
+#define LAPACK_ssysvx LAPACK_GLOBAL(ssysvx,SSYSVX)
+#define LAPACK_dsysvx LAPACK_GLOBAL(dsysvx,DSYSVX)
+#define LAPACK_csysvx LAPACK_GLOBAL(csysvx,CSYSVX)
+#define LAPACK_zsysvx LAPACK_GLOBAL(zsysvx,ZSYSVX)
+#define LAPACK_dsysvxx LAPACK_GLOBAL(dsysvxx,DSYSVXX)
+#define LAPACK_ssysvxx LAPACK_GLOBAL(ssysvxx,SSYSVXX)
+#define LAPACK_zsysvxx LAPACK_GLOBAL(zsysvxx,ZSYSVXX)
+#define LAPACK_csysvxx LAPACK_GLOBAL(csysvxx,CSYSVXX)
+#define LAPACK_chesv LAPACK_GLOBAL(chesv,CHESV)
+#define LAPACK_zhesv LAPACK_GLOBAL(zhesv,ZHESV)
+#define LAPACK_chesvx LAPACK_GLOBAL(chesvx,CHESVX)
+#define LAPACK_zhesvx LAPACK_GLOBAL(zhesvx,ZHESVX)
+#define LAPACK_zhesvxx LAPACK_GLOBAL(zhesvxx,ZHESVXX)
+#define LAPACK_chesvxx LAPACK_GLOBAL(chesvxx,CHESVXX)
+#define LAPACK_sspsv LAPACK_GLOBAL(sspsv,SSPSV)
+#define LAPACK_dspsv LAPACK_GLOBAL(dspsv,DSPSV)
+#define LAPACK_cspsv LAPACK_GLOBAL(cspsv,CSPSV)
+#define LAPACK_zspsv LAPACK_GLOBAL(zspsv,ZSPSV)
+#define LAPACK_sspsvx LAPACK_GLOBAL(sspsvx,SSPSVX)
+#define LAPACK_dspsvx LAPACK_GLOBAL(dspsvx,DSPSVX)
+#define LAPACK_cspsvx LAPACK_GLOBAL(cspsvx,CSPSVX)
+#define LAPACK_zspsvx LAPACK_GLOBAL(zspsvx,ZSPSVX)
+#define LAPACK_chpsv LAPACK_GLOBAL(chpsv,CHPSV)
+#define LAPACK_zhpsv LAPACK_GLOBAL(zhpsv,ZHPSV)
+#define LAPACK_chpsvx LAPACK_GLOBAL(chpsvx,CHPSVX)
+#define LAPACK_zhpsvx LAPACK_GLOBAL(zhpsvx,ZHPSVX)
+#define LAPACK_sgeqrf LAPACK_GLOBAL(sgeqrf,SGEQRF)
+#define LAPACK_dgeqrf LAPACK_GLOBAL(dgeqrf,DGEQRF)
+#define LAPACK_cgeqrf LAPACK_GLOBAL(cgeqrf,CGEQRF)
+#define LAPACK_zgeqrf LAPACK_GLOBAL(zgeqrf,ZGEQRF)
+#define LAPACK_sgeqpf LAPACK_GLOBAL(sgeqpf,SGEQPF)
+#define LAPACK_dgeqpf LAPACK_GLOBAL(dgeqpf,DGEQPF)
+#define LAPACK_cgeqpf LAPACK_GLOBAL(cgeqpf,CGEQPF)
+#define LAPACK_zgeqpf LAPACK_GLOBAL(zgeqpf,ZGEQPF)
+#define LAPACK_sgeqp3 LAPACK_GLOBAL(sgeqp3,SGEQP3)
+#define LAPACK_dgeqp3 LAPACK_GLOBAL(dgeqp3,DGEQP3)
+#define LAPACK_cgeqp3 LAPACK_GLOBAL(cgeqp3,CGEQP3)
+#define LAPACK_zgeqp3 LAPACK_GLOBAL(zgeqp3,ZGEQP3)
+#define LAPACK_sorgqr LAPACK_GLOBAL(sorgqr,SORGQR)
+#define LAPACK_dorgqr LAPACK_GLOBAL(dorgqr,DORGQR)
+#define LAPACK_sormqr LAPACK_GLOBAL(sormqr,SORMQR)
+#define LAPACK_dormqr LAPACK_GLOBAL(dormqr,DORMQR)
+#define LAPACK_cungqr LAPACK_GLOBAL(cungqr,CUNGQR)
+#define LAPACK_zungqr LAPACK_GLOBAL(zungqr,ZUNGQR)
+#define LAPACK_cunmqr LAPACK_GLOBAL(cunmqr,CUNMQR)
+#define LAPACK_zunmqr LAPACK_GLOBAL(zunmqr,ZUNMQR)
+#define LAPACK_sgelqf LAPACK_GLOBAL(sgelqf,SGELQF)
+#define LAPACK_dgelqf LAPACK_GLOBAL(dgelqf,DGELQF)
+#define LAPACK_cgelqf LAPACK_GLOBAL(cgelqf,CGELQF)
+#define LAPACK_zgelqf LAPACK_GLOBAL(zgelqf,ZGELQF)
+#define LAPACK_sorglq LAPACK_GLOBAL(sorglq,SORGLQ)
+#define LAPACK_dorglq LAPACK_GLOBAL(dorglq,DORGLQ)
+#define LAPACK_sormlq LAPACK_GLOBAL(sormlq,SORMLQ)
+#define LAPACK_dormlq LAPACK_GLOBAL(dormlq,DORMLQ)
+#define LAPACK_cunglq LAPACK_GLOBAL(cunglq,CUNGLQ)
+#define LAPACK_zunglq LAPACK_GLOBAL(zunglq,ZUNGLQ)
+#define LAPACK_cunmlq LAPACK_GLOBAL(cunmlq,CUNMLQ)
+#define LAPACK_zunmlq LAPACK_GLOBAL(zunmlq,ZUNMLQ)
+#define LAPACK_sgeqlf LAPACK_GLOBAL(sgeqlf,SGEQLF)
+#define LAPACK_dgeqlf LAPACK_GLOBAL(dgeqlf,DGEQLF)
+#define LAPACK_cgeqlf LAPACK_GLOBAL(cgeqlf,CGEQLF)
+#define LAPACK_zgeqlf LAPACK_GLOBAL(zgeqlf,ZGEQLF)
+#define LAPACK_sorgql LAPACK_GLOBAL(sorgql,SORGQL)
+#define LAPACK_dorgql LAPACK_GLOBAL(dorgql,DORGQL)
+#define LAPACK_cungql LAPACK_GLOBAL(cungql,CUNGQL)
+#define LAPACK_zungql LAPACK_GLOBAL(zungql,ZUNGQL)
+#define LAPACK_sormql LAPACK_GLOBAL(sormql,SORMQL)
+#define LAPACK_dormql LAPACK_GLOBAL(dormql,DORMQL)
+#define LAPACK_cunmql LAPACK_GLOBAL(cunmql,CUNMQL)
+#define LAPACK_zunmql LAPACK_GLOBAL(zunmql,ZUNMQL)
+#define LAPACK_sgerqf LAPACK_GLOBAL(sgerqf,SGERQF)
+#define LAPACK_dgerqf LAPACK_GLOBAL(dgerqf,DGERQF)
+#define LAPACK_cgerqf LAPACK_GLOBAL(cgerqf,CGERQF)
+#define LAPACK_zgerqf LAPACK_GLOBAL(zgerqf,ZGERQF)
+#define LAPACK_sorgrq LAPACK_GLOBAL(sorgrq,SORGRQ)
+#define LAPACK_dorgrq LAPACK_GLOBAL(dorgrq,DORGRQ)
+#define LAPACK_cungrq LAPACK_GLOBAL(cungrq,CUNGRQ)
+#define LAPACK_zungrq LAPACK_GLOBAL(zungrq,ZUNGRQ)
+#define LAPACK_sormrq LAPACK_GLOBAL(sormrq,SORMRQ)
+#define LAPACK_dormrq LAPACK_GLOBAL(dormrq,DORMRQ)
+#define LAPACK_cunmrq LAPACK_GLOBAL(cunmrq,CUNMRQ)
+#define LAPACK_zunmrq LAPACK_GLOBAL(zunmrq,ZUNMRQ)
+#define LAPACK_stzrzf LAPACK_GLOBAL(stzrzf,STZRZF)
+#define LAPACK_dtzrzf LAPACK_GLOBAL(dtzrzf,DTZRZF)
+#define LAPACK_ctzrzf LAPACK_GLOBAL(ctzrzf,CTZRZF)
+#define LAPACK_ztzrzf LAPACK_GLOBAL(ztzrzf,ZTZRZF)
+#define LAPACK_sormrz LAPACK_GLOBAL(sormrz,SORMRZ)
+#define LAPACK_dormrz LAPACK_GLOBAL(dormrz,DORMRZ)
+#define LAPACK_cunmrz LAPACK_GLOBAL(cunmrz,CUNMRZ)
+#define LAPACK_zunmrz LAPACK_GLOBAL(zunmrz,ZUNMRZ)
+#define LAPACK_sggqrf LAPACK_GLOBAL(sggqrf,SGGQRF)
+#define LAPACK_dggqrf LAPACK_GLOBAL(dggqrf,DGGQRF)
+#define LAPACK_cggqrf LAPACK_GLOBAL(cggqrf,CGGQRF)
+#define LAPACK_zggqrf LAPACK_GLOBAL(zggqrf,ZGGQRF)
+#define LAPACK_sggrqf LAPACK_GLOBAL(sggrqf,SGGRQF)
+#define LAPACK_dggrqf LAPACK_GLOBAL(dggrqf,DGGRQF)
+#define LAPACK_cggrqf LAPACK_GLOBAL(cggrqf,CGGRQF)
+#define LAPACK_zggrqf LAPACK_GLOBAL(zggrqf,ZGGRQF)
+#define LAPACK_sgebrd LAPACK_GLOBAL(sgebrd,SGEBRD)
+#define LAPACK_dgebrd LAPACK_GLOBAL(dgebrd,DGEBRD)
+#define LAPACK_cgebrd LAPACK_GLOBAL(cgebrd,CGEBRD)
+#define LAPACK_zgebrd LAPACK_GLOBAL(zgebrd,ZGEBRD)
+#define LAPACK_sgbbrd LAPACK_GLOBAL(sgbbrd,SGBBRD)
+#define LAPACK_dgbbrd LAPACK_GLOBAL(dgbbrd,DGBBRD)
+#define LAPACK_cgbbrd LAPACK_GLOBAL(cgbbrd,CGBBRD)
+#define LAPACK_zgbbrd LAPACK_GLOBAL(zgbbrd,ZGBBRD)
+#define LAPACK_sorgbr LAPACK_GLOBAL(sorgbr,SORGBR)
+#define LAPACK_dorgbr LAPACK_GLOBAL(dorgbr,DORGBR)
+#define LAPACK_sormbr LAPACK_GLOBAL(sormbr,SORMBR)
+#define LAPACK_dormbr LAPACK_GLOBAL(dormbr,DORMBR)
+#define LAPACK_cungbr LAPACK_GLOBAL(cungbr,CUNGBR)
+#define LAPACK_zungbr LAPACK_GLOBAL(zungbr,ZUNGBR)
+#define LAPACK_cunmbr LAPACK_GLOBAL(cunmbr,CUNMBR)
+#define LAPACK_zunmbr LAPACK_GLOBAL(zunmbr,ZUNMBR)
+#define LAPACK_sbdsqr LAPACK_GLOBAL(sbdsqr,SBDSQR)
+#define LAPACK_dbdsqr LAPACK_GLOBAL(dbdsqr,DBDSQR)
+#define LAPACK_cbdsqr LAPACK_GLOBAL(cbdsqr,CBDSQR)
+#define LAPACK_zbdsqr LAPACK_GLOBAL(zbdsqr,ZBDSQR)
+#define LAPACK_sbdsdc LAPACK_GLOBAL(sbdsdc,SBDSDC)
+#define LAPACK_dbdsdc LAPACK_GLOBAL(dbdsdc,DBDSDC)
+#define LAPACK_ssytrd LAPACK_GLOBAL(ssytrd,SSYTRD)
+#define LAPACK_dsytrd LAPACK_GLOBAL(dsytrd,DSYTRD)
+#define LAPACK_sorgtr LAPACK_GLOBAL(sorgtr,SORGTR)
+#define LAPACK_dorgtr LAPACK_GLOBAL(dorgtr,DORGTR)
+#define LAPACK_sormtr LAPACK_GLOBAL(sormtr,SORMTR)
+#define LAPACK_dormtr LAPACK_GLOBAL(dormtr,DORMTR)
+#define LAPACK_chetrd LAPACK_GLOBAL(chetrd,CHETRD)
+#define LAPACK_zhetrd LAPACK_GLOBAL(zhetrd,ZHETRD)
+#define LAPACK_cungtr LAPACK_GLOBAL(cungtr,CUNGTR)
+#define LAPACK_zungtr LAPACK_GLOBAL(zungtr,ZUNGTR)
+#define LAPACK_cunmtr LAPACK_GLOBAL(cunmtr,CUNMTR)
+#define LAPACK_zunmtr LAPACK_GLOBAL(zunmtr,ZUNMTR)
+#define LAPACK_ssptrd LAPACK_GLOBAL(ssptrd,SSPTRD)
+#define LAPACK_dsptrd LAPACK_GLOBAL(dsptrd,DSPTRD)
+#define LAPACK_sopgtr LAPACK_GLOBAL(sopgtr,SOPGTR)
+#define LAPACK_dopgtr LAPACK_GLOBAL(dopgtr,DOPGTR)
+#define LAPACK_sopmtr LAPACK_GLOBAL(sopmtr,SOPMTR)
+#define LAPACK_dopmtr LAPACK_GLOBAL(dopmtr,DOPMTR)
+#define LAPACK_chptrd LAPACK_GLOBAL(chptrd,CHPTRD)
+#define LAPACK_zhptrd LAPACK_GLOBAL(zhptrd,ZHPTRD)
+#define LAPACK_cupgtr LAPACK_GLOBAL(cupgtr,CUPGTR)
+#define LAPACK_zupgtr LAPACK_GLOBAL(zupgtr,ZUPGTR)
+#define LAPACK_cupmtr LAPACK_GLOBAL(cupmtr,CUPMTR)
+#define LAPACK_zupmtr LAPACK_GLOBAL(zupmtr,ZUPMTR)
+#define LAPACK_ssbtrd LAPACK_GLOBAL(ssbtrd,SSBTRD)
+#define LAPACK_dsbtrd LAPACK_GLOBAL(dsbtrd,DSBTRD)
+#define LAPACK_chbtrd LAPACK_GLOBAL(chbtrd,CHBTRD)
+#define LAPACK_zhbtrd LAPACK_GLOBAL(zhbtrd,ZHBTRD)
+#define LAPACK_ssterf LAPACK_GLOBAL(ssterf,SSTERF)
+#define LAPACK_dsterf LAPACK_GLOBAL(dsterf,DSTERF)
+#define LAPACK_ssteqr LAPACK_GLOBAL(ssteqr,SSTEQR)
+#define LAPACK_dsteqr LAPACK_GLOBAL(dsteqr,DSTEQR)
+#define LAPACK_csteqr LAPACK_GLOBAL(csteqr,CSTEQR)
+#define LAPACK_zsteqr LAPACK_GLOBAL(zsteqr,ZSTEQR)
+#define LAPACK_sstemr LAPACK_GLOBAL(sstemr,SSTEMR)
+#define LAPACK_dstemr LAPACK_GLOBAL(dstemr,DSTEMR)
+#define LAPACK_cstemr LAPACK_GLOBAL(cstemr,CSTEMR)
+#define LAPACK_zstemr LAPACK_GLOBAL(zstemr,ZSTEMR)
+#define LAPACK_sstedc LAPACK_GLOBAL(sstedc,SSTEDC)
+#define LAPACK_dstedc LAPACK_GLOBAL(dstedc,DSTEDC)
+#define LAPACK_cstedc LAPACK_GLOBAL(cstedc,CSTEDC)
+#define LAPACK_zstedc LAPACK_GLOBAL(zstedc,ZSTEDC)
+#define LAPACK_sstegr LAPACK_GLOBAL(sstegr,SSTEGR)
+#define LAPACK_dstegr LAPACK_GLOBAL(dstegr,DSTEGR)
+#define LAPACK_cstegr LAPACK_GLOBAL(cstegr,CSTEGR)
+#define LAPACK_zstegr LAPACK_GLOBAL(zstegr,ZSTEGR)
+#define LAPACK_spteqr LAPACK_GLOBAL(spteqr,SPTEQR)
+#define LAPACK_dpteqr LAPACK_GLOBAL(dpteqr,DPTEQR)
+#define LAPACK_cpteqr LAPACK_GLOBAL(cpteqr,CPTEQR)
+#define LAPACK_zpteqr LAPACK_GLOBAL(zpteqr,ZPTEQR)
+#define LAPACK_sstebz LAPACK_GLOBAL(sstebz,SSTEBZ)
+#define LAPACK_dstebz LAPACK_GLOBAL(dstebz,DSTEBZ)
+#define LAPACK_sstein LAPACK_GLOBAL(sstein,SSTEIN)
+#define LAPACK_dstein LAPACK_GLOBAL(dstein,DSTEIN)
+#define LAPACK_cstein LAPACK_GLOBAL(cstein,CSTEIN)
+#define LAPACK_zstein LAPACK_GLOBAL(zstein,ZSTEIN)
+#define LAPACK_sdisna LAPACK_GLOBAL(sdisna,SDISNA)
+#define LAPACK_ddisna LAPACK_GLOBAL(ddisna,DDISNA)
+#define LAPACK_ssygst LAPACK_GLOBAL(ssygst,SSYGST)
+#define LAPACK_dsygst LAPACK_GLOBAL(dsygst,DSYGST)
+#define LAPACK_chegst LAPACK_GLOBAL(chegst,CHEGST)
+#define LAPACK_zhegst LAPACK_GLOBAL(zhegst,ZHEGST)
+#define LAPACK_sspgst LAPACK_GLOBAL(sspgst,SSPGST)
+#define LAPACK_dspgst LAPACK_GLOBAL(dspgst,DSPGST)
+#define LAPACK_chpgst LAPACK_GLOBAL(chpgst,CHPGST)
+#define LAPACK_zhpgst LAPACK_GLOBAL(zhpgst,ZHPGST)
+#define LAPACK_ssbgst LAPACK_GLOBAL(ssbgst,SSBGST)
+#define LAPACK_dsbgst LAPACK_GLOBAL(dsbgst,DSBGST)
+#define LAPACK_chbgst LAPACK_GLOBAL(chbgst,CHBGST)
+#define LAPACK_zhbgst LAPACK_GLOBAL(zhbgst,ZHBGST)
+#define LAPACK_spbstf LAPACK_GLOBAL(spbstf,SPBSTF)
+#define LAPACK_dpbstf LAPACK_GLOBAL(dpbstf,DPBSTF)
+#define LAPACK_cpbstf LAPACK_GLOBAL(cpbstf,CPBSTF)
+#define LAPACK_zpbstf LAPACK_GLOBAL(zpbstf,ZPBSTF)
+#define LAPACK_sgehrd LAPACK_GLOBAL(sgehrd,SGEHRD)
+#define LAPACK_dgehrd LAPACK_GLOBAL(dgehrd,DGEHRD)
+#define LAPACK_cgehrd LAPACK_GLOBAL(cgehrd,CGEHRD)
+#define LAPACK_zgehrd LAPACK_GLOBAL(zgehrd,ZGEHRD)
+#define LAPACK_sorghr LAPACK_GLOBAL(sorghr,SORGHR)
+#define LAPACK_dorghr LAPACK_GLOBAL(dorghr,DORGHR)
+#define LAPACK_sormhr LAPACK_GLOBAL(sormhr,SORMHR)
+#define LAPACK_dormhr LAPACK_GLOBAL(dormhr,DORMHR)
+#define LAPACK_cunghr LAPACK_GLOBAL(cunghr,CUNGHR)
+#define LAPACK_zunghr LAPACK_GLOBAL(zunghr,ZUNGHR)
+#define LAPACK_cunmhr LAPACK_GLOBAL(cunmhr,CUNMHR)
+#define LAPACK_zunmhr LAPACK_GLOBAL(zunmhr,ZUNMHR)
+#define LAPACK_sgebal LAPACK_GLOBAL(sgebal,SGEBAL)
+#define LAPACK_dgebal LAPACK_GLOBAL(dgebal,DGEBAL)
+#define LAPACK_cgebal LAPACK_GLOBAL(cgebal,CGEBAL)
+#define LAPACK_zgebal LAPACK_GLOBAL(zgebal,ZGEBAL)
+#define LAPACK_sgebak LAPACK_GLOBAL(sgebak,SGEBAK)
+#define LAPACK_dgebak LAPACK_GLOBAL(dgebak,DGEBAK)
+#define LAPACK_cgebak LAPACK_GLOBAL(cgebak,CGEBAK)
+#define LAPACK_zgebak LAPACK_GLOBAL(zgebak,ZGEBAK)
+#define LAPACK_shseqr LAPACK_GLOBAL(shseqr,SHSEQR)
+#define LAPACK_dhseqr LAPACK_GLOBAL(dhseqr,DHSEQR)
+#define LAPACK_chseqr LAPACK_GLOBAL(chseqr,CHSEQR)
+#define LAPACK_zhseqr LAPACK_GLOBAL(zhseqr,ZHSEQR)
+#define LAPACK_shsein LAPACK_GLOBAL(shsein,SHSEIN)
+#define LAPACK_dhsein LAPACK_GLOBAL(dhsein,DHSEIN)
+#define LAPACK_chsein LAPACK_GLOBAL(chsein,CHSEIN)
+#define LAPACK_zhsein LAPACK_GLOBAL(zhsein,ZHSEIN)
+#define LAPACK_strevc LAPACK_GLOBAL(strevc,STREVC)
+#define LAPACK_dtrevc LAPACK_GLOBAL(dtrevc,DTREVC)
+#define LAPACK_ctrevc LAPACK_GLOBAL(ctrevc,CTREVC)
+#define LAPACK_ztrevc LAPACK_GLOBAL(ztrevc,ZTREVC)
+#define LAPACK_strsna LAPACK_GLOBAL(strsna,STRSNA)
+#define LAPACK_dtrsna LAPACK_GLOBAL(dtrsna,DTRSNA)
+#define LAPACK_ctrsna LAPACK_GLOBAL(ctrsna,CTRSNA)
+#define LAPACK_ztrsna LAPACK_GLOBAL(ztrsna,ZTRSNA)
+#define LAPACK_strexc LAPACK_GLOBAL(strexc,STREXC)
+#define LAPACK_dtrexc LAPACK_GLOBAL(dtrexc,DTREXC)
+#define LAPACK_ctrexc LAPACK_GLOBAL(ctrexc,CTREXC)
+#define LAPACK_ztrexc LAPACK_GLOBAL(ztrexc,ZTREXC)
+#define LAPACK_strsen LAPACK_GLOBAL(strsen,STRSEN)
+#define LAPACK_dtrsen LAPACK_GLOBAL(dtrsen,DTRSEN)
+#define LAPACK_ctrsen LAPACK_GLOBAL(ctrsen,CTRSEN)
+#define LAPACK_ztrsen LAPACK_GLOBAL(ztrsen,ZTRSEN)
+#define LAPACK_strsyl LAPACK_GLOBAL(strsyl,STRSYL)
+#define LAPACK_dtrsyl LAPACK_GLOBAL(dtrsyl,DTRSYL)
+#define LAPACK_ctrsyl LAPACK_GLOBAL(ctrsyl,CTRSYL)
+#define LAPACK_ztrsyl LAPACK_GLOBAL(ztrsyl,ZTRSYL)
+#define LAPACK_sgghrd LAPACK_GLOBAL(sgghrd,SGGHRD)
+#define LAPACK_dgghrd LAPACK_GLOBAL(dgghrd,DGGHRD)
+#define LAPACK_cgghrd LAPACK_GLOBAL(cgghrd,CGGHRD)
+#define LAPACK_zgghrd LAPACK_GLOBAL(zgghrd,ZGGHRD)
+#define LAPACK_sggbal LAPACK_GLOBAL(sggbal,SGGBAL)
+#define LAPACK_dggbal LAPACK_GLOBAL(dggbal,DGGBAL)
+#define LAPACK_cggbal LAPACK_GLOBAL(cggbal,CGGBAL)
+#define LAPACK_zggbal LAPACK_GLOBAL(zggbal,ZGGBAL)
+#define LAPACK_sggbak LAPACK_GLOBAL(sggbak,SGGBAK)
+#define LAPACK_dggbak LAPACK_GLOBAL(dggbak,DGGBAK)
+#define LAPACK_cggbak LAPACK_GLOBAL(cggbak,CGGBAK)
+#define LAPACK_zggbak LAPACK_GLOBAL(zggbak,ZGGBAK)
+#define LAPACK_shgeqz LAPACK_GLOBAL(shgeqz,SHGEQZ)
+#define LAPACK_dhgeqz LAPACK_GLOBAL(dhgeqz,DHGEQZ)
+#define LAPACK_chgeqz LAPACK_GLOBAL(chgeqz,CHGEQZ)
+#define LAPACK_zhgeqz LAPACK_GLOBAL(zhgeqz,ZHGEQZ)
+#define LAPACK_stgevc LAPACK_GLOBAL(stgevc,STGEVC)
+#define LAPACK_dtgevc LAPACK_GLOBAL(dtgevc,DTGEVC)
+#define LAPACK_ctgevc LAPACK_GLOBAL(ctgevc,CTGEVC)
+#define LAPACK_ztgevc LAPACK_GLOBAL(ztgevc,ZTGEVC)
+#define LAPACK_stgexc LAPACK_GLOBAL(stgexc,STGEXC)
+#define LAPACK_dtgexc LAPACK_GLOBAL(dtgexc,DTGEXC)
+#define LAPACK_ctgexc LAPACK_GLOBAL(ctgexc,CTGEXC)
+#define LAPACK_ztgexc LAPACK_GLOBAL(ztgexc,ZTGEXC)
+#define LAPACK_stgsen LAPACK_GLOBAL(stgsen,STGSEN)
+#define LAPACK_dtgsen LAPACK_GLOBAL(dtgsen,DTGSEN)
+#define LAPACK_ctgsen LAPACK_GLOBAL(ctgsen,CTGSEN)
+#define LAPACK_ztgsen LAPACK_GLOBAL(ztgsen,ZTGSEN)
+#define LAPACK_stgsyl LAPACK_GLOBAL(stgsyl,STGSYL)
+#define LAPACK_dtgsyl LAPACK_GLOBAL(dtgsyl,DTGSYL)
+#define LAPACK_ctgsyl LAPACK_GLOBAL(ctgsyl,CTGSYL)
+#define LAPACK_ztgsyl LAPACK_GLOBAL(ztgsyl,ZTGSYL)
+#define LAPACK_stgsna LAPACK_GLOBAL(stgsna,STGSNA)
+#define LAPACK_dtgsna LAPACK_GLOBAL(dtgsna,DTGSNA)
+#define LAPACK_ctgsna LAPACK_GLOBAL(ctgsna,CTGSNA)
+#define LAPACK_ztgsna LAPACK_GLOBAL(ztgsna,ZTGSNA)
+#define LAPACK_sggsvp LAPACK_GLOBAL(sggsvp,SGGSVP)
+#define LAPACK_dggsvp LAPACK_GLOBAL(dggsvp,DGGSVP)
+#define LAPACK_cggsvp LAPACK_GLOBAL(cggsvp,CGGSVP)
+#define LAPACK_zggsvp LAPACK_GLOBAL(zggsvp,ZGGSVP)
+#define LAPACK_stgsja LAPACK_GLOBAL(stgsja,STGSJA)
+#define LAPACK_dtgsja LAPACK_GLOBAL(dtgsja,DTGSJA)
+#define LAPACK_ctgsja LAPACK_GLOBAL(ctgsja,CTGSJA)
+#define LAPACK_ztgsja LAPACK_GLOBAL(ztgsja,ZTGSJA)
+#define LAPACK_sgels LAPACK_GLOBAL(sgels,SGELS)
+#define LAPACK_dgels LAPACK_GLOBAL(dgels,DGELS)
+#define LAPACK_cgels LAPACK_GLOBAL(cgels,CGELS)
+#define LAPACK_zgels LAPACK_GLOBAL(zgels,ZGELS)
+#define LAPACK_sgelsy LAPACK_GLOBAL(sgelsy,SGELSY)
+#define LAPACK_dgelsy LAPACK_GLOBAL(dgelsy,DGELSY)
+#define LAPACK_cgelsy LAPACK_GLOBAL(cgelsy,CGELSY)
+#define LAPACK_zgelsy LAPACK_GLOBAL(zgelsy,ZGELSY)
+#define LAPACK_sgelss LAPACK_GLOBAL(sgelss,SGELSS)
+#define LAPACK_dgelss LAPACK_GLOBAL(dgelss,DGELSS)
+#define LAPACK_cgelss LAPACK_GLOBAL(cgelss,CGELSS)
+#define LAPACK_zgelss LAPACK_GLOBAL(zgelss,ZGELSS)
+#define LAPACK_sgelsd LAPACK_GLOBAL(sgelsd,SGELSD)
+#define LAPACK_dgelsd LAPACK_GLOBAL(dgelsd,DGELSD)
+#define LAPACK_cgelsd LAPACK_GLOBAL(cgelsd,CGELSD)
+#define LAPACK_zgelsd LAPACK_GLOBAL(zgelsd,ZGELSD)
+#define LAPACK_sgglse LAPACK_GLOBAL(sgglse,SGGLSE)
+#define LAPACK_dgglse LAPACK_GLOBAL(dgglse,DGGLSE)
+#define LAPACK_cgglse LAPACK_GLOBAL(cgglse,CGGLSE)
+#define LAPACK_zgglse LAPACK_GLOBAL(zgglse,ZGGLSE)
+#define LAPACK_sggglm LAPACK_GLOBAL(sggglm,SGGGLM)
+#define LAPACK_dggglm LAPACK_GLOBAL(dggglm,DGGGLM)
+#define LAPACK_cggglm LAPACK_GLOBAL(cggglm,CGGGLM)
+#define LAPACK_zggglm LAPACK_GLOBAL(zggglm,ZGGGLM)
+#define LAPACK_ssyev LAPACK_GLOBAL(ssyev,SSYEV)
+#define LAPACK_dsyev LAPACK_GLOBAL(dsyev,DSYEV)
+#define LAPACK_cheev LAPACK_GLOBAL(cheev,CHEEV)
+#define LAPACK_zheev LAPACK_GLOBAL(zheev,ZHEEV)
+#define LAPACK_ssyevd LAPACK_GLOBAL(ssyevd,SSYEVD)
+#define LAPACK_dsyevd LAPACK_GLOBAL(dsyevd,DSYEVD)
+#define LAPACK_cheevd LAPACK_GLOBAL(cheevd,CHEEVD)
+#define LAPACK_zheevd LAPACK_GLOBAL(zheevd,ZHEEVD)
+#define LAPACK_ssyevx LAPACK_GLOBAL(ssyevx,SSYEVX)
+#define LAPACK_dsyevx LAPACK_GLOBAL(dsyevx,DSYEVX)
+#define LAPACK_cheevx LAPACK_GLOBAL(cheevx,CHEEVX)
+#define LAPACK_zheevx LAPACK_GLOBAL(zheevx,ZHEEVX)
+#define LAPACK_ssyevr LAPACK_GLOBAL(ssyevr,SSYEVR)
+#define LAPACK_dsyevr LAPACK_GLOBAL(dsyevr,DSYEVR)
+#define LAPACK_cheevr LAPACK_GLOBAL(cheevr,CHEEVR)
+#define LAPACK_zheevr LAPACK_GLOBAL(zheevr,ZHEEVR)
+#define LAPACK_sspev LAPACK_GLOBAL(sspev,SSPEV)
+#define LAPACK_dspev LAPACK_GLOBAL(dspev,DSPEV)
+#define LAPACK_chpev LAPACK_GLOBAL(chpev,CHPEV)
+#define LAPACK_zhpev LAPACK_GLOBAL(zhpev,ZHPEV)
+#define LAPACK_sspevd LAPACK_GLOBAL(sspevd,SSPEVD)
+#define LAPACK_dspevd LAPACK_GLOBAL(dspevd,DSPEVD)
+#define LAPACK_chpevd LAPACK_GLOBAL(chpevd,CHPEVD)
+#define LAPACK_zhpevd LAPACK_GLOBAL(zhpevd,ZHPEVD)
+#define LAPACK_sspevx LAPACK_GLOBAL(sspevx,SSPEVX)
+#define LAPACK_dspevx LAPACK_GLOBAL(dspevx,DSPEVX)
+#define LAPACK_chpevx LAPACK_GLOBAL(chpevx,CHPEVX)
+#define LAPACK_zhpevx LAPACK_GLOBAL(zhpevx,ZHPEVX)
+#define LAPACK_ssbev LAPACK_GLOBAL(ssbev,SSBEV)
+#define LAPACK_dsbev LAPACK_GLOBAL(dsbev,DSBEV)
+#define LAPACK_chbev LAPACK_GLOBAL(chbev,CHBEV)
+#define LAPACK_zhbev LAPACK_GLOBAL(zhbev,ZHBEV)
+#define LAPACK_ssbevd LAPACK_GLOBAL(ssbevd,SSBEVD)
+#define LAPACK_dsbevd LAPACK_GLOBAL(dsbevd,DSBEVD)
+#define LAPACK_chbevd LAPACK_GLOBAL(chbevd,CHBEVD)
+#define LAPACK_zhbevd LAPACK_GLOBAL(zhbevd,ZHBEVD)
+#define LAPACK_ssbevx LAPACK_GLOBAL(ssbevx,SSBEVX)
+#define LAPACK_dsbevx LAPACK_GLOBAL(dsbevx,DSBEVX)
+#define LAPACK_chbevx LAPACK_GLOBAL(chbevx,CHBEVX)
+#define LAPACK_zhbevx LAPACK_GLOBAL(zhbevx,ZHBEVX)
+#define LAPACK_sstev LAPACK_GLOBAL(sstev,SSTEV)
+#define LAPACK_dstev LAPACK_GLOBAL(dstev,DSTEV)
+#define LAPACK_sstevd LAPACK_GLOBAL(sstevd,SSTEVD)
+#define LAPACK_dstevd LAPACK_GLOBAL(dstevd,DSTEVD)
+#define LAPACK_sstevx LAPACK_GLOBAL(sstevx,SSTEVX)
+#define LAPACK_dstevx LAPACK_GLOBAL(dstevx,DSTEVX)
+#define LAPACK_sstevr LAPACK_GLOBAL(sstevr,SSTEVR)
+#define LAPACK_dstevr LAPACK_GLOBAL(dstevr,DSTEVR)
+#define LAPACK_sgees LAPACK_GLOBAL(sgees,SGEES)
+#define LAPACK_dgees LAPACK_GLOBAL(dgees,DGEES)
+#define LAPACK_cgees LAPACK_GLOBAL(cgees,CGEES)
+#define LAPACK_zgees LAPACK_GLOBAL(zgees,ZGEES)
+#define LAPACK_sgeesx LAPACK_GLOBAL(sgeesx,SGEESX)
+#define LAPACK_dgeesx LAPACK_GLOBAL(dgeesx,DGEESX)
+#define LAPACK_cgeesx LAPACK_GLOBAL(cgeesx,CGEESX)
+#define LAPACK_zgeesx LAPACK_GLOBAL(zgeesx,ZGEESX)
+#define LAPACK_sgeev LAPACK_GLOBAL(sgeev,SGEEV)
+#define LAPACK_dgeev LAPACK_GLOBAL(dgeev,DGEEV)
+#define LAPACK_cgeev LAPACK_GLOBAL(cgeev,CGEEV)
+#define LAPACK_zgeev LAPACK_GLOBAL(zgeev,ZGEEV)
+#define LAPACK_sgeevx LAPACK_GLOBAL(sgeevx,SGEEVX)
+#define LAPACK_dgeevx LAPACK_GLOBAL(dgeevx,DGEEVX)
+#define LAPACK_cgeevx LAPACK_GLOBAL(cgeevx,CGEEVX)
+#define LAPACK_zgeevx LAPACK_GLOBAL(zgeevx,ZGEEVX)
+#define LAPACK_sgesvd LAPACK_GLOBAL(sgesvd,SGESVD)
+#define LAPACK_dgesvd LAPACK_GLOBAL(dgesvd,DGESVD)
+#define LAPACK_cgesvd LAPACK_GLOBAL(cgesvd,CGESVD)
+#define LAPACK_zgesvd LAPACK_GLOBAL(zgesvd,ZGESVD)
+#define LAPACK_sgesdd LAPACK_GLOBAL(sgesdd,SGESDD)
+#define LAPACK_dgesdd LAPACK_GLOBAL(dgesdd,DGESDD)
+#define LAPACK_cgesdd LAPACK_GLOBAL(cgesdd,CGESDD)
+#define LAPACK_zgesdd LAPACK_GLOBAL(zgesdd,ZGESDD)
+#define LAPACK_dgejsv LAPACK_GLOBAL(dgejsv,DGEJSV)
+#define LAPACK_sgejsv LAPACK_GLOBAL(sgejsv,SGEJSV)
+#define LAPACK_dgesvj LAPACK_GLOBAL(dgesvj,DGESVJ)
+#define LAPACK_sgesvj LAPACK_GLOBAL(sgesvj,SGESVJ)
+#define LAPACK_sggsvd LAPACK_GLOBAL(sggsvd,SGGSVD)
+#define LAPACK_dggsvd LAPACK_GLOBAL(dggsvd,DGGSVD)
+#define LAPACK_cggsvd LAPACK_GLOBAL(cggsvd,CGGSVD)
+#define LAPACK_zggsvd LAPACK_GLOBAL(zggsvd,ZGGSVD)
+#define LAPACK_ssygv LAPACK_GLOBAL(ssygv,SSYGV)
+#define LAPACK_dsygv LAPACK_GLOBAL(dsygv,DSYGV)
+#define LAPACK_chegv LAPACK_GLOBAL(chegv,CHEGV)
+#define LAPACK_zhegv LAPACK_GLOBAL(zhegv,ZHEGV)
+#define LAPACK_ssygvd LAPACK_GLOBAL(ssygvd,SSYGVD)
+#define LAPACK_dsygvd LAPACK_GLOBAL(dsygvd,DSYGVD)
+#define LAPACK_chegvd LAPACK_GLOBAL(chegvd,CHEGVD)
+#define LAPACK_zhegvd LAPACK_GLOBAL(zhegvd,ZHEGVD)
+#define LAPACK_ssygvx LAPACK_GLOBAL(ssygvx,SSYGVX)
+#define LAPACK_dsygvx LAPACK_GLOBAL(dsygvx,DSYGVX)
+#define LAPACK_chegvx LAPACK_GLOBAL(chegvx,CHEGVX)
+#define LAPACK_zhegvx LAPACK_GLOBAL(zhegvx,ZHEGVX)
+#define LAPACK_sspgv LAPACK_GLOBAL(sspgv,SSPGV)
+#define LAPACK_dspgv LAPACK_GLOBAL(dspgv,DSPGV)
+#define LAPACK_chpgv LAPACK_GLOBAL(chpgv,CHPGV)
+#define LAPACK_zhpgv LAPACK_GLOBAL(zhpgv,ZHPGV)
+#define LAPACK_sspgvd LAPACK_GLOBAL(sspgvd,SSPGVD)
+#define LAPACK_dspgvd LAPACK_GLOBAL(dspgvd,DSPGVD)
+#define LAPACK_chpgvd LAPACK_GLOBAL(chpgvd,CHPGVD)
+#define LAPACK_zhpgvd LAPACK_GLOBAL(zhpgvd,ZHPGVD)
+#define LAPACK_sspgvx LAPACK_GLOBAL(sspgvx,SSPGVX)
+#define LAPACK_dspgvx LAPACK_GLOBAL(dspgvx,DSPGVX)
+#define LAPACK_chpgvx LAPACK_GLOBAL(chpgvx,CHPGVX)
+#define LAPACK_zhpgvx LAPACK_GLOBAL(zhpgvx,ZHPGVX)
+#define LAPACK_ssbgv LAPACK_GLOBAL(ssbgv,SSBGV)
+#define LAPACK_dsbgv LAPACK_GLOBAL(dsbgv,DSBGV)
+#define LAPACK_chbgv LAPACK_GLOBAL(chbgv,CHBGV)
+#define LAPACK_zhbgv LAPACK_GLOBAL(zhbgv,ZHBGV)
+#define LAPACK_ssbgvd LAPACK_GLOBAL(ssbgvd,SSBGVD)
+#define LAPACK_dsbgvd LAPACK_GLOBAL(dsbgvd,DSBGVD)
+#define LAPACK_chbgvd LAPACK_GLOBAL(chbgvd,CHBGVD)
+#define LAPACK_zhbgvd LAPACK_GLOBAL(zhbgvd,ZHBGVD)
+#define LAPACK_ssbgvx LAPACK_GLOBAL(ssbgvx,SSBGVX)
+#define LAPACK_dsbgvx LAPACK_GLOBAL(dsbgvx,DSBGVX)
+#define LAPACK_chbgvx LAPACK_GLOBAL(chbgvx,CHBGVX)
+#define LAPACK_zhbgvx LAPACK_GLOBAL(zhbgvx,ZHBGVX)
+#define LAPACK_sgges LAPACK_GLOBAL(sgges,SGGES)
+#define LAPACK_dgges LAPACK_GLOBAL(dgges,DGGES)
+#define LAPACK_cgges LAPACK_GLOBAL(cgges,CGGES)
+#define LAPACK_zgges LAPACK_GLOBAL(zgges,ZGGES)
+#define LAPACK_sggesx LAPACK_GLOBAL(sggesx,SGGESX)
+#define LAPACK_dggesx LAPACK_GLOBAL(dggesx,DGGESX)
+#define LAPACK_cggesx LAPACK_GLOBAL(cggesx,CGGESX)
+#define LAPACK_zggesx LAPACK_GLOBAL(zggesx,ZGGESX)
+#define LAPACK_sggev LAPACK_GLOBAL(sggev,SGGEV)
+#define LAPACK_dggev LAPACK_GLOBAL(dggev,DGGEV)
+#define LAPACK_cggev LAPACK_GLOBAL(cggev,CGGEV)
+#define LAPACK_zggev LAPACK_GLOBAL(zggev,ZGGEV)
+#define LAPACK_sggevx LAPACK_GLOBAL(sggevx,SGGEVX)
+#define LAPACK_dggevx LAPACK_GLOBAL(dggevx,DGGEVX)
+#define LAPACK_cggevx LAPACK_GLOBAL(cggevx,CGGEVX)
+#define LAPACK_zggevx LAPACK_GLOBAL(zggevx,ZGGEVX)
+#define LAPACK_dsfrk LAPACK_GLOBAL(dsfrk,DSFRK)
+#define LAPACK_ssfrk LAPACK_GLOBAL(ssfrk,SSFRK)
+#define LAPACK_zhfrk LAPACK_GLOBAL(zhfrk,ZHFRK)
+#define LAPACK_chfrk LAPACK_GLOBAL(chfrk,CHFRK)
+#define LAPACK_dtfsm LAPACK_GLOBAL(dtfsm,DTFSM)
+#define LAPACK_stfsm LAPACK_GLOBAL(stfsm,STFSM)
+#define LAPACK_ztfsm LAPACK_GLOBAL(ztfsm,ZTFSM)
+#define LAPACK_ctfsm LAPACK_GLOBAL(ctfsm,CTFSM)
+#define LAPACK_dtfttp LAPACK_GLOBAL(dtfttp,DTFTTP)
+#define LAPACK_stfttp LAPACK_GLOBAL(stfttp,STFTTP)
+#define LAPACK_ztfttp LAPACK_GLOBAL(ztfttp,ZTFTTP)
+#define LAPACK_ctfttp LAPACK_GLOBAL(ctfttp,CTFTTP)
+#define LAPACK_dtfttr LAPACK_GLOBAL(dtfttr,DTFTTR)
+#define LAPACK_stfttr LAPACK_GLOBAL(stfttr,STFTTR)
+#define LAPACK_ztfttr LAPACK_GLOBAL(ztfttr,ZTFTTR)
+#define LAPACK_ctfttr LAPACK_GLOBAL(ctfttr,CTFTTR)
+#define LAPACK_dtpttf LAPACK_GLOBAL(dtpttf,DTPTTF)
+#define LAPACK_stpttf LAPACK_GLOBAL(stpttf,STPTTF)
+#define LAPACK_ztpttf LAPACK_GLOBAL(ztpttf,ZTPTTF)
+#define LAPACK_ctpttf LAPACK_GLOBAL(ctpttf,CTPTTF)
+#define LAPACK_dtpttr LAPACK_GLOBAL(dtpttr,DTPTTR)
+#define LAPACK_stpttr LAPACK_GLOBAL(stpttr,STPTTR)
+#define LAPACK_ztpttr LAPACK_GLOBAL(ztpttr,ZTPTTR)
+#define LAPACK_ctpttr LAPACK_GLOBAL(ctpttr,CTPTTR)
+#define LAPACK_dtrttf LAPACK_GLOBAL(dtrttf,DTRTTF)
+#define LAPACK_strttf LAPACK_GLOBAL(strttf,STRTTF)
+#define LAPACK_ztrttf LAPACK_GLOBAL(ztrttf,ZTRTTF)
+#define LAPACK_ctrttf LAPACK_GLOBAL(ctrttf,CTRTTF)
+#define LAPACK_dtrttp LAPACK_GLOBAL(dtrttp,DTRTTP)
+#define LAPACK_strttp LAPACK_GLOBAL(strttp,STRTTP)
+#define LAPACK_ztrttp LAPACK_GLOBAL(ztrttp,ZTRTTP)
+#define LAPACK_ctrttp LAPACK_GLOBAL(ctrttp,CTRTTP)
+#define LAPACK_sgeqrfp LAPACK_GLOBAL(sgeqrfp,SGEQRFP)
+#define LAPACK_dgeqrfp LAPACK_GLOBAL(dgeqrfp,DGEQRFP)
+#define LAPACK_cgeqrfp LAPACK_GLOBAL(cgeqrfp,CGEQRFP)
+#define LAPACK_zgeqrfp LAPACK_GLOBAL(zgeqrfp,ZGEQRFP)
+#define LAPACK_clacgv LAPACK_GLOBAL(clacgv,CLACGV)
+#define LAPACK_zlacgv LAPACK_GLOBAL(zlacgv,ZLACGV)
+#define LAPACK_slarnv LAPACK_GLOBAL(slarnv,SLARNV)
+#define LAPACK_dlarnv LAPACK_GLOBAL(dlarnv,DLARNV)
+#define LAPACK_clarnv LAPACK_GLOBAL(clarnv,CLARNV)
+#define LAPACK_zlarnv LAPACK_GLOBAL(zlarnv,ZLARNV)
+#define LAPACK_sgeqr2 LAPACK_GLOBAL(sgeqr2,SGEQR2)
+#define LAPACK_dgeqr2 LAPACK_GLOBAL(dgeqr2,DGEQR2)
+#define LAPACK_cgeqr2 LAPACK_GLOBAL(cgeqr2,CGEQR2)
+#define LAPACK_zgeqr2 LAPACK_GLOBAL(zgeqr2,ZGEQR2)
+#define LAPACK_slacpy LAPACK_GLOBAL(slacpy,SLACPY)
+#define LAPACK_dlacpy LAPACK_GLOBAL(dlacpy,DLACPY)
+#define LAPACK_clacpy LAPACK_GLOBAL(clacpy,CLACPY)
+#define LAPACK_zlacpy LAPACK_GLOBAL(zlacpy,ZLACPY)
+#define LAPACK_sgetf2 LAPACK_GLOBAL(sgetf2,SGETF2)
+#define LAPACK_dgetf2 LAPACK_GLOBAL(dgetf2,DGETF2)
+#define LAPACK_cgetf2 LAPACK_GLOBAL(cgetf2,CGETF2)
+#define LAPACK_zgetf2 LAPACK_GLOBAL(zgetf2,ZGETF2)
+#define LAPACK_slaswp LAPACK_GLOBAL(slaswp,SLASWP)
+#define LAPACK_dlaswp LAPACK_GLOBAL(dlaswp,DLASWP)
+#define LAPACK_claswp LAPACK_GLOBAL(claswp,CLASWP)
+#define LAPACK_zlaswp LAPACK_GLOBAL(zlaswp,ZLASWP)
+#define LAPACK_slange LAPACK_GLOBAL(slange,SLANGE)
+#define LAPACK_dlange LAPACK_GLOBAL(dlange,DLANGE)
+#define LAPACK_clange LAPACK_GLOBAL(clange,CLANGE)
+#define LAPACK_zlange LAPACK_GLOBAL(zlange,ZLANGE)
+#define LAPACK_clanhe LAPACK_GLOBAL(clanhe,CLANHE)
+#define LAPACK_zlanhe LAPACK_GLOBAL(zlanhe,ZLANHE)
+#define LAPACK_slansy LAPACK_GLOBAL(slansy,SLANSY)
+#define LAPACK_dlansy LAPACK_GLOBAL(dlansy,DLANSY)
+#define LAPACK_clansy LAPACK_GLOBAL(clansy,CLANSY)
+#define LAPACK_zlansy LAPACK_GLOBAL(zlansy,ZLANSY)
+#define LAPACK_slantr LAPACK_GLOBAL(slantr,SLANTR)
+#define LAPACK_dlantr LAPACK_GLOBAL(dlantr,DLANTR)
+#define LAPACK_clantr LAPACK_GLOBAL(clantr,CLANTR)
+#define LAPACK_zlantr LAPACK_GLOBAL(zlantr,ZLANTR)
+#define LAPACK_slamch LAPACK_GLOBAL(slamch,SLAMCH)
+#define LAPACK_dlamch LAPACK_GLOBAL(dlamch,DLAMCH)
+#define LAPACK_sgelq2 LAPACK_GLOBAL(sgelq2,SGELQ2)
+#define LAPACK_dgelq2 LAPACK_GLOBAL(dgelq2,DGELQ2)
+#define LAPACK_cgelq2 LAPACK_GLOBAL(cgelq2,CGELQ2)
+#define LAPACK_zgelq2 LAPACK_GLOBAL(zgelq2,ZGELQ2)
+#define LAPACK_slarfb LAPACK_GLOBAL(slarfb,SLARFB)
+#define LAPACK_dlarfb LAPACK_GLOBAL(dlarfb,DLARFB)
+#define LAPACK_clarfb LAPACK_GLOBAL(clarfb,CLARFB)
+#define LAPACK_zlarfb LAPACK_GLOBAL(zlarfb,ZLARFB)
+#define LAPACK_slarfg LAPACK_GLOBAL(slarfg,SLARFG)
+#define LAPACK_dlarfg LAPACK_GLOBAL(dlarfg,DLARFG)
+#define LAPACK_clarfg LAPACK_GLOBAL(clarfg,CLARFG)
+#define LAPACK_zlarfg LAPACK_GLOBAL(zlarfg,ZLARFG)
+#define LAPACK_slarft LAPACK_GLOBAL(slarft,SLARFT)
+#define LAPACK_dlarft LAPACK_GLOBAL(dlarft,DLARFT)
+#define LAPACK_clarft LAPACK_GLOBAL(clarft,CLARFT)
+#define LAPACK_zlarft LAPACK_GLOBAL(zlarft,ZLARFT)
+#define LAPACK_slarfx LAPACK_GLOBAL(slarfx,SLARFX)
+#define LAPACK_dlarfx LAPACK_GLOBAL(dlarfx,DLARFX)
+#define LAPACK_clarfx LAPACK_GLOBAL(clarfx,CLARFX)
+#define LAPACK_zlarfx LAPACK_GLOBAL(zlarfx,ZLARFX)
+#define LAPACK_slatms LAPACK_GLOBAL(slatms,SLATMS)
+#define LAPACK_dlatms LAPACK_GLOBAL(dlatms,DLATMS)
+#define LAPACK_clatms LAPACK_GLOBAL(clatms,CLATMS)
+#define LAPACK_zlatms LAPACK_GLOBAL(zlatms,ZLATMS)
+#define LAPACK_slag2d LAPACK_GLOBAL(slag2d,SLAG2D)
+#define LAPACK_dlag2s LAPACK_GLOBAL(dlag2s,DLAG2S)
+#define LAPACK_clag2z LAPACK_GLOBAL(clag2z,CLAG2Z)
+#define LAPACK_zlag2c LAPACK_GLOBAL(zlag2c,ZLAG2C)
+#define LAPACK_slauum LAPACK_GLOBAL(slauum,SLAUUM)
+#define LAPACK_dlauum LAPACK_GLOBAL(dlauum,DLAUUM)
+#define LAPACK_clauum LAPACK_GLOBAL(clauum,CLAUUM)
+#define LAPACK_zlauum LAPACK_GLOBAL(zlauum,ZLAUUM)
+#define LAPACK_slagge LAPACK_GLOBAL(slagge,SLAGGE)
+#define LAPACK_dlagge LAPACK_GLOBAL(dlagge,DLAGGE)
+#define LAPACK_clagge LAPACK_GLOBAL(clagge,CLAGGE)
+#define LAPACK_zlagge LAPACK_GLOBAL(zlagge,ZLAGGE)
+#define LAPACK_slaset LAPACK_GLOBAL(slaset,SLASET)
+#define LAPACK_dlaset LAPACK_GLOBAL(dlaset,DLASET)
+#define LAPACK_claset LAPACK_GLOBAL(claset,CLASET)
+#define LAPACK_zlaset LAPACK_GLOBAL(zlaset,ZLASET)
+#define LAPACK_slasrt LAPACK_GLOBAL(slasrt,SLASRT)
+#define LAPACK_dlasrt LAPACK_GLOBAL(dlasrt,DLASRT)
+#define LAPACK_slagsy LAPACK_GLOBAL(slagsy,SLAGSY)
+#define LAPACK_dlagsy LAPACK_GLOBAL(dlagsy,DLAGSY)
+#define LAPACK_clagsy LAPACK_GLOBAL(clagsy,CLAGSY)
+#define LAPACK_zlagsy LAPACK_GLOBAL(zlagsy,ZLAGSY)
+#define LAPACK_claghe LAPACK_GLOBAL(claghe,CLAGHE)
+#define LAPACK_zlaghe LAPACK_GLOBAL(zlaghe,ZLAGHE)
+#define LAPACK_slapmr LAPACK_GLOBAL(slapmr,SLAPMR)
+#define LAPACK_dlapmr LAPACK_GLOBAL(dlapmr,DLAPMR)
+#define LAPACK_clapmr LAPACK_GLOBAL(clapmr,CLAPMR)
+#define LAPACK_zlapmr LAPACK_GLOBAL(zlapmr,ZLAPMR)
+#define LAPACK_slapy2 LAPACK_GLOBAL(slapy2,SLAPY2)
+#define LAPACK_dlapy2 LAPACK_GLOBAL(dlapy2,DLAPY2)
+#define LAPACK_slapy3 LAPACK_GLOBAL(slapy3,SLAPY3)
+#define LAPACK_dlapy3 LAPACK_GLOBAL(dlapy3,DLAPY3)
+#define LAPACK_slartgp LAPACK_GLOBAL(slartgp,SLARTGP)
+#define LAPACK_dlartgp LAPACK_GLOBAL(dlartgp,DLARTGP)
+#define LAPACK_slartgs LAPACK_GLOBAL(slartgs,SLARTGS)
+#define LAPACK_dlartgs LAPACK_GLOBAL(dlartgs,DLARTGS)
+// LAPACK 3.3.0
+#define LAPACK_cbbcsd LAPACK_GLOBAL(cbbcsd,CBBCSD)
+#define LAPACK_cheswapr LAPACK_GLOBAL(cheswapr,CHESWAPR)
+#define LAPACK_chetri2 LAPACK_GLOBAL(chetri2,CHETRI2)
+#define LAPACK_chetri2x LAPACK_GLOBAL(chetri2x,CHETRI2X)
+#define LAPACK_chetrs2 LAPACK_GLOBAL(chetrs2,CHETRS2)
+#define LAPACK_csyconv LAPACK_GLOBAL(csyconv,CSYCONV)
+#define LAPACK_csyswapr LAPACK_GLOBAL(csyswapr,CSYSWAPR)
+#define LAPACK_csytri2 LAPACK_GLOBAL(csytri2,CSYTRI2)
+#define LAPACK_csytri2x LAPACK_GLOBAL(csytri2x,CSYTRI2X)
+#define LAPACK_csytrs2 LAPACK_GLOBAL(csytrs2,CSYTRS2)
+#define LAPACK_cunbdb LAPACK_GLOBAL(cunbdb,CUNBDB)
+#define LAPACK_cuncsd LAPACK_GLOBAL(cuncsd,CUNCSD)
+#define LAPACK_dbbcsd LAPACK_GLOBAL(dbbcsd,DBBCSD)
+#define LAPACK_dorbdb LAPACK_GLOBAL(dorbdb,DORBDB)
+#define LAPACK_dorcsd LAPACK_GLOBAL(dorcsd,DORCSD)
+#define LAPACK_dsyconv LAPACK_GLOBAL(dsyconv,DSYCONV)
+#define LAPACK_dsyswapr LAPACK_GLOBAL(dsyswapr,DSYSWAPR)
+#define LAPACK_dsytri2 LAPACK_GLOBAL(dsytri2,DSYTRI2)
+#define LAPACK_dsytri2x LAPACK_GLOBAL(dsytri2x,DSYTRI2X)
+#define LAPACK_dsytrs2 LAPACK_GLOBAL(dsytrs2,DSYTRS2)
+#define LAPACK_sbbcsd LAPACK_GLOBAL(sbbcsd,SBBCSD)
+#define LAPACK_sorbdb LAPACK_GLOBAL(sorbdb,SORBDB)
+#define LAPACK_sorcsd LAPACK_GLOBAL(sorcsd,SORCSD)
+#define LAPACK_ssyconv LAPACK_GLOBAL(ssyconv,SSYCONV)
+#define LAPACK_ssyswapr LAPACK_GLOBAL(ssyswapr,SSYSWAPR)
+#define LAPACK_ssytri2 LAPACK_GLOBAL(ssytri2,SSYTRI2)
+#define LAPACK_ssytri2x LAPACK_GLOBAL(ssytri2x,SSYTRI2X)
+#define LAPACK_ssytrs2 LAPACK_GLOBAL(ssytrs2,SSYTRS2)
+#define LAPACK_zbbcsd LAPACK_GLOBAL(zbbcsd,ZBBCSD)
+#define LAPACK_zheswapr LAPACK_GLOBAL(zheswapr,ZHESWAPR)
+#define LAPACK_zhetri2 LAPACK_GLOBAL(zhetri2,ZHETRI2)
+#define LAPACK_zhetri2x LAPACK_GLOBAL(zhetri2x,ZHETRI2X)
+#define LAPACK_zhetrs2 LAPACK_GLOBAL(zhetrs2,ZHETRS2)
+#define LAPACK_zsyconv LAPACK_GLOBAL(zsyconv,ZSYCONV)
+#define LAPACK_zsyswapr LAPACK_GLOBAL(zsyswapr,ZSYSWAPR)
+#define LAPACK_zsytri2 LAPACK_GLOBAL(zsytri2,ZSYTRI2)
+#define LAPACK_zsytri2x LAPACK_GLOBAL(zsytri2x,ZSYTRI2X)
+#define LAPACK_zsytrs2 LAPACK_GLOBAL(zsytrs2,ZSYTRS2)
+#define LAPACK_zunbdb LAPACK_GLOBAL(zunbdb,ZUNBDB)
+#define LAPACK_zuncsd LAPACK_GLOBAL(zuncsd,ZUNCSD)
+// LAPACK 3.4.0
+#define LAPACK_sgemqrt LAPACK_GLOBAL(sgemqrt,SGEMQRT)
+#define LAPACK_dgemqrt LAPACK_GLOBAL(dgemqrt,DGEMQRT)
+#define LAPACK_cgemqrt LAPACK_GLOBAL(cgemqrt,CGEMQRT)
+#define LAPACK_zgemqrt LAPACK_GLOBAL(zgemqrt,ZGEMQRT)
+#define LAPACK_sgeqrt LAPACK_GLOBAL(sgeqrt,SGEQRT)
+#define LAPACK_dgeqrt LAPACK_GLOBAL(dgeqrt,DGEQRT)
+#define LAPACK_cgeqrt LAPACK_GLOBAL(cgeqrt,CGEQRT)
+#define LAPACK_zgeqrt LAPACK_GLOBAL(zgeqrt,ZGEQRT)
+#define LAPACK_sgeqrt2 LAPACK_GLOBAL(sgeqrt2,SGEQRT2)
+#define LAPACK_dgeqrt2 LAPACK_GLOBAL(dgeqrt2,DGEQRT2)
+#define LAPACK_cgeqrt2 LAPACK_GLOBAL(cgeqrt2,CGEQRT2)
+#define LAPACK_zgeqrt2 LAPACK_GLOBAL(zgeqrt2,ZGEQRT2)
+#define LAPACK_sgeqrt3 LAPACK_GLOBAL(sgeqrt3,SGEQRT3)
+#define LAPACK_dgeqrt3 LAPACK_GLOBAL(dgeqrt3,DGEQRT3)
+#define LAPACK_cgeqrt3 LAPACK_GLOBAL(cgeqrt3,CGEQRT3)
+#define LAPACK_zgeqrt3 LAPACK_GLOBAL(zgeqrt3,ZGEQRT3)
+#define LAPACK_stpmqrt LAPACK_GLOBAL(stpmqrt,STPMQRT)
+#define LAPACK_dtpmqrt LAPACK_GLOBAL(dtpmqrt,DTPMQRT)
+#define LAPACK_ctpmqrt LAPACK_GLOBAL(ctpmqrt,CTPMQRT)
+#define LAPACK_ztpmqrt LAPACK_GLOBAL(ztpmqrt,ZTPMQRT)
+#define LAPACK_dtpqrt LAPACK_GLOBAL(dtpqrt,DTPQRT)
+#define LAPACK_ctpqrt LAPACK_GLOBAL(ctpqrt,CTPQRT)
+#define LAPACK_ztpqrt LAPACK_GLOBAL(ztpqrt,ZTPQRT)
+#define LAPACK_stpqrt2 LAPACK_GLOBAL(stpqrt2,STPQRT2)
+#define LAPACK_dtpqrt2 LAPACK_GLOBAL(dtpqrt2,DTPQRT2)
+#define LAPACK_ctpqrt2 LAPACK_GLOBAL(ctpqrt2,CTPQRT2)
+#define LAPACK_ztpqrt2 LAPACK_GLOBAL(ztpqrt2,ZTPQRT2)
+#define LAPACK_stprfb LAPACK_GLOBAL(stprfb,STPRFB)
+#define LAPACK_dtprfb LAPACK_GLOBAL(dtprfb,DTPRFB)
+#define LAPACK_ctprfb LAPACK_GLOBAL(ctprfb,CTPRFB)
+#define LAPACK_ztprfb LAPACK_GLOBAL(ztprfb,ZTPRFB)
+// LAPACK 3.X.X
+#define LAPACK_csyr LAPACK_GLOBAL(csyr,CSYR)
+#define LAPACK_zsyr LAPACK_GLOBAL(zsyr,ZSYR)
+
+
+void LAPACK_sgetrf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgetrf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgetrf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_zgetrf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_sgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, float* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, double* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_zgbtrf( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_sgttrf( lapack_int* n, float* dl, float* d, float* du, float* du2,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgttrf( lapack_int* n, double* dl, double* d, double* du,
+                    double* du2, lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgttrf( lapack_int* n, lapack_complex_float* dl,
+                    lapack_complex_float* d, lapack_complex_float* du,
+                    lapack_complex_float* du2, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_zgttrf( lapack_int* n, lapack_complex_double* dl,
+                    lapack_complex_double* d, lapack_complex_double* du,
+                    lapack_complex_double* du2, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_spotrf( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dpotrf( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_cpotrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_zpotrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dpstrf( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* piv, lapack_int* rank, double* tol,
+                    double* work, lapack_int *info );
+void LAPACK_spstrf( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* piv, lapack_int* rank, float* tol, float* work,
+                    lapack_int *info );
+void LAPACK_zpstrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* piv, lapack_int* rank,
+                    double* tol, double* work, lapack_int *info );
+void LAPACK_cpstrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* piv, lapack_int* rank,
+                    float* tol, float* work, lapack_int *info );
+void LAPACK_dpftrf( char* transr, char* uplo, lapack_int* n, double* a,
+                    lapack_int *info );
+void LAPACK_spftrf( char* transr, char* uplo, lapack_int* n, float* a,
+                    lapack_int *info );
+void LAPACK_zpftrf( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int *info );
+void LAPACK_cpftrf( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int *info );
+void LAPACK_spptrf( char* uplo, lapack_int* n, float* ap, lapack_int *info );
+void LAPACK_dpptrf( char* uplo, lapack_int* n, double* ap, lapack_int *info );
+void LAPACK_cpptrf( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_zpptrf( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_spbtrf( char* uplo, lapack_int* n, lapack_int* kd, float* ab,
+                    lapack_int* ldab, lapack_int *info );
+void LAPACK_dpbtrf( char* uplo, lapack_int* n, lapack_int* kd, double* ab,
+                    lapack_int* ldab, lapack_int *info );
+void LAPACK_cpbtrf( char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_int *info );
+void LAPACK_zpbtrf( char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_int *info );
+void LAPACK_spttrf( lapack_int* n, float* d, float* e, lapack_int *info );
+void LAPACK_dpttrf( lapack_int* n, double* d, double* e, lapack_int *info );
+void LAPACK_cpttrf( lapack_int* n, float* d, lapack_complex_float* e,
+                    lapack_int *info );
+void LAPACK_zpttrf( lapack_int* n, double* d, lapack_complex_double* e,
+                    lapack_int *info );
+void LAPACK_ssytrf( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ipiv, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dsytrf( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ipiv, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_csytrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zsytrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_chetrf( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zhetrf( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ssptrf( char* uplo, lapack_int* n, float* ap, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_dsptrf( char* uplo, lapack_int* n, double* ap, lapack_int* ipiv,
+                    lapack_int *info );
+void LAPACK_csptrf( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_zsptrf( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_chptrf( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_zhptrf( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_sgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* a, lapack_int* lda, const lapack_int* ipiv,
+                    float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const lapack_int* ipiv,
+                    double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zgetrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_sgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const float* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const double* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_cgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_float* ab,
+                    lapack_int* ldab, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zgbtrs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_double* ab,
+                    lapack_int* ldab, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_sgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* dl, const float* d, const float* du,
+                    const float* du2, const lapack_int* ipiv, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* dl, const double* d, const double* du,
+                    const double* du2, const lapack_int* ipiv, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du,
+                    const lapack_complex_float* du2, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zgttrs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du,
+                    const lapack_complex_double* du2, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_spotrs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpotrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cpotrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zpotrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_spftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* a, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zpftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cpftrs( char* transr, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_spptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_cpptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zpptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_spbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const float* ab, lapack_int* ldab, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dpbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const double* ab, lapack_int* ldab, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_cpbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zpbtrs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_spttrs( lapack_int* n, lapack_int* nrhs, const float* d,
+                    const float* e, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dpttrs( lapack_int* n, lapack_int* nrhs, const double* d,
+                    const double* e, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_cpttrs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const lapack_complex_float* e, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zpttrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const lapack_complex_double* e,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ssytrs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, const lapack_int* ipiv, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dsytrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const lapack_int* ipiv,
+                    double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_csytrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zsytrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_chetrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_zhetrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_ssptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, const lapack_int* ipiv, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dsptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, const lapack_int* ipiv, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_csptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zsptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_chptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, const lapack_int* ipiv,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zhptrs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, const lapack_int* ipiv,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_strtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dtrtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* a, lapack_int* lda,
+                    double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_ctrtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ztrtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_stptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* ap, float* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_dtptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* ap, double* b,
+                    lapack_int* ldb, lapack_int *info );
+void LAPACK_ctptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* ap,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ztptrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* ap,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_stbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const float* ab,
+                    lapack_int* ldab, float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dtbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const double* ab,
+                    lapack_int* ldab, double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ctbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_ztbtrs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_sgecon( char* norm, lapack_int* n, const float* a, lapack_int* lda,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgecon( char* norm, lapack_int* n, const double* a, lapack_int* lda,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgecon( char* norm, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* anorm, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgecon( char* norm, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, double* anorm, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const float* ab, lapack_int* ldab, const lapack_int* ipiv,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const double* ab, lapack_int* ldab, const lapack_int* ipiv,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgbcon( char* norm, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgtcon( char* norm, lapack_int* n, const float* dl, const float* d,
+                    const float* du, const float* du2, const lapack_int* ipiv,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgtcon( char* norm, lapack_int* n, const double* dl,
+                    const double* d, const double* du, const double* du2,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgtcon( char* norm, lapack_int* n, const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du,
+                    const lapack_complex_float* du2, const lapack_int* ipiv,
+                    float* anorm, float* rcond, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zgtcon( char* norm, lapack_int* n, const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du,
+                    const lapack_complex_double* du2, const lapack_int* ipiv,
+                    double* anorm, double* rcond, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_spocon( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                    float* anorm, float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dpocon( char* uplo, lapack_int* n, const double* a, lapack_int* lda,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cpocon( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* anorm, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zpocon( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, double* anorm, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sppcon( char* uplo, lapack_int* n, const float* ap, float* anorm,
+                    float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dppcon( char* uplo, lapack_int* n, const double* ap, double* anorm,
+                    double* rcond, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cppcon( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    float* anorm, float* rcond, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zppcon( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    double* anorm, double* rcond, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_spbcon( char* uplo, lapack_int* n, lapack_int* kd, const float* ab,
+                    lapack_int* ldab, float* anorm, float* rcond, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dpbcon( char* uplo, lapack_int* n, lapack_int* kd, const double* ab,
+                    lapack_int* ldab, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cpbcon( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    float* anorm, float* rcond, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zpbcon( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    double* anorm, double* rcond, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sptcon( lapack_int* n, const float* d, const float* e, float* anorm,
+                    float* rcond, float* work, lapack_int *info );
+void LAPACK_dptcon( lapack_int* n, const double* d, const double* e,
+                    double* anorm, double* rcond, double* work,
+                    lapack_int *info );
+void LAPACK_cptcon( lapack_int* n, const float* d,
+                    const lapack_complex_float* e, float* anorm, float* rcond,
+                    float* work, lapack_int *info );
+void LAPACK_zptcon( lapack_int* n, const double* d,
+                    const lapack_complex_double* e, double* anorm,
+                    double* rcond, double* work, lapack_int *info );
+void LAPACK_ssycon( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dsycon( char* uplo, lapack_int* n, const double* a, lapack_int* lda,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_csycon( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv, float* anorm,
+                    float* rcond, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zsycon( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv, double* anorm,
+                    double* rcond, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_checon( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv, float* anorm,
+                    float* rcond, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zhecon( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv, double* anorm,
+                    double* rcond, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_sspcon( char* uplo, lapack_int* n, const float* ap,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dspcon( char* uplo, lapack_int* n, const double* ap,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cspcon( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zspcon( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_chpcon( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_int* ipiv, float* anorm, float* rcond,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zhpcon( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_int* ipiv, double* anorm, double* rcond,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_strcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const float* a, lapack_int* lda, float* rcond, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtrcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const double* a, lapack_int* lda, double* rcond,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_ctrcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    float* rcond, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztrcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    double* rcond, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const float* ap, float* rcond, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const double* ap, double* rcond, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_float* ap, float* rcond,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztpcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    const lapack_complex_double* ap, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const float* ab, lapack_int* ldab,
+                    float* rcond, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dtbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const double* ab, lapack_int* ldab,
+                    double* rcond, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const lapack_complex_float* ab,
+                    lapack_int* ldab, float* rcond, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_ztbcon( char* norm, char* uplo, char* diag, lapack_int* n,
+                    lapack_int* kd, const lapack_complex_double* ab,
+                    lapack_int* ldab, double* rcond,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* a, lapack_int* lda, const float* af,
+                    lapack_int* ldaf, const lapack_int* ipiv, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* ferr,
+                    float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const double* af,
+                    lapack_int* ldaf, const lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgerfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const double* a, lapack_int* lda, const double* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const double* r,
+                     const double* c, const double* b, lapack_int* ldb,
+                     double* x, lapack_int* ldx, double* rcond, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const float* a, lapack_int* lda, const float* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const float* r,
+                     const float* c, const float* b, lapack_int* ldb, float* x,
+                     lapack_int* ldx, float* rcond, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const double* r, const double* c,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgerfsx( char* trans, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const float* r, const float* c,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const float* ab, lapack_int* ldab,
+                    const float* afb, lapack_int* ldafb, const lapack_int* ipiv,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const double* ab, lapack_int* ldab,
+                    const double* afb, lapack_int* ldafb,
+                    const lapack_int* ipiv, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_float* ab,
+                    lapack_int* ldab, const lapack_complex_float* afb,
+                    lapack_int* ldafb, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgbrfs( char* trans, lapack_int* n, lapack_int* kl, lapack_int* ku,
+                    lapack_int* nrhs, const lapack_complex_double* ab,
+                    lapack_int* ldab, const lapack_complex_double* afb,
+                    lapack_int* ldafb, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_dgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, const double* ab,
+                     lapack_int* ldab, const double* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const double* r, const double* c,
+                     const double* b, lapack_int* ldb, double* x,
+                     lapack_int* ldx, double* rcond, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, const float* ab,
+                     lapack_int* ldab, const float* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const float* r, const float* c,
+                     const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params, float* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_zgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs,
+                     const lapack_complex_double* ab, lapack_int* ldab,
+                     const lapack_complex_double* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const double* r, const double* c,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgbrfsx( char* trans, char* equed, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs,
+                     const lapack_complex_float* ab, lapack_int* ldab,
+                     const lapack_complex_float* afb, lapack_int* ldafb,
+                     const lapack_int* ipiv, const float* r, const float* c,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* dl, const float* d, const float* du,
+                    const float* dlf, const float* df, const float* duf,
+                    const float* du2, const lapack_int* ipiv, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* ferr,
+                    float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* dl, const double* d, const double* du,
+                    const double* dlf, const double* df, const double* duf,
+                    const double* du2, const lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du,
+                    const lapack_complex_float* dlf,
+                    const lapack_complex_float* df,
+                    const lapack_complex_float* duf,
+                    const lapack_complex_float* du2, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgtrfs( char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du,
+                    const lapack_complex_double* dlf,
+                    const lapack_complex_double* df,
+                    const lapack_complex_double* duf,
+                    const lapack_complex_double* du2, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sporfs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, const float* af, lapack_int* ldaf,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dporfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const double* af,
+                    lapack_int* ldaf, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cporfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zporfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_dporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const double* a, lapack_int* lda, const double* af,
+                     lapack_int* ldaf, const double* s, const double* b,
+                     lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params, double* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_sporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const float* a, lapack_int* lda, const float* af,
+                     lapack_int* ldaf, const float* s, const float* b,
+                     lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const double* s, const lapack_complex_double* b,
+                     lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                     double* rcond, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cporfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const float* s, const lapack_complex_float* b,
+                     lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                     float* rcond, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_spprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, const float* afp, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* ferr,
+                    float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dpprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, const double* afp, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cpprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap,
+                    const lapack_complex_float* afp,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zpprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap,
+                    const lapack_complex_double* afp,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_spbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const float* ab, lapack_int* ldab, const float* afb,
+                    lapack_int* ldafb, const float* b, lapack_int* ldb,
+                    float* x, lapack_int* ldx, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dpbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const double* ab, lapack_int* ldab, const double* afb,
+                    lapack_int* ldafb, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cpbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_complex_float* afb, lapack_int* ldafb,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zpbrfs( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_complex_double* afb, lapack_int* ldafb,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sptrfs( lapack_int* n, lapack_int* nrhs, const float* d,
+                    const float* e, const float* df, const float* ef,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int *info );
+void LAPACK_dptrfs( lapack_int* n, lapack_int* nrhs, const double* d,
+                    const double* e, const double* df, const double* ef,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* ferr, double* berr, double* work,
+                    lapack_int *info );
+void LAPACK_cptrfs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const lapack_complex_float* e, const float* df,
+                    const lapack_complex_float* ef,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zptrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const lapack_complex_double* e,
+                    const double* df, const lapack_complex_double* ef,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_ssyrfs( char* uplo, lapack_int* n, lapack_int* nrhs, const float* a,
+                    lapack_int* lda, const float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const float* b, lapack_int* ldb,
+                    float* x, lapack_int* ldx, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dsyrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, const double* af,
+                    lapack_int* ldaf, const lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_csyrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zsyrfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dsyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const double* a, lapack_int* lda, const double* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const double* s,
+                     const double* b, lapack_int* ldb, double* x,
+                     lapack_int* ldx, double* rcond, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_ssyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const float* a, lapack_int* lda, const float* af,
+                     lapack_int* ldaf, const lapack_int* ipiv, const float* s,
+                     const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params, float* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_zsyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const double* s,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_csyrfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const float* s,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_cherfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zherfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* af, lapack_int* ldaf,
+                    const lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_zherfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_complex_double* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const double* s,
+                     const lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cherfsx( char* uplo, char* equed, lapack_int* n, lapack_int* nrhs,
+                     const lapack_complex_float* a, lapack_int* lda,
+                     const lapack_complex_float* af, lapack_int* ldaf,
+                     const lapack_int* ipiv, const float* s,
+                     const lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* berr, lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_ssprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, const float* afp, const lapack_int* ipiv,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dsprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, const double* afp, const lapack_int* ipiv,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_csprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap,
+                    const lapack_complex_float* afp, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zsprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap,
+                    const lapack_complex_double* afp, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_chprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap,
+                    const lapack_complex_float* afp, const lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhprfs( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap,
+                    const lapack_complex_double* afp, const lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* ferr,
+                    double* berr, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_strrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* a, lapack_int* lda,
+                    const float* b, lapack_int* ldb, const float* x,
+                    lapack_int* ldx, float* ferr, float* berr, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtrrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* a, lapack_int* lda,
+                    const double* b, lapack_int* ldb, const double* x,
+                    lapack_int* ldx, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctrrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* b,
+                    lapack_int* ldb, const lapack_complex_float* x,
+                    lapack_int* ldx, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztrrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* b,
+                    lapack_int* ldb, const lapack_complex_double* x,
+                    lapack_int* ldx, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const float* ap, const float* b,
+                    lapack_int* ldb, const float* x, lapack_int* ldx,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dtprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const double* ap, const double* b,
+                    lapack_int* ldb, const double* x, lapack_int* ldx,
+                    double* ferr, double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_float* ap,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    const lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztprfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* nrhs, const lapack_complex_double* ap,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    const lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_stbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const float* ab,
+                    lapack_int* ldab, const float* b, lapack_int* ldb,
+                    const float* x, lapack_int* ldx, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dtbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs, const double* ab,
+                    lapack_int* ldab, const double* b, lapack_int* ldb,
+                    const double* x, lapack_int* ldx, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    const lapack_complex_float* x, lapack_int* ldx, float* ferr,
+                    float* berr, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztbrfs( char* uplo, char* trans, char* diag, lapack_int* n,
+                    lapack_int* kd, lapack_int* nrhs,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    const lapack_complex_double* x, lapack_int* ldx,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgetri( lapack_int* n, float* a, lapack_int* lda,
+                    const lapack_int* ipiv, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgetri( lapack_int* n, double* a, lapack_int* lda,
+                    const lapack_int* ipiv, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgetri( lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zgetri( lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    const lapack_int* ipiv, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_spotri( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dpotri( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_cpotri( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_zpotri( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dpftri( char* transr, char* uplo, lapack_int* n, double* a,
+                    lapack_int *info );
+void LAPACK_spftri( char* transr, char* uplo, lapack_int* n, float* a,
+                    lapack_int *info );
+void LAPACK_zpftri( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int *info );
+void LAPACK_cpftri( char* transr, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int *info );
+void LAPACK_spptri( char* uplo, lapack_int* n, float* ap, lapack_int *info );
+void LAPACK_dpptri( char* uplo, lapack_int* n, double* ap, lapack_int *info );
+void LAPACK_cpptri( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_zpptri( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_ssytri( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    const lapack_int* ipiv, float* work, lapack_int *info );
+void LAPACK_dsytri( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    const lapack_int* ipiv, double* work, lapack_int *info );
+void LAPACK_csytri( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zsytri( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_chetri( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zhetri( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, const lapack_int* ipiv,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_ssptri( char* uplo, lapack_int* n, float* ap,
+                    const lapack_int* ipiv, float* work, lapack_int *info );
+void LAPACK_dsptri( char* uplo, lapack_int* n, double* ap,
+                    const lapack_int* ipiv, double* work, lapack_int *info );
+void LAPACK_csptri( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    const lapack_int* ipiv, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zsptri( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    const lapack_int* ipiv, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_chptri( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    const lapack_int* ipiv, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zhptri( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    const lapack_int* ipiv, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_strtri( char* uplo, char* diag, lapack_int* n, float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dtrtri( char* uplo, char* diag, lapack_int* n, double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_ctrtri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_ztrtri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dtftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    double* a, lapack_int *info );
+void LAPACK_stftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    float* a, lapack_int *info );
+void LAPACK_ztftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_double* a, lapack_int *info );
+void LAPACK_ctftri( char* transr, char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_float* a, lapack_int *info );
+void LAPACK_stptri( char* uplo, char* diag, lapack_int* n, float* ap,
+                    lapack_int *info );
+void LAPACK_dtptri( char* uplo, char* diag, lapack_int* n, double* ap,
+                    lapack_int *info );
+void LAPACK_ctptri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_float* ap, lapack_int *info );
+void LAPACK_ztptri( char* uplo, char* diag, lapack_int* n,
+                    lapack_complex_double* ap, lapack_int *info );
+void LAPACK_sgeequ( lapack_int* m, lapack_int* n, const float* a,
+                    lapack_int* lda, float* r, float* c, float* rowcnd,
+                    float* colcnd, float* amax, lapack_int *info );
+void LAPACK_dgeequ( lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, double* r, double* c, double* rowcnd,
+                    double* colcnd, double* amax, lapack_int *info );
+void LAPACK_cgeequ( lapack_int* m, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* r, float* c, float* rowcnd,
+                    float* colcnd, float* amax, lapack_int *info );
+void LAPACK_zgeequ( lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* r,
+                    double* c, double* rowcnd, double* colcnd, double* amax,
+                    lapack_int *info );
+void LAPACK_dgeequb( lapack_int* m, lapack_int* n, const double* a,
+                     lapack_int* lda, double* r, double* c, double* rowcnd,
+                     double* colcnd, double* amax, lapack_int *info );
+void LAPACK_sgeequb( lapack_int* m, lapack_int* n, const float* a,
+                     lapack_int* lda, float* r, float* c, float* rowcnd,
+                     float* colcnd, float* amax, lapack_int *info );
+void LAPACK_zgeequb( lapack_int* m, lapack_int* n,
+                     const lapack_complex_double* a, lapack_int* lda, double* r,
+                     double* c, double* rowcnd, double* colcnd, double* amax,
+                     lapack_int *info );
+void LAPACK_cgeequb( lapack_int* m, lapack_int* n,
+                     const lapack_complex_float* a, lapack_int* lda, float* r,
+                     float* c, float* rowcnd, float* colcnd, float* amax,
+                     lapack_int *info );
+void LAPACK_sgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const float* ab, lapack_int* ldab, float* r,
+                    float* c, float* rowcnd, float* colcnd, float* amax,
+                    lapack_int *info );
+void LAPACK_dgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const double* ab, lapack_int* ldab,
+                    double* r, double* c, double* rowcnd, double* colcnd,
+                    double* amax, lapack_int *info );
+void LAPACK_cgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const lapack_complex_float* ab,
+                    lapack_int* ldab, float* r, float* c, float* rowcnd,
+                    float* colcnd, float* amax, lapack_int *info );
+void LAPACK_zgbequ( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const lapack_complex_double* ab,
+                    lapack_int* ldab, double* r, double* c, double* rowcnd,
+                    double* colcnd, double* amax, lapack_int *info );
+void LAPACK_dgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const double* ab, lapack_int* ldab,
+                     double* r, double* c, double* rowcnd, double* colcnd,
+                     double* amax, lapack_int *info );
+void LAPACK_sgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const float* ab, lapack_int* ldab,
+                     float* r, float* c, float* rowcnd, float* colcnd,
+                     float* amax, lapack_int *info );
+void LAPACK_zgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const lapack_complex_double* ab,
+                     lapack_int* ldab, double* r, double* c, double* rowcnd,
+                     double* colcnd, double* amax, lapack_int *info );
+void LAPACK_cgbequb( lapack_int* m, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, const lapack_complex_float* ab,
+                     lapack_int* ldab, float* r, float* c, float* rowcnd,
+                     float* colcnd, float* amax, lapack_int *info );
+void LAPACK_spoequ( lapack_int* n, const float* a, lapack_int* lda, float* s,
+                    float* scond, float* amax, lapack_int *info );
+void LAPACK_dpoequ( lapack_int* n, const double* a, lapack_int* lda, double* s,
+                    double* scond, double* amax, lapack_int *info );
+void LAPACK_cpoequ( lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, float* s, float* scond, float* amax,
+                    lapack_int *info );
+void LAPACK_zpoequ( lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, double* s, double* scond, double* amax,
+                    lapack_int *info );
+void LAPACK_dpoequb( lapack_int* n, const double* a, lapack_int* lda, double* s,
+                     double* scond, double* amax, lapack_int *info );
+void LAPACK_spoequb( lapack_int* n, const float* a, lapack_int* lda, float* s,
+                     float* scond, float* amax, lapack_int *info );
+void LAPACK_zpoequb( lapack_int* n, const lapack_complex_double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     lapack_int *info );
+void LAPACK_cpoequb( lapack_int* n, const lapack_complex_float* a,
+                     lapack_int* lda, float* s, float* scond, float* amax,
+                     lapack_int *info );
+void LAPACK_sppequ( char* uplo, lapack_int* n, const float* ap, float* s,
+                    float* scond, float* amax, lapack_int *info );
+void LAPACK_dppequ( char* uplo, lapack_int* n, const double* ap, double* s,
+                    double* scond, double* amax, lapack_int *info );
+void LAPACK_cppequ( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    float* s, float* scond, float* amax, lapack_int *info );
+void LAPACK_zppequ( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    double* s, double* scond, double* amax, lapack_int *info );
+void LAPACK_spbequ( char* uplo, lapack_int* n, lapack_int* kd, const float* ab,
+                    lapack_int* ldab, float* s, float* scond, float* amax,
+                    lapack_int *info );
+void LAPACK_dpbequ( char* uplo, lapack_int* n, lapack_int* kd, const double* ab,
+                    lapack_int* ldab, double* s, double* scond, double* amax,
+                    lapack_int *info );
+void LAPACK_cpbequ( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_float* ab, lapack_int* ldab, float* s,
+                    float* scond, float* amax, lapack_int *info );
+void LAPACK_zpbequ( char* uplo, lapack_int* n, lapack_int* kd,
+                    const lapack_complex_double* ab, lapack_int* ldab,
+                    double* s, double* scond, double* amax, lapack_int *info );
+void LAPACK_dsyequb( char* uplo, lapack_int* n, const double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     double* work, lapack_int *info );
+void LAPACK_ssyequb( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                     float* s, float* scond, float* amax, float* work,
+                     lapack_int *info );
+void LAPACK_zsyequb( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_csyequb( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                     lapack_int* lda, float* s, float* scond, float* amax,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_zheequb( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                     lapack_int* lda, double* s, double* scond, double* amax,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_cheequb( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                     lapack_int* lda, float* s, float* scond, float* amax,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_sgesv( lapack_int* n, lapack_int* nrhs, float* a, lapack_int* lda,
+                   lapack_int* ipiv, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dgesv( lapack_int* n, lapack_int* nrhs, double* a, lapack_int* lda,
+                   lapack_int* ipiv, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cgesv( lapack_int* n, lapack_int* nrhs, lapack_complex_float* a,
+                   lapack_int* lda, lapack_int* ipiv, lapack_complex_float* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_zgesv( lapack_int* n, lapack_int* nrhs, lapack_complex_double* a,
+                   lapack_int* lda, lapack_int* ipiv, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_dsgesv( lapack_int* n, lapack_int* nrhs, double* a, lapack_int* lda,
+                    lapack_int* ipiv, double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* work, float* swork,
+                    lapack_int* iter, lapack_int *info );
+void LAPACK_zcgesv( lapack_int* n, lapack_int* nrhs, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    lapack_complex_double* work, lapack_complex_float* swork,
+                    double* rwork, lapack_int* iter, lapack_int *info );
+void LAPACK_sgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, float* r, float* c, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, double* r, double* c,
+                    double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, float* r, float* c,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgesvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, char* equed, double* r, double* c,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                     double* rcond, double* rpvgrw, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* r, float* c,
+                     float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* rpvgrw, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgesvxx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* r, float* c,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, float* ab, lapack_int* ldab,
+                   lapack_int* ipiv, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, double* ab, lapack_int* ldab,
+                   lapack_int* ipiv, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, lapack_complex_float* ab, lapack_int* ldab,
+                   lapack_int* ipiv, lapack_complex_float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_zgbsv( lapack_int* n, lapack_int* kl, lapack_int* ku,
+                   lapack_int* nrhs, lapack_complex_double* ab,
+                   lapack_int* ldab, lapack_int* ipiv, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_sgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, float* ab,
+                    lapack_int* ldab, float* afb, lapack_int* ldafb,
+                    lapack_int* ipiv, char* equed, float* r, float* c, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, double* ab,
+                    lapack_int* ldab, double* afb, lapack_int* ldafb,
+                    lapack_int* ipiv, char* equed, double* r, double* c,
+                    double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, lapack_complex_float* ab,
+                    lapack_int* ldab, lapack_complex_float* afb,
+                    lapack_int* ldafb, lapack_int* ipiv, char* equed, float* r,
+                    float* c, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgbsvx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, lapack_int* nrhs, lapack_complex_double* ab,
+                    lapack_int* ldab, lapack_complex_double* afb,
+                    lapack_int* ldafb, lapack_int* ipiv, char* equed, double* r,
+                    double* c, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, double* ab,
+                     lapack_int* ldab, double* afb, lapack_int* ldafb,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     double* b, lapack_int* ldb, double* x, lapack_int* ldx,
+                     double* rcond, double* rpvgrw, double* berr,
+                     lapack_int* n_err_bnds, double* err_bnds_norm,
+                     double* err_bnds_comp, lapack_int* nparams, double* params,
+                     double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, float* ab,
+                     lapack_int* ldab, float* afb, lapack_int* ldafb,
+                     lapack_int* ipiv, char* equed, float* r, float* c,
+                     float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                     float* rcond, float* rpvgrw, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs,
+                     lapack_complex_double* ab, lapack_int* ldab,
+                     lapack_complex_double* afb, lapack_int* ldafb,
+                     lapack_int* ipiv, char* equed, double* r, double* c,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cgbsvxx( char* fact, char* trans, lapack_int* n, lapack_int* kl,
+                     lapack_int* ku, lapack_int* nrhs, lapack_complex_float* ab,
+                     lapack_int* ldab, lapack_complex_float* afb,
+                     lapack_int* ldafb, lapack_int* ipiv, char* equed, float* r,
+                     float* c, lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sgtsv( lapack_int* n, lapack_int* nrhs, float* dl, float* d,
+                   float* du, float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dgtsv( lapack_int* n, lapack_int* nrhs, double* dl, double* d,
+                   double* du, double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cgtsv( lapack_int* n, lapack_int* nrhs, lapack_complex_float* dl,
+                   lapack_complex_float* d, lapack_complex_float* du,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zgtsv( lapack_int* n, lapack_int* nrhs, lapack_complex_double* dl,
+                   lapack_complex_double* d, lapack_complex_double* du,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_sgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const float* dl, const float* d, const float* du,
+                    float* dlf, float* df, float* duf, float* du2,
+                    lapack_int* ipiv, const float* b, lapack_int* ldb, float* x,
+                    lapack_int* ldx, float* rcond, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const double* dl, const double* d, const double* du,
+                    double* dlf, double* df, double* duf, double* du2,
+                    lapack_int* ipiv, const double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* rcond, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* dl,
+                    const lapack_complex_float* d,
+                    const lapack_complex_float* du, lapack_complex_float* dlf,
+                    lapack_complex_float* df, lapack_complex_float* duf,
+                    lapack_complex_float* du2, lapack_int* ipiv,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgtsvx( char* fact, char* trans, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* dl,
+                    const lapack_complex_double* d,
+                    const lapack_complex_double* du, lapack_complex_double* dlf,
+                    lapack_complex_double* df, lapack_complex_double* duf,
+                    lapack_complex_double* du2, lapack_int* ipiv,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sposv( char* uplo, lapack_int* n, lapack_int* nrhs, float* a,
+                   lapack_int* lda, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dposv( char* uplo, lapack_int* n, lapack_int* nrhs, double* a,
+                   lapack_int* lda, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cposv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zposv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dsposv( char* uplo, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* work, float* swork,
+                    lapack_int* iter, lapack_int *info );
+void LAPACK_zcposv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx,
+                    lapack_complex_double* work, lapack_complex_float* swork,
+                    double* rwork, lapack_int* iter, lapack_int *info );
+void LAPACK_sposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                    char* equed, float* s, float* b, lapack_int* ldb, float* x,
+                    lapack_int* ldx, float* rcond, float* ferr, float* berr,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                    char* equed, double* s, double* b, lapack_int* ldb,
+                    double* x, lapack_int* ldx, double* rcond, double* ferr,
+                    double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf, char* equed,
+                    float* s, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zposvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf, char* equed,
+                    double* s, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_dposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                     char* equed, double* s, double* b, lapack_int* ldb,
+                     double* x, lapack_int* ldx, double* rcond, double* rpvgrw,
+                     double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params, double* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_sposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                     char* equed, float* s, float* b, lapack_int* ldb, float* x,
+                     lapack_int* ldx, float* rcond, float* rpvgrw, float* berr,
+                     lapack_int* n_err_bnds, float* err_bnds_norm,
+                     float* err_bnds_comp, lapack_int* nparams, float* params,
+                     float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_zposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf, char* equed,
+                     double* s, lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_cposvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf, char* equed,
+                     float* s, lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sppsv( char* uplo, lapack_int* n, lapack_int* nrhs, float* ap,
+                   float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dppsv( char* uplo, lapack_int* n, lapack_int* nrhs, double* ap,
+                   double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cppsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* ap, lapack_complex_float* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_zppsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* ap, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_sppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    float* ap, float* afp, char* equed, float* s, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    double* ap, double* afp, char* equed, double* s, double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* ap, lapack_complex_float* afp,
+                    char* equed, float* s, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zppsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* ap, lapack_complex_double* afp,
+                    char* equed, double* s, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_spbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   float* ab, lapack_int* ldab, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dpbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   double* ab, lapack_int* ldab, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cpbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   lapack_complex_float* ab, lapack_int* ldab,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zpbsv( char* uplo, lapack_int* n, lapack_int* kd, lapack_int* nrhs,
+                   lapack_complex_double* ab, lapack_int* ldab,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_spbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, float* ab, lapack_int* ldab, float* afb,
+                    lapack_int* ldafb, char* equed, float* s, float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dpbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, double* ab, lapack_int* ldab, double* afb,
+                    lapack_int* ldafb, char* equed, double* s, double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_cpbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, lapack_complex_float* ab,
+                    lapack_int* ldab, lapack_complex_float* afb,
+                    lapack_int* ldafb, char* equed, float* s,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zpbsvx( char* fact, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_int* nrhs, lapack_complex_double* ab,
+                    lapack_int* ldab, lapack_complex_double* afb,
+                    lapack_int* ldafb, char* equed, double* s,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sptsv( lapack_int* n, lapack_int* nrhs, float* d, float* e,
+                   float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_dptsv( lapack_int* n, lapack_int* nrhs, double* d, double* e,
+                   double* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_cptsv( lapack_int* n, lapack_int* nrhs, float* d,
+                   lapack_complex_float* e, lapack_complex_float* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_zptsv( lapack_int* n, lapack_int* nrhs, double* d,
+                   lapack_complex_double* e, lapack_complex_double* b,
+                   lapack_int* ldb, lapack_int *info );
+void LAPACK_sptsvx( char* fact, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const float* e, float* df, float* ef, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int *info );
+void LAPACK_dptsvx( char* fact, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const double* e, double* df, double* ef,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* rcond, double* ferr, double* berr,
+                    double* work, lapack_int *info );
+void LAPACK_cptsvx( char* fact, lapack_int* n, lapack_int* nrhs, const float* d,
+                    const lapack_complex_float* e, float* df,
+                    lapack_complex_float* ef, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zptsvx( char* fact, lapack_int* n, lapack_int* nrhs,
+                    const double* d, const lapack_complex_double* e, double* df,
+                    lapack_complex_double* ef, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_ssysv( char* uplo, lapack_int* n, lapack_int* nrhs, float* a,
+                   lapack_int* lda, lapack_int* ipiv, float* b, lapack_int* ldb,
+                   float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dsysv( char* uplo, lapack_int* n, lapack_int* nrhs, double* a,
+                   lapack_int* lda, lapack_int* ipiv, double* b,
+                   lapack_int* ldb, double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_csysv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_zsysv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_ssysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* a, lapack_int* lda, float* af,
+                    lapack_int* ldaf, lapack_int* ipiv, const float* b,
+                    lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                    float* ferr, float* berr, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dsysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* a, lapack_int* lda, double* af,
+                    lapack_int* ldaf, lapack_int* ipiv, const double* b,
+                    lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                    double* ferr, double* berr, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_csysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zsysvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_dsysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     double* a, lapack_int* lda, double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* s, double* b,
+                     lapack_int* ldb, double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params, double* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_ssysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     float* a, lapack_int* lda, float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* s, float* b,
+                     lapack_int* ldb, float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params, float* work,
+                     lapack_int* iwork, lapack_int *info );
+void LAPACK_zsysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* s,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_csysvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* s,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_chesv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_zhesv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_chesvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhesvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* af, lapack_int* ldaf,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_zhesvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, double* s,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* x, lapack_int* ldx, double* rcond,
+                     double* rpvgrw, double* berr, lapack_int* n_err_bnds,
+                     double* err_bnds_norm, double* err_bnds_comp,
+                     lapack_int* nparams, double* params,
+                     lapack_complex_double* work, double* rwork,
+                     lapack_int *info );
+void LAPACK_chesvxx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* af, lapack_int* ldaf,
+                     lapack_int* ipiv, char* equed, float* s,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* x, lapack_int* ldx, float* rcond,
+                     float* rpvgrw, float* berr, lapack_int* n_err_bnds,
+                     float* err_bnds_norm, float* err_bnds_comp,
+                     lapack_int* nparams, float* params,
+                     lapack_complex_float* work, float* rwork,
+                     lapack_int *info );
+void LAPACK_sspsv( char* uplo, lapack_int* n, lapack_int* nrhs, float* ap,
+                   lapack_int* ipiv, float* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_dspsv( char* uplo, lapack_int* n, lapack_int* nrhs, double* ap,
+                   lapack_int* ipiv, double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_cspsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* ap, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zspsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* ap, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_sspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const float* ap, float* afp, lapack_int* ipiv,
+                    const float* b, lapack_int* ldb, float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr, float* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const double* ap, double* afp, lapack_int* ipiv,
+                    const double* b, lapack_int* ldb, double* x,
+                    lapack_int* ldx, double* rcond, double* ferr, double* berr,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, lapack_complex_float* afp,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zspsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, lapack_complex_double* afp,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_chpsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* ap, lapack_int* ipiv,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int *info );
+void LAPACK_zhpsv( char* uplo, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* ap, lapack_int* ipiv,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_int *info );
+void LAPACK_chpsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_float* ap, lapack_complex_float* afp,
+                    lapack_int* ipiv, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* x, lapack_int* ldx,
+                    float* rcond, float* ferr, float* berr,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhpsvx( char* fact, char* uplo, lapack_int* n, lapack_int* nrhs,
+                    const lapack_complex_double* ap, lapack_complex_double* afp,
+                    lapack_int* ipiv, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* x, lapack_int* ldx,
+                    double* rcond, double* ferr, double* berr,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sgeqrf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgeqrf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgeqrf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgeqrf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgeqpf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* jpvt, float* tau, float* work,
+                    lapack_int *info );
+void LAPACK_dgeqpf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* jpvt, double* tau, double* work,
+                    lapack_int *info );
+void LAPACK_cgeqpf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    float* rwork, lapack_int *info );
+void LAPACK_zgeqpf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgeqp3( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* jpvt, float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgeqp3( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* jpvt, double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgeqp3( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int *info );
+void LAPACK_zgeqp3( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* jpvt,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int *info );
+void LAPACK_sorgqr( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgqr( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungqr( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungqr( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmqr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgelqf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgelqf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgelqf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgelqf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sorglq( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorglq( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cunglq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zunglq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmlq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgeqlf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgeqlf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgeqlf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgeqlf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sorgql( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgql( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungql( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungql( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cunmql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmql( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgerqf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgerqf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgerqf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgerqf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sorgrq( lapack_int* m, lapack_int* n, lapack_int* k, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgrq( lapack_int* m, lapack_int* n, lapack_int* k, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungrq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungrq( lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cunmrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmrq( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_stzrzf( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dtzrzf( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ctzrzf( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ztzrzf( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l, const float* a,
+                    lapack_int* lda, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dormrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l, const double* a,
+                    lapack_int* lda, const double* tau, double* c,
+                    lapack_int* ldc, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmrz( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* k, lapack_int* l,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sggqrf( lapack_int* n, lapack_int* m, lapack_int* p, float* a,
+                    lapack_int* lda, float* taua, float* b, lapack_int* ldb,
+                    float* taub, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dggqrf( lapack_int* n, lapack_int* m, lapack_int* p, double* a,
+                    lapack_int* lda, double* taua, double* b, lapack_int* ldb,
+                    double* taub, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cggqrf( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* taua, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* taub,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zggqrf( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* taua, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* taub,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sggrqf( lapack_int* m, lapack_int* p, lapack_int* n, float* a,
+                    lapack_int* lda, float* taua, float* b, lapack_int* ldb,
+                    float* taub, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dggrqf( lapack_int* m, lapack_int* p, lapack_int* n, double* a,
+                    lapack_int* lda, double* taua, double* b, lapack_int* ldb,
+                    double* taub, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cggrqf( lapack_int* m, lapack_int* p, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* taua, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* taub,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zggrqf( lapack_int* m, lapack_int* p, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* taua, lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* taub,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgebrd( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* d, float* e, float* tauq, float* taup, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgebrd( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* d, double* e, double* tauq, double* taup,
+                    double* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_cgebrd( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, float* d, float* e,
+                    lapack_complex_float* tauq, lapack_complex_float* taup,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zgebrd( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, double* d, double* e,
+                    lapack_complex_double* tauq, lapack_complex_double* taup,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, float* ab, lapack_int* ldab,
+                    float* d, float* e, float* q, lapack_int* ldq, float* pt,
+                    lapack_int* ldpt, float* c, lapack_int* ldc, float* work,
+                    lapack_int *info );
+void LAPACK_dgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, double* ab,
+                    lapack_int* ldab, double* d, double* e, double* q,
+                    lapack_int* ldq, double* pt, lapack_int* ldpt, double* c,
+                    lapack_int* ldc, double* work, lapack_int *info );
+void LAPACK_cgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, lapack_complex_float* ab,
+                    lapack_int* ldab, float* d, float* e,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* pt, lapack_int* ldpt,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgbbrd( char* vect, lapack_int* m, lapack_int* n, lapack_int* ncc,
+                    lapack_int* kl, lapack_int* ku, lapack_complex_double* ab,
+                    lapack_int* ldab, double* d, double* e,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* pt, lapack_int* ldpt,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_sorgbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    float* a, lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorgbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    double* a, lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k, const float* a,
+                    lapack_int* lda, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dormbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k, const double* a,
+                    lapack_int* lda, const double* tau, double* c,
+                    lapack_int* ldc, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cungbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zungbr( char* vect, lapack_int* m, lapack_int* n, lapack_int* k,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmbr( char* vect, char* side, char* trans, lapack_int* m,
+                    lapack_int* n, lapack_int* k,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, float* d, float* e,
+                    float* vt, lapack_int* ldvt, float* u, lapack_int* ldu,
+                    float* c, lapack_int* ldc, float* work, lapack_int *info );
+void LAPACK_dbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, double* d, double* e,
+                    double* vt, lapack_int* ldvt, double* u, lapack_int* ldu,
+                    double* c, lapack_int* ldc, double* work,
+                    lapack_int *info );
+void LAPACK_cbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, float* d, float* e,
+                    lapack_complex_float* vt, lapack_int* ldvt,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* c, lapack_int* ldc, float* work,
+                    lapack_int *info );
+void LAPACK_zbdsqr( char* uplo, lapack_int* n, lapack_int* ncvt,
+                    lapack_int* nru, lapack_int* ncc, double* d, double* e,
+                    lapack_complex_double* vt, lapack_int* ldvt,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* c, lapack_int* ldc, double* work,
+                    lapack_int *info );
+void LAPACK_sbdsdc( char* uplo, char* compq, lapack_int* n, float* d, float* e,
+                    float* u, lapack_int* ldu, float* vt, lapack_int* ldvt,
+                    float* q, lapack_int* iq, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dbdsdc( char* uplo, char* compq, lapack_int* n, double* d,
+                    double* e, double* u, lapack_int* ldu, double* vt,
+                    lapack_int* ldvt, double* q, lapack_int* iq, double* work,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ssytrd( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    float* d, float* e, float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dsytrd( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    double* d, double* e, double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sorgtr( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    const float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dorgtr( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    const double* tau, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_sormtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const float* a, lapack_int* lda,
+                    const float* tau, float* c, lapack_int* ldc, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dormtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const double* a, lapack_int* lda,
+                    const double* tau, double* c, lapack_int* ldc, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_chetrd( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, float* d, float* e,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zhetrd( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, double* d, double* e,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cungtr( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zungtr( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_zunmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ssptrd( char* uplo, lapack_int* n, float* ap, float* d, float* e,
+                    float* tau, lapack_int *info );
+void LAPACK_dsptrd( char* uplo, lapack_int* n, double* ap, double* d, double* e,
+                    double* tau, lapack_int *info );
+void LAPACK_sopgtr( char* uplo, lapack_int* n, const float* ap,
+                    const float* tau, float* q, lapack_int* ldq, float* work,
+                    lapack_int *info );
+void LAPACK_dopgtr( char* uplo, lapack_int* n, const double* ap,
+                    const double* tau, double* q, lapack_int* ldq, double* work,
+                    lapack_int *info );
+void LAPACK_sopmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const float* ap, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int *info );
+void LAPACK_dopmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const double* ap, const double* tau,
+                    double* c, lapack_int* ldc, double* work,
+                    lapack_int *info );
+void LAPACK_chptrd( char* uplo, lapack_int* n, lapack_complex_float* ap,
+                    float* d, float* e, lapack_complex_float* tau,
+                    lapack_int *info );
+void LAPACK_zhptrd( char* uplo, lapack_int* n, lapack_complex_double* ap,
+                    double* d, double* e, lapack_complex_double* tau,
+                    lapack_int *info );
+void LAPACK_cupgtr( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_complex_float* tau, lapack_complex_float* q,
+                    lapack_int* ldq, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zupgtr( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_complex_double* tau, lapack_complex_double* q,
+                    lapack_int* ldq, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_cupmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* ap,
+                    const lapack_complex_float* tau, lapack_complex_float* c,
+                    lapack_int* ldc, lapack_complex_float* work,
+                    lapack_int *info );
+void LAPACK_zupmtr( char* side, char* uplo, char* trans, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* ap,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_ssbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    float* ab, lapack_int* ldab, float* d, float* e, float* q,
+                    lapack_int* ldq, float* work, lapack_int *info );
+void LAPACK_dsbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    double* ab, lapack_int* ldab, double* d, double* e,
+                    double* q, lapack_int* ldq, double* work,
+                    lapack_int *info );
+void LAPACK_chbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_float* ab, lapack_int* ldab, float* d,
+                    float* e, lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zhbtrd( char* vect, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_double* ab, lapack_int* ldab, double* d,
+                    double* e, lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_ssterf( lapack_int* n, float* d, float* e, lapack_int *info );
+void LAPACK_dsterf( lapack_int* n, double* d, double* e, lapack_int *info );
+void LAPACK_ssteqr( char* compz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dsteqr( char* compz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_csteqr( char* compz, lapack_int* n, float* d, float* e,
+                    lapack_complex_float* z, lapack_int* ldz, float* work,
+                    lapack_int *info );
+void LAPACK_zsteqr( char* compz, lapack_int* n, double* d, double* e,
+                    lapack_complex_double* z, lapack_int* ldz, double* work,
+                    lapack_int *info );
+void LAPACK_sstemr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    lapack_int* nzc, lapack_int* isuppz, lapack_logical* tryrac,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dstemr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, lapack_int* m, double* w, double* z,
+                    lapack_int* ldz, lapack_int* nzc, lapack_int* isuppz,
+                    lapack_logical* tryrac, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cstemr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_int* nzc, lapack_int* isuppz,
+                    lapack_logical* tryrac, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_zstemr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, lapack_int* m, double* w,
+                    lapack_complex_double* z, lapack_int* ldz, lapack_int* nzc,
+                    lapack_int* isuppz, lapack_logical* tryrac, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_sstedc( char* compz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dstedc( char* compz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cstedc( char* compz, lapack_int* n, float* d, float* e,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zstedc( char* compz, lapack_int* n, double* d, double* e,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sstegr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, lapack_int* isuppz, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_dstegr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, lapack_int* isuppz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_cstegr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_int* isuppz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_zstegr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_int* isuppz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_spteqr( char* compz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dpteqr( char* compz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_cpteqr( char* compz, lapack_int* n, float* d, float* e,
+                    lapack_complex_float* z, lapack_int* ldz, float* work,
+                    lapack_int *info );
+void LAPACK_zpteqr( char* compz, lapack_int* n, double* d, double* e,
+                    lapack_complex_double* z, lapack_int* ldz, double* work,
+                    lapack_int *info );
+void LAPACK_sstebz( char* range, char* order, lapack_int* n, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    const float* d, const float* e, lapack_int* m,
+                    lapack_int* nsplit, float* w, lapack_int* iblock,
+                    lapack_int* isplit, float* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dstebz( char* range, char* order, lapack_int* n, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    const double* d, const double* e, lapack_int* m,
+                    lapack_int* nsplit, double* w, lapack_int* iblock,
+                    lapack_int* isplit, double* work, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_sstein( lapack_int* n, const float* d, const float* e,
+                    lapack_int* m, const float* w, const lapack_int* iblock,
+                    const lapack_int* isplit, float* z, lapack_int* ldz,
+                    float* work, lapack_int* iwork, lapack_int* ifailv,
+                    lapack_int *info );
+void LAPACK_dstein( lapack_int* n, const double* d, const double* e,
+                    lapack_int* m, const double* w, const lapack_int* iblock,
+                    const lapack_int* isplit, double* z, lapack_int* ldz,
+                    double* work, lapack_int* iwork, lapack_int* ifailv,
+                    lapack_int *info );
+void LAPACK_cstein( lapack_int* n, const float* d, const float* e,
+                    lapack_int* m, const float* w, const lapack_int* iblock,
+                    const lapack_int* isplit, lapack_complex_float* z,
+                    lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifailv, lapack_int *info );
+void LAPACK_zstein( lapack_int* n, const double* d, const double* e,
+                    lapack_int* m, const double* w, const lapack_int* iblock,
+                    const lapack_int* isplit, lapack_complex_double* z,
+                    lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifailv, lapack_int *info );
+void LAPACK_sdisna( char* job, lapack_int* m, lapack_int* n, const float* d,
+                    float* sep, lapack_int *info );
+void LAPACK_ddisna( char* job, lapack_int* m, lapack_int* n, const double* d,
+                    double* sep, lapack_int *info );
+void LAPACK_ssygst( lapack_int* itype, char* uplo, lapack_int* n, float* a,
+                    lapack_int* lda, const float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_dsygst( lapack_int* itype, char* uplo, lapack_int* n, double* a,
+                    lapack_int* lda, const double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_chegst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_zhegst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int *info );
+void LAPACK_sspgst( lapack_int* itype, char* uplo, lapack_int* n, float* ap,
+                    const float* bp, lapack_int *info );
+void LAPACK_dspgst( lapack_int* itype, char* uplo, lapack_int* n, double* ap,
+                    const double* bp, lapack_int *info );
+void LAPACK_chpgst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, const lapack_complex_float* bp,
+                    lapack_int *info );
+void LAPACK_zhpgst( lapack_int* itype, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, const lapack_complex_double* bp,
+                    lapack_int *info );
+void LAPACK_ssbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, float* ab, lapack_int* ldab,
+                    const float* bb, lapack_int* ldbb, float* x,
+                    lapack_int* ldx, float* work, lapack_int *info );
+void LAPACK_dsbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, double* ab, lapack_int* ldab,
+                    const double* bb, lapack_int* ldbb, double* x,
+                    lapack_int* ldx, double* work, lapack_int *info );
+void LAPACK_chbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_float* ab, lapack_int* ldab,
+                    const lapack_complex_float* bb, lapack_int* ldbb,
+                    lapack_complex_float* x, lapack_int* ldx,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhbgst( char* vect, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_double* ab, lapack_int* ldab,
+                    const lapack_complex_double* bb, lapack_int* ldbb,
+                    lapack_complex_double* x, lapack_int* ldx,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_spbstf( char* uplo, lapack_int* n, lapack_int* kb, float* bb,
+                    lapack_int* ldbb, lapack_int *info );
+void LAPACK_dpbstf( char* uplo, lapack_int* n, lapack_int* kb, double* bb,
+                    lapack_int* ldbb, lapack_int *info );
+void LAPACK_cpbstf( char* uplo, lapack_int* n, lapack_int* kb,
+                    lapack_complex_float* bb, lapack_int* ldbb,
+                    lapack_int *info );
+void LAPACK_zpbstf( char* uplo, lapack_int* n, lapack_int* kb,
+                    lapack_complex_double* bb, lapack_int* ldbb,
+                    lapack_int *info );
+void LAPACK_sgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi, float* a,
+                    lapack_int* lda, float* tau, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi, double* a,
+                    lapack_int* lda, double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zgehrd( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sorghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi, float* a,
+                    lapack_int* lda, const float* tau, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dorghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi, double* a,
+                    lapack_int* lda, const double* tau, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sormhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, const float* a,
+                    lapack_int* lda, const float* tau, float* c,
+                    lapack_int* ldc, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dormhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, const double* a,
+                    lapack_int* lda, const double* tau, double* c,
+                    lapack_int* ldc, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zunghr( lapack_int* n, lapack_int* ilo, lapack_int* ihi,
+                    lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cunmhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* tau, lapack_complex_float* c,
+                    lapack_int* ldc, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zunmhr( char* side, char* trans, lapack_int* m, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* tau, lapack_complex_double* c,
+                    lapack_int* ldc, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sgebal( char* job, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ilo, lapack_int* ihi, float* scale,
+                    lapack_int *info );
+void LAPACK_dgebal( char* job, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ilo, lapack_int* ihi, double* scale,
+                    lapack_int *info );
+void LAPACK_cgebal( char* job, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ilo, lapack_int* ihi,
+                    float* scale, lapack_int *info );
+void LAPACK_zgebal( char* job, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ilo, lapack_int* ihi,
+                    double* scale, lapack_int *info );
+void LAPACK_sgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* scale, lapack_int* m,
+                    float* v, lapack_int* ldv, lapack_int *info );
+void LAPACK_dgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* scale, lapack_int* m,
+                    double* v, lapack_int* ldv, lapack_int *info );
+void LAPACK_cgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* scale, lapack_int* m,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_zgebak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* scale, lapack_int* m,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_shseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, float* h, lapack_int* ldh, float* wr,
+                    float* wi, float* z, lapack_int* ldz, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dhseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, double* h, lapack_int* ldh, double* wr,
+                    double* wi, double* z, lapack_int* ldz, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_chseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_float* h, lapack_int* ldh,
+                    lapack_complex_float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zhseqr( char* job, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_double* h, lapack_int* ldh,
+                    lapack_complex_double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_shsein( char* job, char* eigsrc, char* initv,
+                    lapack_logical* select, lapack_int* n, const float* h,
+                    lapack_int* ldh, float* wr, const float* wi, float* vl,
+                    lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, float* work,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_dhsein( char* job, char* eigsrc, char* initv,
+                    lapack_logical* select, lapack_int* n, const double* h,
+                    lapack_int* ldh, double* wr, const double* wi, double* vl,
+                    lapack_int* ldvl, double* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, double* work,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_chsein( char* job, char* eigsrc, char* initv,
+                    const lapack_logical* select, lapack_int* n,
+                    const lapack_complex_float* h, lapack_int* ldh,
+                    lapack_complex_float* w, lapack_complex_float* vl,
+                    lapack_int* ldvl, lapack_complex_float* vr,
+                    lapack_int* ldvr, lapack_int* mm, lapack_int* m,
+                    lapack_complex_float* work, float* rwork,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_zhsein( char* job, char* eigsrc, char* initv,
+                    const lapack_logical* select, lapack_int* n,
+                    const lapack_complex_double* h, lapack_int* ldh,
+                    lapack_complex_double* w, lapack_complex_double* vl,
+                    lapack_int* ldvl, lapack_complex_double* vr,
+                    lapack_int* ldvr, lapack_int* mm, lapack_int* m,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int* ifaill, lapack_int* ifailr, lapack_int *info );
+void LAPACK_strevc( char* side, char* howmny, lapack_logical* select,
+                    lapack_int* n, const float* t, lapack_int* ldt, float* vl,
+                    lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, float* work,
+                    lapack_int *info );
+void LAPACK_dtrevc( char* side, char* howmny, lapack_logical* select,
+                    lapack_int* n, const double* t, lapack_int* ldt, double* vl,
+                    lapack_int* ldvl, double* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, double* work,
+                    lapack_int *info );
+void LAPACK_ctrevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* vl, lapack_int* ldvl,
+                    lapack_complex_float* vr, lapack_int* ldvr, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztrevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* vl, lapack_int* ldvl,
+                    lapack_complex_double* vr, lapack_int* ldvr, lapack_int* mm,
+                    lapack_int* m, lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_strsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const float* t, lapack_int* ldt,
+                    const float* vl, lapack_int* ldvl, const float* vr,
+                    lapack_int* ldvr, float* s, float* sep, lapack_int* mm,
+                    lapack_int* m, float* work, lapack_int* ldwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dtrsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const double* t, lapack_int* ldt,
+                    const double* vl, lapack_int* ldvl, const double* vr,
+                    lapack_int* ldvr, double* s, double* sep, lapack_int* mm,
+                    lapack_int* m, double* work, lapack_int* ldwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctrsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_float* t,
+                    lapack_int* ldt, const lapack_complex_float* vl,
+                    lapack_int* ldvl, const lapack_complex_float* vr,
+                    lapack_int* ldvr, float* s, float* sep, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work,
+                    lapack_int* ldwork, float* rwork, lapack_int *info );
+void LAPACK_ztrsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_double* t,
+                    lapack_int* ldt, const lapack_complex_double* vl,
+                    lapack_int* ldvl, const lapack_complex_double* vr,
+                    lapack_int* ldvr, double* s, double* sep, lapack_int* mm,
+                    lapack_int* m, lapack_complex_double* work,
+                    lapack_int* ldwork, double* rwork, lapack_int *info );
+void LAPACK_strexc( char* compq, lapack_int* n, float* t, lapack_int* ldt,
+                    float* q, lapack_int* ldq, lapack_int* ifst,
+                    lapack_int* ilst, float* work, lapack_int *info );
+void LAPACK_dtrexc( char* compq, lapack_int* n, double* t, lapack_int* ldt,
+                    double* q, lapack_int* ldq, lapack_int* ifst,
+                    lapack_int* ilst, double* work, lapack_int *info );
+void LAPACK_ctrexc( char* compq, lapack_int* n, lapack_complex_float* t,
+                    lapack_int* ldt, lapack_complex_float* q, lapack_int* ldq,
+                    lapack_int* ifst, lapack_int* ilst, lapack_int *info );
+void LAPACK_ztrexc( char* compq, lapack_int* n, lapack_complex_double* t,
+                    lapack_int* ldt, lapack_complex_double* q, lapack_int* ldq,
+                    lapack_int* ifst, lapack_int* ilst, lapack_int *info );
+void LAPACK_strsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, float* t, lapack_int* ldt, float* q,
+                    lapack_int* ldq, float* wr, float* wi, lapack_int* m,
+                    float* s, float* sep, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dtrsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, double* t, lapack_int* ldt, double* q,
+                    lapack_int* ldq, double* wr, double* wi, lapack_int* m,
+                    double* s, double* sep, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_ctrsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* w, lapack_int* m, float* s,
+                    float* sep, lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_ztrsen( char* job, char* compq, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* w, lapack_int* m, double* s,
+                    double* sep, lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_strsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const float* a, lapack_int* lda,
+                    const float* b, lapack_int* ldb, float* c, lapack_int* ldc,
+                    float* scale, lapack_int *info );
+void LAPACK_dtrsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const double* a, lapack_int* lda,
+                    const double* b, lapack_int* ldb, double* c,
+                    lapack_int* ldc, double* scale, lapack_int *info );
+void LAPACK_ctrsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* b,
+                    lapack_int* ldb, lapack_complex_float* c, lapack_int* ldc,
+                    float* scale, lapack_int *info );
+void LAPACK_ztrsyl( char* trana, char* tranb, lapack_int* isgn, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* b,
+                    lapack_int* ldb, lapack_complex_double* c, lapack_int* ldc,
+                    double* scale, lapack_int *info );
+void LAPACK_sgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* q, lapack_int* ldq, float* z,
+                    lapack_int* ldz, lapack_int *info );
+void LAPACK_dgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* q, lapack_int* ldq, double* z,
+                    lapack_int* ldz, lapack_int *info );
+void LAPACK_cgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_int *info );
+void LAPACK_zgghrd( char* compq, char* compz, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_int *info );
+void LAPACK_sggbal( char* job, lapack_int* n, float* a, lapack_int* lda,
+                    float* b, lapack_int* ldb, lapack_int* ilo, lapack_int* ihi,
+                    float* lscale, float* rscale, float* work,
+                    lapack_int *info );
+void LAPACK_dggbal( char* job, lapack_int* n, double* a, lapack_int* lda,
+                    double* b, lapack_int* ldb, lapack_int* ilo,
+                    lapack_int* ihi, double* lscale, double* rscale,
+                    double* work, lapack_int *info );
+void LAPACK_cggbal( char* job, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                    lapack_int* ilo, lapack_int* ihi, float* lscale,
+                    float* rscale, float* work, lapack_int *info );
+void LAPACK_zggbal( char* job, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                    lapack_int* ilo, lapack_int* ihi, double* lscale,
+                    double* rscale, double* work, lapack_int *info );
+void LAPACK_sggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* lscale, const float* rscale,
+                    lapack_int* m, float* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_dggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* lscale, const double* rscale,
+                    lapack_int* m, double* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_cggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const float* lscale, const float* rscale,
+                    lapack_int* m, lapack_complex_float* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_zggbak( char* job, char* side, lapack_int* n, lapack_int* ilo,
+                    lapack_int* ihi, const double* lscale, const double* rscale,
+                    lapack_int* m, lapack_complex_double* v, lapack_int* ldv,
+                    lapack_int *info );
+void LAPACK_shgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, float* h, lapack_int* ldh,
+                    float* t, lapack_int* ldt, float* alphar, float* alphai,
+                    float* beta, float* q, lapack_int* ldq, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dhgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, double* h,
+                    lapack_int* ldh, double* t, lapack_int* ldt, double* alphar,
+                    double* alphai, double* beta, double* q, lapack_int* ldq,
+                    double* z, lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_chgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, lapack_complex_float* h,
+                    lapack_int* ldh, lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zhgeqz( char* job, char* compq, char* compz, lapack_int* n,
+                    lapack_int* ilo, lapack_int* ihi, lapack_complex_double* h,
+                    lapack_int* ldh, lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_stgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const float* s, lapack_int* lds,
+                    const float* p, lapack_int* ldp, float* vl,
+                    lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, float* work,
+                    lapack_int *info );
+void LAPACK_dtgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const double* s, lapack_int* lds,
+                    const double* p, lapack_int* ldp, double* vl,
+                    lapack_int* ldvl, double* vr, lapack_int* ldvr,
+                    lapack_int* mm, lapack_int* m, double* work,
+                    lapack_int *info );
+void LAPACK_ctgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_float* s,
+                    lapack_int* lds, const lapack_complex_float* p,
+                    lapack_int* ldp, lapack_complex_float* vl, lapack_int* ldvl,
+                    lapack_complex_float* vr, lapack_int* ldvr, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work, float* rwork,
+                    lapack_int *info );
+void LAPACK_ztgevc( char* side, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_double* s,
+                    lapack_int* lds, const lapack_complex_double* p,
+                    lapack_int* ldp, lapack_complex_double* vl,
+                    lapack_int* ldvl, lapack_complex_double* vr,
+                    lapack_int* ldvr, lapack_int* mm, lapack_int* m,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int *info );
+void LAPACK_stgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* q, lapack_int* ldq, float* z, lapack_int* ldz,
+                    lapack_int* ifst, lapack_int* ilst, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dtgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* q, lapack_int* ldq, double* z, lapack_int* ldz,
+                    lapack_int* ifst, lapack_int* ilst, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_ctgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz, lapack_int* ifst,
+                    lapack_int* ilst, lapack_int *info );
+void LAPACK_ztgexc( lapack_logical* wantq, lapack_logical* wantz, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz, lapack_int* ifst,
+                    lapack_int* ilst, lapack_int *info );
+void LAPACK_stgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* alphar, float* alphai, float* beta,
+                    float* q, lapack_int* ldq, float* z, lapack_int* ldz,
+                    lapack_int* m, float* pl, float* pr, float* dif,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dtgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* alphar, double* alphai,
+                    double* beta, double* q, lapack_int* ldq, double* z,
+                    lapack_int* ldz, lapack_int* m, double* pl, double* pr,
+                    double* dif, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_ctgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* z, lapack_int* ldz, lapack_int* m,
+                    float* pl, float* pr, float* dif,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_ztgsen( lapack_int* ijob, lapack_logical* wantq,
+                    lapack_logical* wantz, const lapack_logical* select,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* z, lapack_int* ldz, lapack_int* m,
+                    double* pl, double* pr, double* dif,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_stgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const float* a, lapack_int* lda, const float* b,
+                    lapack_int* ldb, float* c, lapack_int* ldc, const float* d,
+                    lapack_int* ldd, const float* e, lapack_int* lde, float* f,
+                    lapack_int* ldf, float* scale, float* dif, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_dtgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const double* a, lapack_int* lda, const double* b,
+                    lapack_int* ldb, double* c, lapack_int* ldc,
+                    const double* d, lapack_int* ldd, const double* e,
+                    lapack_int* lde, double* f, lapack_int* ldf, double* scale,
+                    double* dif, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ctgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    const lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    const lapack_complex_float* d, lapack_int* ldd,
+                    const lapack_complex_float* e, lapack_int* lde,
+                    lapack_complex_float* f, lapack_int* ldf, float* scale,
+                    float* dif, lapack_complex_float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ztgsyl( char* trans, lapack_int* ijob, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    const lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    const lapack_complex_double* d, lapack_int* ldd,
+                    const lapack_complex_double* e, lapack_int* lde,
+                    lapack_complex_double* f, lapack_int* ldf, double* scale,
+                    double* dif, lapack_complex_double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_stgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const float* a, lapack_int* lda,
+                    const float* b, lapack_int* ldb, const float* vl,
+                    lapack_int* ldvl, const float* vr, lapack_int* ldvr,
+                    float* s, float* dif, lapack_int* mm, lapack_int* m,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dtgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const double* a, lapack_int* lda,
+                    const double* b, lapack_int* ldb, const double* vl,
+                    lapack_int* ldvl, const double* vr, lapack_int* ldvr,
+                    double* s, double* dif, lapack_int* mm, lapack_int* m,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_ctgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, const lapack_complex_float* b,
+                    lapack_int* ldb, const lapack_complex_float* vl,
+                    lapack_int* ldvl, const lapack_complex_float* vr,
+                    lapack_int* ldvr, float* s, float* dif, lapack_int* mm,
+                    lapack_int* m, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_ztgsna( char* job, char* howmny, const lapack_logical* select,
+                    lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, const lapack_complex_double* b,
+                    lapack_int* ldb, const lapack_complex_double* vl,
+                    lapack_int* ldvl, const lapack_complex_double* vr,
+                    lapack_int* ldvr, double* s, double* dif, lapack_int* mm,
+                    lapack_int* m, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_sggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, float* a, lapack_int* lda,
+                    float* b, lapack_int* ldb, float* tola, float* tolb,
+                    lapack_int* k, lapack_int* l, float* u, lapack_int* ldu,
+                    float* v, lapack_int* ldv, float* q, lapack_int* ldq,
+                    lapack_int* iwork, float* tau, float* work,
+                    lapack_int *info );
+void LAPACK_dggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, double* a, lapack_int* lda,
+                    double* b, lapack_int* ldb, double* tola, double* tolb,
+                    lapack_int* k, lapack_int* l, double* u, lapack_int* ldu,
+                    double* v, lapack_int* ldv, double* q, lapack_int* ldq,
+                    lapack_int* iwork, double* tau, double* work,
+                    lapack_int *info );
+void LAPACK_cggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                    float* tola, float* tolb, lapack_int* k, lapack_int* l,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_complex_float* q, lapack_int* ldq, lapack_int* iwork,
+                    float* rwork, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zggsvp( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                    double* tola, double* tolb, lapack_int* k, lapack_int* l,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_int* iwork, double* rwork,
+                    lapack_complex_double* tau, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_stgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* tola, float* tolb, float* alpha, float* beta,
+                    float* u, lapack_int* ldu, float* v, lapack_int* ldv,
+                    float* q, lapack_int* ldq, float* work, lapack_int* ncycle,
+                    lapack_int *info );
+void LAPACK_dtgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* tola, double* tolb, double* alpha, double* beta,
+                    double* u, lapack_int* ldu, double* v, lapack_int* ldv,
+                    double* q, lapack_int* ldq, double* work,
+                    lapack_int* ncycle, lapack_int *info );
+void LAPACK_ctgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* tola,
+                    float* tolb, float* alpha, float* beta,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* work, lapack_int* ncycle,
+                    lapack_int *info );
+void LAPACK_ztgsja( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* p, lapack_int* n, lapack_int* k, lapack_int* l,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* tola,
+                    double* tolb, double* alpha, double* beta,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* work, lapack_int* ncycle,
+                    lapack_int *info );
+void LAPACK_sgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                   float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                   double* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_cgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_zgels( char* trans, lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_sgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb,
+                    lapack_int* jpvt, float* rcond, lapack_int* rank,
+                    float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb,
+                    lapack_int* jpvt, double* rcond, lapack_int* rank,
+                    double* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_cgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, lapack_int* jpvt,
+                    float* rcond, lapack_int* rank, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int *info );
+void LAPACK_zgelsy( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, lapack_int* jpvt,
+                    double* rcond, lapack_int* rank,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int *info );
+void LAPACK_zgelss( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_dgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* s,
+                    float* rcond, lapack_int* rank, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_zgelsd( lapack_int* m, lapack_int* n, lapack_int* nrhs,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* s,
+                    double* rcond, lapack_int* rank,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_sgglse( lapack_int* m, lapack_int* n, lapack_int* p, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* c,
+                    float* d, float* x, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dgglse( lapack_int* m, lapack_int* n, lapack_int* p, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* c,
+                    double* d, double* x, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cgglse( lapack_int* m, lapack_int* n, lapack_int* p,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* c, lapack_complex_float* d,
+                    lapack_complex_float* x, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zgglse( lapack_int* m, lapack_int* n, lapack_int* p,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* c, lapack_complex_double* d,
+                    lapack_complex_double* x, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sggglm( lapack_int* n, lapack_int* m, lapack_int* p, float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb, float* d,
+                    float* x, float* y, float* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_dggglm( lapack_int* n, lapack_int* m, lapack_int* p, double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb, double* d,
+                    double* x, double* y, double* work, lapack_int* lwork,
+                    lapack_int *info );
+void LAPACK_cggglm( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* d, lapack_complex_float* x,
+                    lapack_complex_float* y, lapack_complex_float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_zggglm( lapack_int* n, lapack_int* m, lapack_int* p,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* d, lapack_complex_double* x,
+                    lapack_complex_double* y, lapack_complex_double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_ssyev( char* jobz, char* uplo, lapack_int* n, float* a,
+                   lapack_int* lda, float* w, float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_dsyev( char* jobz, char* uplo, lapack_int* n, double* a,
+                   lapack_int* lda, double* w, double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_cheev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda, float* w,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zheev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda, double* w,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_int *info );
+void LAPACK_ssyevd( char* jobz, char* uplo, lapack_int* n, float* a,
+                    lapack_int* lda, float* w, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dsyevd( char* jobz, char* uplo, lapack_int* n, double* a,
+                    lapack_int* lda, double* w, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cheevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* w,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zheevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* w,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssyevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    float* a, lapack_int* lda, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dsyevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    double* a, lapack_int* lda, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, double* z, lapack_int* ldz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_cheevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_zheevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_ssyevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    float* a, lapack_int* lda, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    lapack_int* isuppz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dsyevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    double* a, lapack_int* lda, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, double* z, lapack_int* ldz,
+                    lapack_int* isuppz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_cheevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_int* isuppz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zheevr( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_int* isuppz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sspev( char* jobz, char* uplo, lapack_int* n, float* ap, float* w,
+                   float* z, lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dspev( char* jobz, char* uplo, lapack_int* n, double* ap, double* w,
+                   double* z, lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_chpev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* ap, float* w, lapack_complex_float* z,
+                   lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                   lapack_int *info );
+void LAPACK_zhpev( char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* ap, double* w,
+                   lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_sspevd( char* jobz, char* uplo, lapack_int* n, float* ap, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dspevd( char* jobz, char* uplo, lapack_int* n, double* ap,
+                    double* w, double* z, lapack_int* ldz, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_chpevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* lrwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_zhpevd( char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sspevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    float* ap, float* vl, float* vu, lapack_int* il,
+                    lapack_int* iu, float* abstol, lapack_int* m, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dspevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    double* ap, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_chpevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_zhpevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_ssbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   float* ab, lapack_int* ldab, float* w, float* z,
+                   lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dsbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   double* ab, lapack_int* ldab, double* w, double* z,
+                   lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_chbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   lapack_complex_float* ab, lapack_int* ldab, float* w,
+                   lapack_complex_float* z, lapack_int* ldz,
+                   lapack_complex_float* work, float* rwork, lapack_int *info );
+void LAPACK_zhbev( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                   lapack_complex_double* ab, lapack_int* ldab, double* w,
+                   lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_ssbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    float* ab, lapack_int* ldab, float* w, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dsbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    double* ab, lapack_int* ldab, double* w, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_chbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_float* ab, lapack_int* ldab, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhbevd( char* jobz, char* uplo, lapack_int* n, lapack_int* kd,
+                    lapack_complex_double* ab, lapack_int* ldab, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, float* ab, lapack_int* ldab, float* q,
+                    lapack_int* ldq, float* vl, float* vu, lapack_int* il,
+                    lapack_int* iu, float* abstol, lapack_int* m, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dsbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, double* ab, lapack_int* ldab, double* q,
+                    lapack_int* ldq, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_chbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* q, lapack_int* ldq, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_zhbevx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* kd, lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* q, lapack_int* ldq, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_sstev( char* jobz, lapack_int* n, float* d, float* e, float* z,
+                   lapack_int* ldz, float* work, lapack_int *info );
+void LAPACK_dstev( char* jobz, lapack_int* n, double* d, double* e, double* z,
+                   lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_sstevd( char* jobz, lapack_int* n, float* d, float* e, float* z,
+                    lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_dstevd( char* jobz, lapack_int* n, double* d, double* e, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_sstevx( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dstevx( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_sstevr( char* jobz, char* range, lapack_int* n, float* d, float* e,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, lapack_int* isuppz, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_dstevr( char* jobz, char* range, lapack_int* n, double* d,
+                    double* e, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, lapack_int* isuppz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sgees( char* jobvs, char* sort, LAPACK_S_SELECT2 select,
+                   lapack_int* n, float* a, lapack_int* lda, lapack_int* sdim,
+                   float* wr, float* wi, float* vs, lapack_int* ldvs,
+                   float* work, lapack_int* lwork, lapack_logical* bwork,
+                   lapack_int *info );
+void LAPACK_dgees( char* jobvs, char* sort, LAPACK_D_SELECT2 select,
+                   lapack_int* n, double* a, lapack_int* lda, lapack_int* sdim,
+                   double* wr, double* wi, double* vs, lapack_int* ldvs,
+                   double* work, lapack_int* lwork, lapack_logical* bwork,
+                   lapack_int *info );
+void LAPACK_cgees( char* jobvs, char* sort, LAPACK_C_SELECT1 select,
+                   lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                   lapack_int* sdim, lapack_complex_float* w,
+                   lapack_complex_float* vs, lapack_int* ldvs,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_logical* bwork, lapack_int *info );
+void LAPACK_zgees( char* jobvs, char* sort, LAPACK_Z_SELECT1 select,
+                   lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                   lapack_int* sdim, lapack_complex_double* w,
+                   lapack_complex_double* vs, lapack_int* ldvs,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_logical* bwork, lapack_int *info );
+void LAPACK_sgeesx( char* jobvs, char* sort, LAPACK_S_SELECT2 select,
+                    char* sense, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* sdim, float* wr, float* wi, float* vs,
+                    lapack_int* ldvs, float* rconde, float* rcondv, float* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_logical* bwork, lapack_int *info );
+void LAPACK_dgeesx( char* jobvs, char* sort, LAPACK_D_SELECT2 select,
+                    char* sense, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* sdim, double* wr, double* wi, double* vs,
+                    lapack_int* ldvs, double* rconde, double* rcondv,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_cgeesx( char* jobvs, char* sort, LAPACK_C_SELECT1 select,
+                    char* sense, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* sdim, lapack_complex_float* w,
+                    lapack_complex_float* vs, lapack_int* ldvs, float* rconde,
+                    float* rcondv, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_zgeesx( char* jobvs, char* sort, LAPACK_Z_SELECT1 select,
+                    char* sense, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* sdim, lapack_complex_double* w,
+                    lapack_complex_double* vs, lapack_int* ldvs, double* rconde,
+                    double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_sgeev( char* jobvl, char* jobvr, lapack_int* n, float* a,
+                   lapack_int* lda, float* wr, float* wi, float* vl,
+                   lapack_int* ldvl, float* vr, lapack_int* ldvr, float* work,
+                   lapack_int* lwork, lapack_int *info );
+void LAPACK_dgeev( char* jobvl, char* jobvr, lapack_int* n, double* a,
+                   lapack_int* lda, double* wr, double* wi, double* vl,
+                   lapack_int* ldvl, double* vr, lapack_int* ldvr, double* work,
+                   lapack_int* lwork, lapack_int *info );
+void LAPACK_cgeev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* w, lapack_complex_float* vl,
+                   lapack_int* ldvl, lapack_complex_float* vr, lapack_int* ldvr,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zgeev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* w, lapack_complex_double* vl,
+                   lapack_int* ldvl, lapack_complex_double* vr,
+                   lapack_int* ldvr, lapack_complex_double* work,
+                   lapack_int* lwork, double* rwork, lapack_int *info );
+void LAPACK_sgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, float* a, lapack_int* lda, float* wr,
+                    float* wi, float* vl, lapack_int* ldvl, float* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    float* scale, float* abnrm, float* rconde, float* rcondv,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_dgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, double* a, lapack_int* lda, double* wr,
+                    double* wi, double* vl, lapack_int* ldvl, double* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    double* scale, double* abnrm, double* rconde,
+                    double* rcondv, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_cgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* w, lapack_complex_float* vl,
+                    lapack_int* ldvl, lapack_complex_float* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    float* scale, float* abnrm, float* rconde, float* rcondv,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgeevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* w, lapack_complex_double* vl,
+                    lapack_int* ldvl, lapack_complex_double* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    double* scale, double* abnrm, double* rconde,
+                    double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int *info );
+void LAPACK_sgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    float* a, lapack_int* lda, float* s, float* u,
+                    lapack_int* ldu, float* vt, lapack_int* ldvt, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_dgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    double* a, lapack_int* lda, double* s, double* u,
+                    lapack_int* ldu, double* vt, lapack_int* ldvt, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_cgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* s,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* vt, lapack_int* ldvt,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int *info );
+void LAPACK_zgesvd( char* jobu, char* jobvt, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* s,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* vt, lapack_int* ldvt,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int *info );
+void LAPACK_sgesdd( char* jobz, lapack_int* m, lapack_int* n, float* a,
+                    lapack_int* lda, float* s, float* u, lapack_int* ldu,
+                    float* vt, lapack_int* ldvt, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dgesdd( char* jobz, lapack_int* m, lapack_int* n, double* a,
+                    lapack_int* lda, double* s, double* u, lapack_int* ldu,
+                    double* vt, lapack_int* ldvt, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_cgesdd( char* jobz, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda, float* s,
+                    lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* vt, lapack_int* ldvt,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_zgesdd( char* jobz, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda, double* s,
+                    lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* vt, lapack_int* ldvt,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* iwork, lapack_int *info );
+void LAPACK_dgejsv( char* joba, char* jobu, char* jobv, char* jobr, char* jobt,
+                    char* jobp, lapack_int* m, lapack_int* n, double* a,
+                    lapack_int* lda, double* sva, double* u, lapack_int* ldu,
+                    double* v, lapack_int* ldv, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_sgejsv( char* joba, char* jobu, char* jobv, char* jobr, char* jobt,
+                    char* jobp, lapack_int* m, lapack_int* n, float* a,
+                    lapack_int* lda, float* sva, float* u, lapack_int* ldu,
+                    float* v, lapack_int* ldv, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_dgesvj( char* joba, char* jobu, char* jobv, lapack_int* m,
+                    lapack_int* n, double* a, lapack_int* lda, double* sva,
+                    lapack_int* mv, double* v, lapack_int* ldv, double* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sgesvj( char* joba, char* jobu, char* jobv, lapack_int* m,
+                    lapack_int* n, float* a, lapack_int* lda, float* sva,
+                    lapack_int* mv, float* v, lapack_int* ldv, float* work,
+                    lapack_int* lwork, lapack_int *info );
+void LAPACK_sggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* alpha, float* beta, float* u, lapack_int* ldu,
+                    float* v, lapack_int* ldv, float* q, lapack_int* ldq,
+                    float* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_dggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* alpha, double* beta, double* u, lapack_int* ldu,
+                    double* v, lapack_int* ldv, double* q, lapack_int* ldq,
+                    double* work, lapack_int* iwork, lapack_int *info );
+void LAPACK_cggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* alpha,
+                    float* beta, lapack_complex_float* u, lapack_int* ldu,
+                    lapack_complex_float* v, lapack_int* ldv,
+                    lapack_complex_float* q, lapack_int* ldq,
+                    lapack_complex_float* work, float* rwork, lapack_int* iwork,
+                    lapack_int *info );
+void LAPACK_zggsvd( char* jobu, char* jobv, char* jobq, lapack_int* m,
+                    lapack_int* n, lapack_int* p, lapack_int* k, lapack_int* l,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* alpha,
+                    double* beta, lapack_complex_double* u, lapack_int* ldu,
+                    lapack_complex_double* v, lapack_int* ldv,
+                    lapack_complex_double* q, lapack_int* ldq,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int *info );
+void LAPACK_ssygv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                   float* w, float* work, lapack_int* lwork, lapack_int *info );
+void LAPACK_dsygv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                   double* w, double* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_chegv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb, float* w,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zhegv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb, double* w,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_int *info );
+void LAPACK_ssygvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    float* w, float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dsygvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* w, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_chegvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* w,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhegvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* w,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssygvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* vl, float* vu, lapack_int* il,
+                    lapack_int* iu, float* abstol, lapack_int* m, float* w,
+                    float* z, lapack_int* ldz, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_dsygvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_chegvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, float* vl,
+                    float* vu, lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_zhegvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_sspgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   float* ap, float* bp, float* w, float* z, lapack_int* ldz,
+                   float* work, lapack_int *info );
+void LAPACK_dspgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   double* ap, double* bp, double* w, double* z,
+                   lapack_int* ldz, double* work, lapack_int *info );
+void LAPACK_chpgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_float* ap, lapack_complex_float* bp, float* w,
+                   lapack_complex_float* z, lapack_int* ldz,
+                   lapack_complex_float* work, float* rwork, lapack_int *info );
+void LAPACK_zhpgv( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                   lapack_complex_double* ap, lapack_complex_double* bp,
+                   double* w, lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_sspgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    float* ap, float* bp, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dspgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    double* ap, double* bp, double* w, double* z,
+                    lapack_int* ldz, double* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_int* liwork, lapack_int *info );
+void LAPACK_chpgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_float* ap, lapack_complex_float* bp,
+                    float* w, lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhpgvd( lapack_int* itype, char* jobz, char* uplo, lapack_int* n,
+                    lapack_complex_double* ap, lapack_complex_double* bp,
+                    double* w, lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_sspgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, float* ap, float* bp, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* iwork, lapack_int* ifail,
+                    lapack_int *info );
+void LAPACK_dspgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, double* ap, double* bp, double* vl,
+                    double* vu, lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, double* z, lapack_int* ldz,
+                    double* work, lapack_int* iwork, lapack_int* ifail,
+                    lapack_int *info );
+void LAPACK_chpgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_float* ap,
+                    lapack_complex_float* bp, float* vl, float* vu,
+                    lapack_int* il, lapack_int* iu, float* abstol,
+                    lapack_int* m, float* w, lapack_complex_float* z,
+                    lapack_int* ldz, lapack_complex_float* work, float* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_zhpgvx( lapack_int* itype, char* jobz, char* range, char* uplo,
+                    lapack_int* n, lapack_complex_double* ap,
+                    lapack_complex_double* bp, double* vl, double* vu,
+                    lapack_int* il, lapack_int* iu, double* abstol,
+                    lapack_int* m, double* w, lapack_complex_double* z,
+                    lapack_int* ldz, lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_ssbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, float* ab, lapack_int* ldab, float* bb,
+                   lapack_int* ldbb, float* w, float* z, lapack_int* ldz,
+                   float* work, lapack_int *info );
+void LAPACK_dsbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, double* ab, lapack_int* ldab, double* bb,
+                   lapack_int* ldbb, double* w, double* z, lapack_int* ldz,
+                   double* work, lapack_int *info );
+void LAPACK_chbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, lapack_complex_float* ab, lapack_int* ldab,
+                   lapack_complex_float* bb, lapack_int* ldbb, float* w,
+                   lapack_complex_float* z, lapack_int* ldz,
+                   lapack_complex_float* work, float* rwork, lapack_int *info );
+void LAPACK_zhbgv( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                   lapack_int* kb, lapack_complex_double* ab, lapack_int* ldab,
+                   lapack_complex_double* bb, lapack_int* ldbb, double* w,
+                   lapack_complex_double* z, lapack_int* ldz,
+                   lapack_complex_double* work, double* rwork,
+                   lapack_int *info );
+void LAPACK_ssbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, float* ab, lapack_int* ldab, float* bb,
+                    lapack_int* ldbb, float* w, float* z, lapack_int* ldz,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_dsbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, double* ab, lapack_int* ldab, double* bb,
+                    lapack_int* ldbb, double* w, double* z, lapack_int* ldz,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_chbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_float* ab, lapack_int* ldab,
+                    lapack_complex_float* bb, lapack_int* ldbb, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* lrwork, lapack_int* iwork, lapack_int* liwork,
+                    lapack_int *info );
+void LAPACK_zhbgvd( char* jobz, char* uplo, lapack_int* n, lapack_int* ka,
+                    lapack_int* kb, lapack_complex_double* ab, lapack_int* ldab,
+                    lapack_complex_double* bb, lapack_int* ldbb, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_int *info );
+void LAPACK_ssbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, float* ab, lapack_int* ldab,
+                    float* bb, lapack_int* ldbb, float* q, lapack_int* ldq,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w, float* z,
+                    lapack_int* ldz, float* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_dsbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, double* ab,
+                    lapack_int* ldab, double* bb, lapack_int* ldbb, double* q,
+                    lapack_int* ldq, double* vl, double* vu, lapack_int* il,
+                    lapack_int* iu, double* abstol, lapack_int* m, double* w,
+                    double* z, lapack_int* ldz, double* work, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_chbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, lapack_complex_float* ab,
+                    lapack_int* ldab, lapack_complex_float* bb,
+                    lapack_int* ldbb, lapack_complex_float* q, lapack_int* ldq,
+                    float* vl, float* vu, lapack_int* il, lapack_int* iu,
+                    float* abstol, lapack_int* m, float* w,
+                    lapack_complex_float* z, lapack_int* ldz,
+                    lapack_complex_float* work, float* rwork, lapack_int* iwork,
+                    lapack_int* ifail, lapack_int *info );
+void LAPACK_zhbgvx( char* jobz, char* range, char* uplo, lapack_int* n,
+                    lapack_int* ka, lapack_int* kb, lapack_complex_double* ab,
+                    lapack_int* ldab, lapack_complex_double* bb,
+                    lapack_int* ldbb, lapack_complex_double* q, lapack_int* ldq,
+                    double* vl, double* vu, lapack_int* il, lapack_int* iu,
+                    double* abstol, lapack_int* m, double* w,
+                    lapack_complex_double* z, lapack_int* ldz,
+                    lapack_complex_double* work, double* rwork,
+                    lapack_int* iwork, lapack_int* ifail, lapack_int *info );
+void LAPACK_sgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_S_SELECT3 selctg, lapack_int* n, float* a,
+                   lapack_int* lda, float* b, lapack_int* ldb, lapack_int* sdim,
+                   float* alphar, float* alphai, float* beta, float* vsl,
+                   lapack_int* ldvsl, float* vsr, lapack_int* ldvsr,
+                   float* work, lapack_int* lwork, lapack_logical* bwork,
+                   lapack_int *info );
+void LAPACK_dgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_D_SELECT3 selctg, lapack_int* n, double* a,
+                   lapack_int* lda, double* b, lapack_int* ldb,
+                   lapack_int* sdim, double* alphar, double* alphai,
+                   double* beta, double* vsl, lapack_int* ldvsl, double* vsr,
+                   lapack_int* ldvsr, double* work, lapack_int* lwork,
+                   lapack_logical* bwork, lapack_int *info );
+void LAPACK_cgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_C_SELECT2 selctg, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb, lapack_int* sdim,
+                   lapack_complex_float* alpha, lapack_complex_float* beta,
+                   lapack_complex_float* vsl, lapack_int* ldvsl,
+                   lapack_complex_float* vsr, lapack_int* ldvsr,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_logical* bwork, lapack_int *info );
+void LAPACK_zgges( char* jobvsl, char* jobvsr, char* sort,
+                   LAPACK_Z_SELECT2 selctg, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb, lapack_int* sdim,
+                   lapack_complex_double* alpha, lapack_complex_double* beta,
+                   lapack_complex_double* vsl, lapack_int* ldvsl,
+                   lapack_complex_double* vsr, lapack_int* ldvsr,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_logical* bwork, lapack_int *info );
+void LAPACK_sggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_S_SELECT3 selctg, char* sense, lapack_int* n,
+                    float* a, lapack_int* lda, float* b, lapack_int* ldb,
+                    lapack_int* sdim, float* alphar, float* alphai, float* beta,
+                    float* vsl, lapack_int* ldvsl, float* vsr,
+                    lapack_int* ldvsr, float* rconde, float* rcondv,
+                    float* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_dggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_D_SELECT3 selctg, char* sense, lapack_int* n,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    lapack_int* sdim, double* alphar, double* alphai,
+                    double* beta, double* vsl, lapack_int* ldvsl, double* vsr,
+                    lapack_int* ldvsr, double* rconde, double* rcondv,
+                    double* work, lapack_int* lwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_cggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_C_SELECT2 selctg, char* sense, lapack_int* n,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb, lapack_int* sdim,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* vsl, lapack_int* ldvsl,
+                    lapack_complex_float* vsr, lapack_int* ldvsr, float* rconde,
+                    float* rcondv, lapack_complex_float* work,
+                    lapack_int* lwork, float* rwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_zggesx( char* jobvsl, char* jobvsr, char* sort,
+                    LAPACK_Z_SELECT2 selctg, char* sense, lapack_int* n,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb, lapack_int* sdim,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* vsl, lapack_int* ldvsl,
+                    lapack_complex_double* vsr, lapack_int* ldvsr,
+                    double* rconde, double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_int* liwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_sggev( char* jobvl, char* jobvr, lapack_int* n, float* a,
+                   lapack_int* lda, float* b, lapack_int* ldb, float* alphar,
+                   float* alphai, float* beta, float* vl, lapack_int* ldvl,
+                   float* vr, lapack_int* ldvr, float* work, lapack_int* lwork,
+                   lapack_int *info );
+void LAPACK_dggev( char* jobvl, char* jobvr, lapack_int* n, double* a,
+                   lapack_int* lda, double* b, lapack_int* ldb, double* alphar,
+                   double* alphai, double* beta, double* vl, lapack_int* ldvl,
+                   double* vr, lapack_int* ldvr, double* work,
+                   lapack_int* lwork, lapack_int *info );
+void LAPACK_cggev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_float* a, lapack_int* lda,
+                   lapack_complex_float* b, lapack_int* ldb,
+                   lapack_complex_float* alpha, lapack_complex_float* beta,
+                   lapack_complex_float* vl, lapack_int* ldvl,
+                   lapack_complex_float* vr, lapack_int* ldvr,
+                   lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                   lapack_int *info );
+void LAPACK_zggev( char* jobvl, char* jobvr, lapack_int* n,
+                   lapack_complex_double* a, lapack_int* lda,
+                   lapack_complex_double* b, lapack_int* ldb,
+                   lapack_complex_double* alpha, lapack_complex_double* beta,
+                   lapack_complex_double* vl, lapack_int* ldvl,
+                   lapack_complex_double* vr, lapack_int* ldvr,
+                   lapack_complex_double* work, lapack_int* lwork,
+                   double* rwork, lapack_int *info );
+void LAPACK_sggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, float* alphar, float* alphai, float* beta,
+                    float* vl, lapack_int* ldvl, float* vr, lapack_int* ldvr,
+                    lapack_int* ilo, lapack_int* ihi, float* lscale,
+                    float* rscale, float* abnrm, float* bbnrm, float* rconde,
+                    float* rcondv, float* work, lapack_int* lwork,
+                    lapack_int* iwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_dggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, double* alphar, double* alphai,
+                    double* beta, double* vl, lapack_int* ldvl, double* vr,
+                    lapack_int* ldvr, lapack_int* ilo, lapack_int* ihi,
+                    double* lscale, double* rscale, double* abnrm,
+                    double* bbnrm, double* rconde, double* rcondv, double* work,
+                    lapack_int* lwork, lapack_int* iwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_cggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* vl, lapack_int* ldvl,
+                    lapack_complex_float* vr, lapack_int* ldvr, lapack_int* ilo,
+                    lapack_int* ihi, float* lscale, float* rscale, float* abnrm,
+                    float* bbnrm, float* rconde, float* rcondv,
+                    lapack_complex_float* work, lapack_int* lwork, float* rwork,
+                    lapack_int* iwork, lapack_logical* bwork,
+                    lapack_int *info );
+void LAPACK_zggevx( char* balanc, char* jobvl, char* jobvr, char* sense,
+                    lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* vl, lapack_int* ldvl,
+                    lapack_complex_double* vr, lapack_int* ldvr,
+                    lapack_int* ilo, lapack_int* ihi, double* lscale,
+                    double* rscale, double* abnrm, double* bbnrm,
+                    double* rconde, double* rcondv, lapack_complex_double* work,
+                    lapack_int* lwork, double* rwork, lapack_int* iwork,
+                    lapack_logical* bwork, lapack_int *info );
+void LAPACK_dsfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, double* alpha, const double* a,
+                   lapack_int* lda, double* beta, double* c );
+void LAPACK_ssfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, float* alpha, const float* a, lapack_int* lda,
+                   float* beta, float* c );
+void LAPACK_zhfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, double* alpha, const lapack_complex_double* a,
+                   lapack_int* lda, double* beta, lapack_complex_double* c );
+void LAPACK_chfrk( char* transr, char* uplo, char* trans, lapack_int* n,
+                   lapack_int* k, float* alpha, const lapack_complex_float* a,
+                   lapack_int* lda, float* beta, lapack_complex_float* c );
+void LAPACK_dtfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n, double* alpha,
+                   const double* a, double* b, lapack_int* ldb );
+void LAPACK_stfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n, float* alpha,
+                   const float* a, float* b, lapack_int* ldb );
+void LAPACK_ztfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n,
+                   lapack_complex_double* alpha, const lapack_complex_double* a,
+                   lapack_complex_double* b, lapack_int* ldb );
+void LAPACK_ctfsm( char* transr, char* side, char* uplo, char* trans,
+                   char* diag, lapack_int* m, lapack_int* n,
+                   lapack_complex_float* alpha, const lapack_complex_float* a,
+                   lapack_complex_float* b, lapack_int* ldb );
+void LAPACK_dtfttp( char* transr, char* uplo, lapack_int* n, const double* arf,
+                    double* ap, lapack_int *info );
+void LAPACK_stfttp( char* transr, char* uplo, lapack_int* n, const float* arf,
+                    float* ap, lapack_int *info );
+void LAPACK_ztfttp( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* arf, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_ctfttp( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* arf, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_dtfttr( char* transr, char* uplo, lapack_int* n, const double* arf,
+                    double* a, lapack_int* lda, lapack_int *info );
+void LAPACK_stfttr( char* transr, char* uplo, lapack_int* n, const float* arf,
+                    float* a, lapack_int* lda, lapack_int *info );
+void LAPACK_ztfttr( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* arf, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_ctfttr( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* arf, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_dtpttf( char* transr, char* uplo, lapack_int* n, const double* ap,
+                    double* arf, lapack_int *info );
+void LAPACK_stpttf( char* transr, char* uplo, lapack_int* n, const float* ap,
+                    float* arf, lapack_int *info );
+void LAPACK_ztpttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* ap, lapack_complex_double* arf,
+                    lapack_int *info );
+void LAPACK_ctpttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* ap, lapack_complex_float* arf,
+                    lapack_int *info );
+void LAPACK_dtpttr( char* uplo, lapack_int* n, const double* ap, double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_stpttr( char* uplo, lapack_int* n, const float* ap, float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_ztpttr( char* uplo, lapack_int* n, const lapack_complex_double* ap,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_ctpttr( char* uplo, lapack_int* n, const lapack_complex_float* ap,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dtrttf( char* transr, char* uplo, lapack_int* n, const double* a,
+                    lapack_int* lda, double* arf, lapack_int *info );
+void LAPACK_strttf( char* transr, char* uplo, lapack_int* n, const float* a,
+                    lapack_int* lda, float* arf, lapack_int *info );
+void LAPACK_ztrttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* arf, lapack_int *info );
+void LAPACK_ctrttf( char* transr, char* uplo, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* arf, lapack_int *info );
+void LAPACK_dtrttp( char* uplo, lapack_int* n, const double* a, lapack_int* lda,
+                    double* ap, lapack_int *info );
+void LAPACK_strttp( char* uplo, lapack_int* n, const float* a, lapack_int* lda,
+                    float* ap, lapack_int *info );
+void LAPACK_ztrttp( char* uplo, lapack_int* n, const lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* ap,
+                    lapack_int *info );
+void LAPACK_ctrttp( char* uplo, lapack_int* n, const lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* ap,
+                    lapack_int *info );
+void LAPACK_sgeqrfp( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* tau, float* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_dgeqrfp( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* tau, double* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_cgeqrfp( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* tau,
+                     lapack_complex_float* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_zgeqrfp( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* tau,
+                     lapack_complex_double* work, lapack_int* lwork,
+                     lapack_int *info );
+void LAPACK_clacgv( lapack_int* n, lapack_complex_float* x, lapack_int* incx );
+void LAPACK_zlacgv( lapack_int* n, lapack_complex_double* x, lapack_int* incx );
+void LAPACK_slarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    float* x );
+void LAPACK_dlarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    double* x );
+void LAPACK_clarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    lapack_complex_float* x );
+void LAPACK_zlarnv( lapack_int* idist, lapack_int* iseed, lapack_int* n,
+                    lapack_complex_double* x );
+void LAPACK_sgeqr2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int *info );
+void LAPACK_dgeqr2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int *info );
+void LAPACK_cgeqr2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgeqr2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slacpy( char* uplo, lapack_int* m, lapack_int* n, const float* a,
+                    lapack_int* lda, float* b, lapack_int* ldb );
+void LAPACK_dlacpy( char* uplo, lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, double* b, lapack_int* ldb );
+void LAPACK_clacpy( char* uplo, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb );
+void LAPACK_zlacpy( char* uplo, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb );
+void LAPACK_sgetf2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_dgetf2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int* ipiv, lapack_int *info );
+void LAPACK_cgetf2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_zgetf2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* ipiv, lapack_int *info );
+void LAPACK_slaswp( lapack_int* n, float* a, lapack_int* lda, lapack_int* k1,
+                    lapack_int* k2, const lapack_int* ipiv, lapack_int* incx );
+void LAPACK_dlaswp( lapack_int* n, double* a, lapack_int* lda, lapack_int* k1,
+                    lapack_int* k2, const lapack_int* ipiv, lapack_int* incx );
+void LAPACK_claswp( lapack_int* n, lapack_complex_float* a, lapack_int* lda,
+                    lapack_int* k1, lapack_int* k2, const lapack_int* ipiv,
+                    lapack_int* incx );
+void LAPACK_zlaswp( lapack_int* n, lapack_complex_double* a, lapack_int* lda,
+                    lapack_int* k1, lapack_int* k2, const lapack_int* ipiv,
+                    lapack_int* incx );
+float LAPACK_slange( char* norm, lapack_int* m, lapack_int* n, const float* a,
+                    lapack_int* lda, float* work );
+double LAPACK_dlange( char* norm, lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, double* work );
+float LAPACK_clange( char* norm, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda, float* work );
+double LAPACK_zlange( char* norm, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* work );
+float LAPACK_clanhe( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda, float* work );
+double LAPACK_zlanhe( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* work );
+float LAPACK_slansy( char* norm, char* uplo, lapack_int* n, const float* a,
+                    lapack_int* lda, float* work );
+double LAPACK_dlansy( char* norm, char* uplo, lapack_int* n, const double* a,
+                    lapack_int* lda, double* work );
+float LAPACK_clansy( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_float* a, lapack_int* lda, float* work );
+double LAPACK_zlansy( char* norm, char* uplo, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda, double* work );
+float LAPACK_slantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const float* a, lapack_int* lda, float* work );
+double LAPACK_dlantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const double* a, lapack_int* lda, double* work );
+float LAPACK_clantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const lapack_complex_float* a, lapack_int* lda,
+                    float* work );
+double LAPACK_zlantr( char* norm, char* uplo, char* diag, lapack_int* m,
+                    lapack_int* n, const lapack_complex_double* a, lapack_int* lda,
+                    double* work );
+float LAPACK_slamch( char* cmach );
+double LAPACK_dlamch( char* cmach );
+void LAPACK_sgelq2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                    float* tau, float* work, lapack_int *info );
+void LAPACK_dgelq2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                    double* tau, double* work, lapack_int *info );
+void LAPACK_cgelq2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_complex_float* tau,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgelq2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_complex_double* tau,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, const float* v,
+                    lapack_int* ldv, const float* t, lapack_int* ldt, float* c,
+                    lapack_int* ldc, float* work, lapack_int* ldwork );
+void LAPACK_dlarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k,
+                    const double* v, lapack_int* ldv, const double* t,
+                    lapack_int* ldt, double* c, lapack_int* ldc, double* work,
+                    lapack_int* ldwork );
+void LAPACK_clarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k,
+                    const lapack_complex_float* v, lapack_int* ldv,
+                    const lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work, lapack_int* ldwork );
+void LAPACK_zlarfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k,
+                    const lapack_complex_double* v, lapack_int* ldv,
+                    const lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work, lapack_int* ldwork );
+void LAPACK_slarfg( lapack_int* n, float* alpha, float* x, lapack_int* incx,
+                    float* tau );
+void LAPACK_dlarfg( lapack_int* n, double* alpha, double* x, lapack_int* incx,
+                    double* tau );
+void LAPACK_clarfg( lapack_int* n, lapack_complex_float* alpha,
+                    lapack_complex_float* x, lapack_int* incx,
+                    lapack_complex_float* tau );
+void LAPACK_zlarfg( lapack_int* n, lapack_complex_double* alpha,
+                    lapack_complex_double* x, lapack_int* incx,
+                    lapack_complex_double* tau );
+void LAPACK_slarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const float* v, lapack_int* ldv, const float* tau, float* t,
+                    lapack_int* ldt );
+void LAPACK_dlarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const double* v, lapack_int* ldv, const double* tau,
+                    double* t, lapack_int* ldt );
+void LAPACK_clarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const lapack_complex_float* v, lapack_int* ldv,
+                    const lapack_complex_float* tau, lapack_complex_float* t,
+                    lapack_int* ldt );
+void LAPACK_zlarft( char* direct, char* storev, lapack_int* n, lapack_int* k,
+                    const lapack_complex_double* v, lapack_int* ldv,
+                    const lapack_complex_double* tau, lapack_complex_double* t,
+                    lapack_int* ldt );
+void LAPACK_slarfx( char* side, lapack_int* m, lapack_int* n, const float* v,
+                    float* tau, float* c, lapack_int* ldc, float* work );
+void LAPACK_dlarfx( char* side, lapack_int* m, lapack_int* n, const double* v,
+                    double* tau, double* c, lapack_int* ldc, double* work );
+void LAPACK_clarfx( char* side, lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* v, lapack_complex_float* tau,
+                    lapack_complex_float* c, lapack_int* ldc,
+                    lapack_complex_float* work );
+void LAPACK_zlarfx( char* side, lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* v, lapack_complex_double* tau,
+                    lapack_complex_double* c, lapack_int* ldc,
+                    lapack_complex_double* work );
+void LAPACK_slatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, float* d, lapack_int* mode, float* cond,
+                    float* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    float* a, lapack_int* lda, float* work, lapack_int *info );
+void LAPACK_dlatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, double* d, lapack_int* mode, double* cond,
+                    double* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    double* a, lapack_int* lda, double* work,
+                    lapack_int *info );
+void LAPACK_clatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, float* d, lapack_int* mode, float* cond,
+                    float* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlatms( lapack_int* m, lapack_int* n, char* dist, lapack_int* iseed,
+                    char* sym, double* d, lapack_int* mode, double* cond,
+                    double* dmax, lapack_int* kl, lapack_int* ku, char* pack,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slag2d( lapack_int* m, lapack_int* n, const float* sa,
+                    lapack_int* ldsa, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dlag2s( lapack_int* m, lapack_int* n, const double* a,
+                    lapack_int* lda, float* sa, lapack_int* ldsa,
+                    lapack_int *info );
+void LAPACK_clag2z( lapack_int* m, lapack_int* n,
+                    const lapack_complex_float* sa, lapack_int* ldsa,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_zlag2c( lapack_int* m, lapack_int* n,
+                    const lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_float* sa, lapack_int* ldsa,
+                    lapack_int *info );
+void LAPACK_slauum( char* uplo, lapack_int* n, float* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_dlauum( char* uplo, lapack_int* n, double* a, lapack_int* lda,
+                    lapack_int *info );
+void LAPACK_clauum( char* uplo, lapack_int* n, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_zlauum( char* uplo, lapack_int* n, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int *info );
+void LAPACK_slagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const float* d, float* a, lapack_int* lda,
+                    lapack_int* iseed, float* work, lapack_int *info );
+void LAPACK_dlagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const double* d, double* a, lapack_int* lda,
+                    lapack_int* iseed, double* work, lapack_int *info );
+void LAPACK_clagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const float* d, lapack_complex_float* a,
+                    lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlagge( lapack_int* m, lapack_int* n, lapack_int* kl,
+                    lapack_int* ku, const double* d, lapack_complex_double* a,
+                    lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_slaset( char* uplo, lapack_int* m, lapack_int* n, float* alpha,
+                    float* beta, float* a, lapack_int* lda );
+void LAPACK_dlaset( char* uplo, lapack_int* m, lapack_int* n, double* alpha,
+                    double* beta, double* a, lapack_int* lda );
+void LAPACK_claset( char* uplo, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* alpha, lapack_complex_float* beta,
+                    lapack_complex_float* a, lapack_int* lda );
+void LAPACK_zlaset( char* uplo, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* alpha, lapack_complex_double* beta,
+                    lapack_complex_double* a, lapack_int* lda );
+void LAPACK_slasrt( char* id, lapack_int* n, float* d, lapack_int *info );
+void LAPACK_dlasrt( char* id, lapack_int* n, double* d, lapack_int *info );
+void LAPACK_claghe( lapack_int* n, lapack_int* k, const float* d,
+                    lapack_complex_float* a, lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlaghe( lapack_int* n, lapack_int* k, const double* d,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int* iseed, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_slagsy( lapack_int* n, lapack_int* k, const float* d, float* a,
+                    lapack_int* lda, lapack_int* iseed, float* work,
+                    lapack_int *info );
+void LAPACK_dlagsy( lapack_int* n, lapack_int* k, const double* d, double* a,
+                    lapack_int* lda, lapack_int* iseed, double* work,
+                    lapack_int *info );
+void LAPACK_clagsy( lapack_int* n, lapack_int* k, const float* d,
+                    lapack_complex_float* a, lapack_int* lda, lapack_int* iseed,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zlagsy( lapack_int* n, lapack_int* k, const double* d,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_int* iseed, lapack_complex_double* work,
+                    lapack_int *info );
+void LAPACK_slapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    float* x, lapack_int* ldx, lapack_int* k );
+void LAPACK_dlapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    double* x, lapack_int* ldx, lapack_int* k );
+void LAPACK_clapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    lapack_complex_float* x, lapack_int* ldx, lapack_int* k );
+void LAPACK_zlapmr( lapack_logical* forwrd, lapack_int* m, lapack_int* n,
+                    lapack_complex_double* x, lapack_int* ldx, lapack_int* k );
+float LAPACK_slapy2( float* x, float* y );
+double LAPACK_dlapy2( double* x, double* y );
+float LAPACK_slapy3( float* x, float* y, float* z );
+double LAPACK_dlapy3( double* x, double* y, double* z );
+void LAPACK_slartgp( float* f, float* g, float* cs, float* sn, float* r );
+void LAPACK_dlartgp( double* f, double* g, double* cs, double* sn, double* r );
+void LAPACK_slartgs( float* x, float* y, float* sigma, float* cs, float* sn );
+void LAPACK_dlartgs( double* x, double* y, double* sigma, double* cs,
+                     double* sn );
+// LAPACK 3.3.0
+void LAPACK_cbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    float* theta, float* phi,
+                    lapack_complex_float* u1, lapack_int* ldu1,
+                    lapack_complex_float* u2, lapack_int* ldu2,
+                    lapack_complex_float* v1t, lapack_int* ldv1t,
+                    lapack_complex_float* v2t, lapack_int* ldv2t,
+                    float* b11d, float* b11e, float* b12d,
+                    float* b12e, float* b21d, float* b21e,
+                    float* b22d, float* b22e, float* rwork,
+                    lapack_int* lrwork , lapack_int *info );
+void LAPACK_cheswapr( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_chetri2( char* uplo, lapack_int* n,
+                     lapack_complex_float* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_chetri2x( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_float* work, lapack_int* nb , lapack_int *info );
+void LAPACK_chetrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const lapack_complex_float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* work , lapack_int *info );
+void LAPACK_csyconv( char* uplo, char* way,
+                     lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_float* work , lapack_int *info );
+void LAPACK_csyswapr( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_csytri2( char* uplo, lapack_int* n,
+                     lapack_complex_float* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_csytri2x( char* uplo, lapack_int* n,
+                      lapack_complex_float* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_float* work, lapack_int* nb , lapack_int *info );
+void LAPACK_csytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const lapack_complex_float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* work , lapack_int *info );
+void LAPACK_cunbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    lapack_complex_float* x11, lapack_int* ldx11,
+                    lapack_complex_float* x12, lapack_int* ldx12,
+                    lapack_complex_float* x21, lapack_int* ldx21,
+                    lapack_complex_float* x22, lapack_int* ldx22,
+                    float* theta, float* phi,
+                    lapack_complex_float* taup1,
+                    lapack_complex_float* taup2,
+                    lapack_complex_float* tauq1,
+                    lapack_complex_float* tauq2,
+                    lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_cuncsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, lapack_complex_float* x11,
+                    lapack_int* ldx11, lapack_complex_float* x12,
+                    lapack_int* ldx12, lapack_complex_float* x21,
+                    lapack_int* ldx21, lapack_complex_float* x22,
+                    lapack_int* ldx22, float* theta,
+                    lapack_complex_float* u1, lapack_int* ldu1,
+                    lapack_complex_float* u2, lapack_int* ldu2,
+                    lapack_complex_float* v1t, lapack_int* ldv1t,
+                    lapack_complex_float* v2t, lapack_int* ldv2t,
+                    lapack_complex_float* work, lapack_int* lwork,
+                    float* rwork, lapack_int* lrwork,
+                    lapack_int* iwork , lapack_int *info );
+void LAPACK_dbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    double* theta, double* phi, double* u1,
+                    lapack_int* ldu1, double* u2, lapack_int* ldu2,
+                    double* v1t, lapack_int* ldv1t, double* v2t,
+                    lapack_int* ldv2t, double* b11d, double* b11e,
+                    double* b12d, double* b12e, double* b21d,
+                    double* b21e, double* b22d, double* b22e,
+                    double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_dorbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    double* x11, lapack_int* ldx11, double* x12,
+                    lapack_int* ldx12, double* x21, lapack_int* ldx21,
+                    double* x22, lapack_int* ldx22, double* theta,
+                    double* phi, double* taup1, double* taup2,
+                    double* tauq1, double* tauq2, double* work,
+                    lapack_int* lwork , lapack_int *info );
+void LAPACK_dorcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, double* x11, lapack_int* ldx11,
+                    double* x12, lapack_int* ldx12, double* x21,
+                    lapack_int* ldx21, double* x22, lapack_int* ldx22,
+                    double* theta, double* u1, lapack_int* ldu1,
+                    double* u2, lapack_int* ldu2, double* v1t,
+                    lapack_int* ldv1t, double* v2t, lapack_int* ldv2t,
+                    double* work, lapack_int* lwork,
+                    lapack_int* iwork , lapack_int *info );
+void LAPACK_dsyconv( char* uplo, char* way,
+                     lapack_int* n, double* a, lapack_int* lda,
+                     const lapack_int* ipiv, double* work , lapack_int *info );
+void LAPACK_dsyswapr( char* uplo, lapack_int* n,
+                      double* a, lapack_int* i1, lapack_int* i2 );
+void LAPACK_dsytri2( char* uplo, lapack_int* n,
+                     double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_dsytri2x( char* uplo, lapack_int* n,
+                      double* a, lapack_int* lda,
+                      const lapack_int* ipiv, double* work,
+                      lapack_int* nb , lapack_int *info );
+void LAPACK_dsytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const double* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     double* b, lapack_int* ldb, double* work , lapack_int *info );
+void LAPACK_sbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    float* theta, float* phi, float* u1,
+                    lapack_int* ldu1, float* u2, lapack_int* ldu2,
+                    float* v1t, lapack_int* ldv1t, float* v2t,
+                    lapack_int* ldv2t, float* b11d, float* b11e,
+                    float* b12d, float* b12e, float* b21d,
+                    float* b21e, float* b22d, float* b22e,
+                    float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_sorbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    float* x11, lapack_int* ldx11, float* x12,
+                    lapack_int* ldx12, float* x21, lapack_int* ldx21,
+                    float* x22, lapack_int* ldx22, float* theta,
+                    float* phi, float* taup1, float* taup2,
+                    float* tauq1, float* tauq2, float* work,
+                    lapack_int* lwork , lapack_int *info );
+void LAPACK_sorcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, float* x11, lapack_int* ldx11,
+                    float* x12, lapack_int* ldx12, float* x21,
+                    lapack_int* ldx21, float* x22, lapack_int* ldx22,
+                    float* theta, float* u1, lapack_int* ldu1,
+                    float* u2, lapack_int* ldu2, float* v1t,
+                    lapack_int* ldv1t, float* v2t, lapack_int* ldv2t,
+                    float* work, lapack_int* lwork,
+                    lapack_int* iwork , lapack_int *info );
+void LAPACK_ssyconv( char* uplo, char* way,
+                     lapack_int* n, float* a, lapack_int* lda,
+                     const lapack_int* ipiv, float* work , lapack_int *info );
+void LAPACK_ssyswapr( char* uplo, lapack_int* n,
+                      float* a, lapack_int* i1, lapack_int* i2 );
+void LAPACK_ssytri2( char* uplo, lapack_int* n,
+                     float* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_float* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_ssytri2x( char* uplo, lapack_int* n,
+                      float* a, lapack_int* lda,
+                      const lapack_int* ipiv, float* work,
+                      lapack_int* nb , lapack_int *info );
+void LAPACK_ssytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs, const float* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     float* b, lapack_int* ldb, float* work , lapack_int *info );
+void LAPACK_zbbcsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    double* theta, double* phi,
+                    lapack_complex_double* u1, lapack_int* ldu1,
+                    lapack_complex_double* u2, lapack_int* ldu2,
+                    lapack_complex_double* v1t, lapack_int* ldv1t,
+                    lapack_complex_double* v2t, lapack_int* ldv2t,
+                    double* b11d, double* b11e, double* b12d,
+                    double* b12e, double* b21d, double* b21e,
+                    double* b22d, double* b22e, double* rwork,
+                    lapack_int* lrwork , lapack_int *info );
+void LAPACK_zheswapr( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_zhetri2( char* uplo, lapack_int* n,
+                     lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_zhetri2x( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_double* work, lapack_int* nb , lapack_int *info );
+void LAPACK_zhetrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* work , lapack_int *info );
+void LAPACK_zsyconv( char* uplo, char* way,
+                     lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, const lapack_int* ipiv,
+                     lapack_complex_double* work , lapack_int *info );
+void LAPACK_zsyswapr( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* i1,
+                      lapack_int* i2 );
+void LAPACK_zsytri2( char* uplo, lapack_int* n,
+                     lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_zsytri2x( char* uplo, lapack_int* n,
+                      lapack_complex_double* a, lapack_int* lda,
+                      const lapack_int* ipiv,
+                      lapack_complex_double* work, lapack_int* nb , lapack_int *info );
+void LAPACK_zsytrs2( char* uplo, lapack_int* n,
+                     lapack_int* nrhs,
+                     const lapack_complex_double* a, lapack_int* lda,
+                     const lapack_int* ipiv,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* work , lapack_int *info );
+void LAPACK_zunbdb( char* trans, char* signs,
+                    lapack_int* m, lapack_int* p, lapack_int* q,
+                    lapack_complex_double* x11, lapack_int* ldx11,
+                    lapack_complex_double* x12, lapack_int* ldx12,
+                    lapack_complex_double* x21, lapack_int* ldx21,
+                    lapack_complex_double* x22, lapack_int* ldx22,
+                    double* theta, double* phi,
+                    lapack_complex_double* taup1,
+                    lapack_complex_double* taup2,
+                    lapack_complex_double* tauq1,
+                    lapack_complex_double* tauq2,
+                    lapack_complex_double* work, lapack_int* lwork , lapack_int *info );
+void LAPACK_zuncsd( char* jobu1, char* jobu2,
+                    char* jobv1t, char* jobv2t, char* trans,
+                    char* signs, lapack_int* m, lapack_int* p,
+                    lapack_int* q, lapack_complex_double* x11,
+                    lapack_int* ldx11, lapack_complex_double* x12,
+                    lapack_int* ldx12, lapack_complex_double* x21,
+                    lapack_int* ldx21, lapack_complex_double* x22,
+                    lapack_int* ldx22, double* theta,
+                    lapack_complex_double* u1, lapack_int* ldu1,
+                    lapack_complex_double* u2, lapack_int* ldu2,
+                    lapack_complex_double* v1t, lapack_int* ldv1t,
+                    lapack_complex_double* v2t, lapack_int* ldv2t,
+                    lapack_complex_double* work, lapack_int* lwork,
+                    double* rwork, lapack_int* lrwork,
+                    lapack_int* iwork , lapack_int *info );
+// LAPACK 3.4.0
+void LAPACK_sgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb, const float* v,
+                     lapack_int* ldv, const float* t, lapack_int* ldt, float* c,
+                     lapack_int* ldc, float* work, lapack_int *info );
+void LAPACK_dgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb, const double* v,
+                     lapack_int* ldv, const double* t, lapack_int* ldt,
+                     double* c, lapack_int* ldc, double* work,
+                     lapack_int *info );
+void LAPACK_cgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb,
+                     const lapack_complex_float* v, lapack_int* ldv,
+                     const lapack_complex_float* t, lapack_int* ldt,
+                     lapack_complex_float* c, lapack_int* ldc,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgemqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* nb,
+                     const lapack_complex_double* v, lapack_int* ldv,
+                     const lapack_complex_double* t, lapack_int* ldt,
+                     lapack_complex_double* c, lapack_int* ldc,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_sgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb, float* a,
+                    lapack_int* lda, float* t, lapack_int* ldt, float* work,
+                    lapack_int *info );
+void LAPACK_dgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb, double* a,
+                    lapack_int* lda, double* t, lapack_int* ldt, double* work,
+                    lapack_int *info );
+void LAPACK_cgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_zgeqrt( lapack_int* m, lapack_int* n, lapack_int* nb,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_sgeqrt2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_dgeqrt2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_cgeqrt2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_zgeqrt2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_sgeqrt3( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_dgeqrt3( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* t, lapack_int* ldt, lapack_int *info );
+void LAPACK_cgeqrt3( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_zgeqrt3( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_stpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const float* v, lapack_int* ldv, const float* t,
+                     lapack_int* ldt, float* a, lapack_int* lda, float* b,
+                     lapack_int* ldb, float* work, lapack_int *info );
+void LAPACK_dtpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const double* v, lapack_int* ldv, const double* t,
+                     lapack_int* ldt, double* a, lapack_int* lda, double* b,
+                     lapack_int* ldb, double* work, lapack_int *info );
+void LAPACK_ctpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const lapack_complex_float* v, lapack_int* ldv,
+                     const lapack_complex_float* t, lapack_int* ldt,
+                     lapack_complex_float* a, lapack_int* lda,
+                     lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* work, lapack_int *info );
+void LAPACK_ztpmqrt( char* side, char* trans, lapack_int* m, lapack_int* n,
+                     lapack_int* k, lapack_int* l, lapack_int* nb,
+                     const lapack_complex_double* v, lapack_int* ldv,
+                     const lapack_complex_double* t, lapack_int* ldt,
+                     lapack_complex_double* a, lapack_int* lda,
+                     lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* work, lapack_int *info );
+void LAPACK_dtpqrt( lapack_int* m, lapack_int* n, lapack_int* l, lapack_int* nb,
+                    double* a, lapack_int* lda, double* b, lapack_int* ldb,
+                    double* t, lapack_int* ldt, double* work,
+                    lapack_int *info );
+void LAPACK_ctpqrt( lapack_int* m, lapack_int* n, lapack_int* l, lapack_int* nb,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* t, lapack_complex_float* b,
+                    lapack_int* ldb, lapack_int* ldt,
+                    lapack_complex_float* work, lapack_int *info );
+void LAPACK_ztpqrt( lapack_int* m, lapack_int* n, lapack_int* l, lapack_int* nb,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* work, lapack_int *info );
+void LAPACK_stpqrt2( lapack_int* m, lapack_int* n, float* a, lapack_int* lda,
+                     float* b, lapack_int* ldb, float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_dtpqrt2( lapack_int* m, lapack_int* n, double* a, lapack_int* lda,
+                     double* b, lapack_int* ldb, double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_ctpqrt2( lapack_int* m, lapack_int* n, lapack_complex_float* a,
+                     lapack_int* lda, lapack_complex_float* b, lapack_int* ldb,
+                     lapack_complex_float* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_ztpqrt2( lapack_int* m, lapack_int* n, lapack_complex_double* a,
+                     lapack_int* lda, lapack_complex_double* b, lapack_int* ldb,
+                     lapack_complex_double* t, lapack_int* ldt,
+                     lapack_int *info );
+void LAPACK_stprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const float* v, lapack_int* ldv, const float* t,
+                    lapack_int* ldt, float* a, lapack_int* lda, float* b,
+                    lapack_int* ldb, const float* mywork,
+                    lapack_int* myldwork );
+void LAPACK_dtprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const double* v, lapack_int* ldv, const double* t,
+                    lapack_int* ldt, double* a, lapack_int* lda, double* b,
+                    lapack_int* ldb, const double* mywork,
+                    lapack_int* myldwork );
+void LAPACK_ctprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const lapack_complex_float* v, lapack_int* ldv,
+                    const lapack_complex_float* t, lapack_int* ldt,
+                    lapack_complex_float* a, lapack_int* lda,
+                    lapack_complex_float* b, lapack_int* ldb,
+                    const float* mywork, lapack_int* myldwork );
+void LAPACK_ztprfb( char* side, char* trans, char* direct, char* storev,
+                    lapack_int* m, lapack_int* n, lapack_int* k, lapack_int* l,
+                    const lapack_complex_double* v, lapack_int* ldv,
+                    const lapack_complex_double* t, lapack_int* ldt,
+                    lapack_complex_double* a, lapack_int* lda,
+                    lapack_complex_double* b, lapack_int* ldb,
+                    const double* mywork, lapack_int* myldwork );
+// LAPACK 3.X.X
+void LAPACK_csyr( char* uplo, lapack_int* n, lapack_complex_float* alpha,
+                      const lapack_complex_float* x, lapack_int* incx,
+                      lapack_complex_float* a, lapack_int* lda );
+void LAPACK_zsyr( char* uplo, lapack_int* n, lapack_complex_double* alpha,
+                      const lapack_complex_double* x, lapack_int* incx,
+                      lapack_complex_double* a, lapack_int* lda );
+
+#ifdef __cplusplus
+}
+#endif /* __cplusplus */
+
+#endif /* _LAPACKE_H_ */
+
+#endif /* _MKL_LAPACKE_H_ */
diff --git a/Eigen/src/misc/lapacke_mangling.h b/Eigen/src/misc/lapacke_mangling.h
new file mode 100644
index 0000000..6211fd1
--- /dev/null
+++ b/Eigen/src/misc/lapacke_mangling.h
@@ -0,0 +1,17 @@
+#ifndef LAPACK_HEADER_INCLUDED
+#define LAPACK_HEADER_INCLUDED
+
+#ifndef LAPACK_GLOBAL
+#if defined(LAPACK_GLOBAL_PATTERN_LC) || defined(ADD_)
+#define LAPACK_GLOBAL(lcname,UCNAME)  lcname##_
+#elif defined(LAPACK_GLOBAL_PATTERN_UC) || defined(UPPER)
+#define LAPACK_GLOBAL(lcname,UCNAME)  UCNAME
+#elif defined(LAPACK_GLOBAL_PATTERN_MC) || defined(NOCHANGE)
+#define LAPACK_GLOBAL(lcname,UCNAME)  lcname
+#else
+#define LAPACK_GLOBAL(lcname,UCNAME)  lcname##_
+#endif
+#endif
+
+#endif
+
diff --git a/Eigen/src/plugins/ArrayCwiseBinaryOps.h b/Eigen/src/plugins/ArrayCwiseBinaryOps.h
index 1951286..1f8a531 100644
--- a/Eigen/src/plugins/ArrayCwiseBinaryOps.h
+++ b/Eigen/src/plugins/ArrayCwiseBinaryOps.h
@@ -1,12 +1,14 @@
+
 /** \returns an expression of the coefficient wise product of \c *this and \a other
   *
   * \sa MatrixBase::cwiseProduct
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)
 operator*(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)(derived(), other.derived());
+  return EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient wise quotient of \c *this and \a other
@@ -14,10 +16,11 @@
   * \sa MatrixBase::cwiseQuotient
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_quotient_op<Scalar,typename OtherDerived::Scalar>, const Derived, const OtherDerived>
 operator/(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
+  return CwiseBinaryOp<internal::scalar_quotient_op<Scalar,typename OtherDerived::Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise min of \c *this and \a other
@@ -27,13 +30,14 @@
   *
   * \sa max()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(min,internal::scalar_min_op)
+EIGEN_MAKE_CWISE_BINARY_OP(min,min)
 
 /** \returns an expression of the coefficient-wise min of \c *this and scalar \a other
   *
   * \sa max()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived,
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived,
                                         const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> >
 #ifdef EIGEN_PARSED_BY_DOXYGEN
 min
@@ -52,13 +56,14 @@
   *
   * \sa min()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(max,internal::scalar_max_op)
+EIGEN_MAKE_CWISE_BINARY_OP(max,max)
 
 /** \returns an expression of the coefficient-wise max of \c *this and scalar \a other
   *
   * \sa min()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived,
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived,
                                         const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> >
 #ifdef EIGEN_PARSED_BY_DOXYGEN
 max
@@ -70,33 +75,62 @@
   return (max)(Derived::PlainObject::Constant(rows(), cols(), other));
 }
 
+/** \returns an expression of the coefficient-wise power of \c *this to the given array of \a exponents.
+  *
+  * This function computes the coefficient-wise power.
+  *
+  * Example: \include Cwise_array_power_array.cpp
+  * Output: \verbinclude Cwise_array_power_array.out
+  */
+EIGEN_MAKE_CWISE_BINARY_OP(pow,pow)
 
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(pow,pow)
+#else
+/** \returns an expression of the coefficients of \c *this rasied to the constant power \a exponent
+  *
+  * \tparam T is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression.
+  *
+  * This function computes the coefficient-wise power. The function MatrixBase::pow() in the
+  * unsupported module MatrixFunctions computes the matrix power.
+  *
+  * Example: \include Cwise_pow.cpp
+  * Output: \verbinclude Cwise_pow.out
+  *
+  * \sa ArrayBase::pow(ArrayBase), square(), cube(), exp(), log()
+  */
+template<typename T>
+const CwiseBinaryOp<internal::scalar_pow_op<Scalar,T>,Derived,Constant<T> > pow(const T& exponent) const;
+#endif
+
+
+// TODO code generating macros could be moved to Macros.h and could include generation of documentation
 #define EIGEN_MAKE_CWISE_COMP_OP(OP, COMPARATOR) \
 template<typename OtherDerived> \
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived> \
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<Scalar, typename OtherDerived::Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived> \
 OP(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
 { \
-  return CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived>(derived(), other.derived()); \
+  return CwiseBinaryOp<internal::scalar_cmp_op<Scalar, typename OtherDerived::Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const OtherDerived>(derived(), other.derived()); \
 }\
-typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> > Cmp ## COMPARATOR ## ReturnType; \
-typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_ ## COMPARATOR>, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject>, const Derived > RCmp ## COMPARATOR ## ReturnType; \
-EIGEN_STRONG_INLINE const Cmp ## COMPARATOR ## ReturnType \
+typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,Scalar, internal::cmp_ ## COMPARATOR>, const Derived, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject> > Cmp ## COMPARATOR ## ReturnType; \
+typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,Scalar, internal::cmp_ ## COMPARATOR>, const CwiseNullaryOp<internal::scalar_constant_op<Scalar>, PlainObject>, const Derived > RCmp ## COMPARATOR ## ReturnType; \
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Cmp ## COMPARATOR ## ReturnType \
 OP(const Scalar& s) const { \
   return this->OP(Derived::PlainObject::Constant(rows(), cols(), s)); \
 } \
-friend EIGEN_STRONG_INLINE const RCmp ## COMPARATOR ## ReturnType \
+EIGEN_DEVICE_FUNC friend EIGEN_STRONG_INLINE const RCmp ## COMPARATOR ## ReturnType \
 OP(const Scalar& s, const Derived& d) { \
   return Derived::PlainObject::Constant(d.rows(), d.cols(), s).OP(d); \
 }
 
 #define EIGEN_MAKE_CWISE_COMP_R_OP(OP, R_OP, RCOMPARATOR) \
 template<typename OtherDerived> \
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived> \
+EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_cmp_op<typename OtherDerived::Scalar, Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived> \
 OP(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const \
 { \
-  return CwiseBinaryOp<internal::scalar_cmp_op<Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived>(other.derived(), derived()); \
+  return CwiseBinaryOp<internal::scalar_cmp_op<typename OtherDerived::Scalar, Scalar, internal::cmp_##RCOMPARATOR>, const OtherDerived, const Derived>(other.derived(), derived()); \
 } \
-\
+EIGEN_DEVICE_FUNC \
 inline const RCmp ## RCOMPARATOR ## ReturnType \
 OP(const Scalar& s) const { \
   return Derived::PlainObject::Constant(rows(), cols(), s).R_OP(*this); \
@@ -107,6 +141,7 @@
 }
 
 
+
 /** \returns an expression of the coefficient-wise \< operator of *this and \a other
   *
   * Example: \include Cwise_less.cpp
@@ -171,83 +206,127 @@
   */
 EIGEN_MAKE_CWISE_COMP_OP(operator!=, NEQ)
 
+
 #undef EIGEN_MAKE_CWISE_COMP_OP
 #undef EIGEN_MAKE_CWISE_COMP_R_OP
 
 // scalar addition
-
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP(operator+,sum)
+#else
 /** \returns an expression of \c *this with each coeff incremented by the constant \a scalar
   *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  *
   * Example: \include Cwise_plus.cpp
   * Output: \verbinclude Cwise_plus.out
   *
   * \sa operator+=(), operator-()
   */
-inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>
-operator+(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>(derived(), internal::scalar_add_op<Scalar>(scalar));
-}
+template<typename T>
+const CwiseBinaryOp<internal::scalar_sum_op<Scalar,T>,Derived,Constant<T> > operator+(const T& scalar) const;
+/** \returns an expression of \a expr with each coeff incremented by the constant \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T> friend
+const CwiseBinaryOp<internal::scalar_sum_op<T,Scalar>,Constant<T>,Derived> operator+(const T& scalar, const StorageBaseType& expr);
+#endif
 
-friend inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>
-operator+(const Scalar& scalar,const EIGEN_CURRENT_STORAGE_BASE_CLASS<Derived>& other)
-{
-  return other + scalar;
-}
-
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP(operator-,difference)
+#else
 /** \returns an expression of \c *this with each coeff decremented by the constant \a scalar
   *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  *
   * Example: \include Cwise_minus.cpp
   * Output: \verbinclude Cwise_minus.out
   *
-  * \sa operator+(), operator-=()
+  * \sa operator+=(), operator-()
   */
-inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const Derived>
-operator-(const Scalar& scalar) const
-{
-  return *this + (-scalar);
-}
-
-friend inline const CwiseUnaryOp<internal::scalar_add_op<Scalar>, const CwiseUnaryOp<internal::scalar_opposite_op<Scalar>, const Derived> >
-operator-(const Scalar& scalar,const EIGEN_CURRENT_STORAGE_BASE_CLASS<Derived>& other)
-{
-  return (-other) + scalar;
-}
-
-/** \returns an expression of the coefficient-wise && operator of *this and \a other
+template<typename T>
+const CwiseBinaryOp<internal::scalar_difference_op<Scalar,T>,Derived,Constant<T> > operator-(const T& scalar) const;
+/** \returns an expression of the constant matrix of value \a scalar decremented by the coefficients of \a expr
   *
-  * \warning this operator is for expression of bool only.
-  *
-  * Example: \include Cwise_boolean_and.cpp
-  * Output: \verbinclude Cwise_boolean_and.out
-  *
-  * \sa operator||(), select()
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
   */
+template<typename T> friend
+const CwiseBinaryOp<internal::scalar_difference_op<T,Scalar>,Constant<T>,Derived> operator-(const T& scalar, const StorageBaseType& expr);
+#endif
+
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+  EIGEN_MAKE_SCALAR_BINARY_OP_ONTHELEFT(operator/,quotient)
+#else
+  /**
+    * \brief Component-wise division of the scalar \a s by array elements of \a a.
+    *
+    * \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar).
+    */
+  template<typename T> friend
+  inline const CwiseBinaryOp<internal::scalar_quotient_op<T,Scalar>,Constant<T>,Derived>
+  operator/(const T& s,const StorageBaseType& a);
+#endif
+
+/** \returns an expression of the coefficient-wise ^ operator of *this and \a other
+ *
+ * \warning this operator is for expression of bool only.
+ *
+ * Example: \include Cwise_boolean_xor.cpp
+ * Output: \verbinclude Cwise_boolean_xor.out
+ *
+ * \sa operator&&(), select()
+ */
 template<typename OtherDerived>
-inline const CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>
-operator&&(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+EIGEN_DEVICE_FUNC
+inline const CwiseBinaryOp<internal::scalar_boolean_xor_op, const Derived, const OtherDerived>
+operator^(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
   EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
                       THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
-  return CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>(derived(),other.derived());
+  return CwiseBinaryOp<internal::scalar_boolean_xor_op, const Derived, const OtherDerived>(derived(),other.derived());
 }
 
-/** \returns an expression of the coefficient-wise || operator of *this and \a other
+// NOTE disabled until we agree on argument order
+#if 0
+/** \cpp11 \returns an expression of the coefficient-wise polygamma function.
   *
-  * \warning this operator is for expression of bool only.
+  * \specialfunctions_module
   *
-  * Example: \include Cwise_boolean_or.cpp
-  * Output: \verbinclude Cwise_boolean_or.out
+  * It returns the \a n -th derivative of the digamma(psi) evaluated at \c *this.
   *
-  * \sa operator&&(), select()
+  * \warning Be careful with the order of the parameters: x.polygamma(n) is equivalent to polygamma(n,x)
+  *
+  * \sa Eigen::polygamma()
   */
-template<typename OtherDerived>
-inline const CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>
-operator||(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+template<typename DerivedN>
+inline const CwiseBinaryOp<internal::scalar_polygamma_op<Scalar>, const DerivedN, const Derived>
+polygamma(const EIGEN_CURRENT_STORAGE_BASE_CLASS<DerivedN> &n) const
 {
-  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
-                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
-  return CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>(derived(),other.derived());
+  return CwiseBinaryOp<internal::scalar_polygamma_op<Scalar>, const DerivedN, const Derived>(n.derived(), this->derived());
 }
+#endif
 
-
+/** \returns an expression of the coefficient-wise zeta function.
+  *
+  * \specialfunctions_module
+  *
+  * It returns the Riemann zeta function of two arguments \c *this and \a q:
+  *
+  * \param *this is the exposent, it must be > 1
+  * \param q is the shift, it must be > 0
+  *
+  * \note This function supports only float and double scalar types. To support other scalar types, the user has
+  * to provide implementations of zeta(T,T) for any scalar type T to be supported.
+  *
+  * This method is an alias for zeta(*this,q);
+  *
+  * \sa Eigen::zeta()
+  */
+template<typename DerivedQ>
+inline const CwiseBinaryOp<internal::scalar_zeta_op<Scalar>, const Derived, const DerivedQ>
+zeta(const EIGEN_CURRENT_STORAGE_BASE_CLASS<DerivedQ> &q) const
+{
+  return CwiseBinaryOp<internal::scalar_zeta_op<Scalar>, const Derived, const DerivedQ>(this->derived(), q.derived());
+}
diff --git a/Eigen/src/plugins/ArrayCwiseUnaryOps.h b/Eigen/src/plugins/ArrayCwiseUnaryOps.h
index 1c3ed3f..ebaa3f1 100644
--- a/Eigen/src/plugins/ArrayCwiseUnaryOps.h
+++ b/Eigen/src/plugins/ArrayCwiseUnaryOps.h
@@ -1,16 +1,62 @@
 
 
+typedef CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived> AbsReturnType;
+typedef CwiseUnaryOp<internal::scalar_arg_op<Scalar>, const Derived> ArgReturnType;
+typedef CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived> Abs2ReturnType;
+typedef CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived> SqrtReturnType;
+typedef CwiseUnaryOp<internal::scalar_rsqrt_op<Scalar>, const Derived> RsqrtReturnType;
+typedef CwiseUnaryOp<internal::scalar_sign_op<Scalar>, const Derived> SignReturnType;
+typedef CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived> InverseReturnType;
+typedef CwiseUnaryOp<internal::scalar_boolean_not_op<Scalar>, const Derived> BooleanNotReturnType;
+
+typedef CwiseUnaryOp<internal::scalar_exp_op<Scalar>, const Derived> ExpReturnType;
+typedef CwiseUnaryOp<internal::scalar_log_op<Scalar>, const Derived> LogReturnType;
+typedef CwiseUnaryOp<internal::scalar_log1p_op<Scalar>, const Derived> Log1pReturnType;
+typedef CwiseUnaryOp<internal::scalar_log10_op<Scalar>, const Derived> Log10ReturnType;
+typedef CwiseUnaryOp<internal::scalar_cos_op<Scalar>, const Derived> CosReturnType;
+typedef CwiseUnaryOp<internal::scalar_sin_op<Scalar>, const Derived> SinReturnType;
+typedef CwiseUnaryOp<internal::scalar_tan_op<Scalar>, const Derived> TanReturnType;
+typedef CwiseUnaryOp<internal::scalar_acos_op<Scalar>, const Derived> AcosReturnType;
+typedef CwiseUnaryOp<internal::scalar_asin_op<Scalar>, const Derived> AsinReturnType;
+typedef CwiseUnaryOp<internal::scalar_atan_op<Scalar>, const Derived> AtanReturnType;
+typedef CwiseUnaryOp<internal::scalar_tanh_op<Scalar>, const Derived> TanhReturnType;
+typedef CwiseUnaryOp<internal::scalar_sinh_op<Scalar>, const Derived> SinhReturnType;
+typedef CwiseUnaryOp<internal::scalar_cosh_op<Scalar>, const Derived> CoshReturnType;
+typedef CwiseUnaryOp<internal::scalar_square_op<Scalar>, const Derived> SquareReturnType;
+typedef CwiseUnaryOp<internal::scalar_cube_op<Scalar>, const Derived> CubeReturnType;
+typedef CwiseUnaryOp<internal::scalar_round_op<Scalar>, const Derived> RoundReturnType;
+typedef CwiseUnaryOp<internal::scalar_floor_op<Scalar>, const Derived> FloorReturnType;
+typedef CwiseUnaryOp<internal::scalar_ceil_op<Scalar>, const Derived> CeilReturnType;
+typedef CwiseUnaryOp<internal::scalar_isnan_op<Scalar>, const Derived> IsNaNReturnType;
+typedef CwiseUnaryOp<internal::scalar_isinf_op<Scalar>, const Derived> IsInfReturnType;
+typedef CwiseUnaryOp<internal::scalar_isfinite_op<Scalar>, const Derived> IsFiniteReturnType;
+
 /** \returns an expression of the coefficient-wise absolute value of \c *this
   *
   * Example: \include Cwise_abs.cpp
   * Output: \verbinclude Cwise_abs.out
   *
-  * \sa abs2()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_abs">Math functions</a>, abs2()
   */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const AbsReturnType
 abs() const
 {
-  return derived();
+  return AbsReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise phase angle of \c *this
+  *
+  * Example: \include Cwise_arg.cpp
+  * Output: \verbinclude Cwise_arg.out
+  *
+  * \sa abs()
+  */
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const ArgReturnType
+arg() const
+{
+  return ArgReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise squared absolute value of \c *this
@@ -18,104 +64,162 @@
   * Example: \include Cwise_abs2.cpp
   * Output: \verbinclude Cwise_abs2.out
   *
-  * \sa abs(), square()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_abs2">Math functions</a>, abs(), square()
   */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const Abs2ReturnType
 abs2() const
 {
-  return derived();
+  return Abs2ReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise exponential of *this.
   *
+  * This function computes the coefficient-wise exponential. The function MatrixBase::exp() in the
+  * unsupported module MatrixFunctions computes the matrix exponential.
+  *
   * Example: \include Cwise_exp.cpp
   * Output: \verbinclude Cwise_exp.out
   *
-  * \sa pow(), log(), sin(), cos()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_exp">Math functions</a>, pow(), log(), sin(), cos()
   */
-inline const CwiseUnaryOp<internal::scalar_exp_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const ExpReturnType
 exp() const
 {
-  return derived();
+  return ExpReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise logarithm of *this.
   *
+  * This function computes the coefficient-wise logarithm. The function MatrixBase::log() in the
+  * unsupported module MatrixFunctions computes the matrix logarithm.
+  *
   * Example: \include Cwise_log.cpp
   * Output: \verbinclude Cwise_log.out
   *
-  * \sa exp()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_log">Math functions</a>, exp()
   */
-inline const CwiseUnaryOp<internal::scalar_log_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const LogReturnType
 log() const
 {
-  return derived();
+  return LogReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise logarithm of 1 plus \c *this.
+  *
+  * In exact arithmetic, \c x.log() is equivalent to \c (x+1).log(),
+  * however, with finite precision, this function is much more accurate when \c x is close to zero.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_log1p">Math functions</a>, log()
+  */
+EIGEN_DEVICE_FUNC
+inline const Log1pReturnType
+log1p() const
+{
+  return Log1pReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise base-10 logarithm of *this.
+  *
+  * This function computes the coefficient-wise base-10 logarithm.
+  *
+  * Example: \include Cwise_log10.cpp
+  * Output: \verbinclude Cwise_log10.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_log10">Math functions</a>, log()
+  */
+EIGEN_DEVICE_FUNC
+inline const Log10ReturnType
+log10() const
+{
+  return Log10ReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise square root of *this.
   *
+  * This function computes the coefficient-wise square root. The function MatrixBase::sqrt() in the
+  * unsupported module MatrixFunctions computes the matrix square root.
+  *
+  * Example: \include Cwise_sqrt.cpp
+  * Output: \verbinclude Cwise_sqrt.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_sqrt">Math functions</a>, pow(), square()
+  */
+EIGEN_DEVICE_FUNC
+inline const SqrtReturnType
+sqrt() const
+{
+  return SqrtReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise inverse square root of *this.
+  *
+  * This function computes the coefficient-wise inverse square root.
+  *
   * Example: \include Cwise_sqrt.cpp
   * Output: \verbinclude Cwise_sqrt.out
   *
   * \sa pow(), square()
   */
-inline const CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived>
-sqrt() const
+EIGEN_DEVICE_FUNC
+inline const RsqrtReturnType
+rsqrt() const
 {
-  return derived();
+  return RsqrtReturnType(derived());
 }
 
+/** \returns an expression of the coefficient-wise signum of *this.
+  *
+  * This function computes the coefficient-wise signum.
+  *
+  * Example: \include Cwise_sign.cpp
+  * Output: \verbinclude Cwise_sign.out
+  *
+  * \sa pow(), square()
+  */
+EIGEN_DEVICE_FUNC
+inline const SignReturnType
+sign() const
+{
+  return SignReturnType(derived());
+}
+
+
 /** \returns an expression of the coefficient-wise cosine of *this.
   *
+  * This function computes the coefficient-wise cosine. The function MatrixBase::cos() in the
+  * unsupported module MatrixFunctions computes the matrix cosine.
+  *
   * Example: \include Cwise_cos.cpp
   * Output: \verbinclude Cwise_cos.out
   *
-  * \sa sin(), acos()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_cos">Math functions</a>, sin(), acos()
   */
-inline const CwiseUnaryOp<internal::scalar_cos_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const CosReturnType
 cos() const
 {
-  return derived();
+  return CosReturnType(derived());
 }
 
 
 /** \returns an expression of the coefficient-wise sine of *this.
   *
+  * This function computes the coefficient-wise sine. The function MatrixBase::sin() in the
+  * unsupported module MatrixFunctions computes the matrix sine.
+  *
   * Example: \include Cwise_sin.cpp
   * Output: \verbinclude Cwise_sin.out
   *
-  * \sa cos(), asin()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_sin">Math functions</a>, cos(), asin()
   */
-inline const CwiseUnaryOp<internal::scalar_sin_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const SinReturnType
 sin() const
 {
-  return derived();
-}
-
-/** \returns an expression of the coefficient-wise arc cosine of *this.
-  *
-  * Example: \include Cwise_acos.cpp
-  * Output: \verbinclude Cwise_acos.out
-  *
-  * \sa cos(), asin()
-  */
-inline const CwiseUnaryOp<internal::scalar_acos_op<Scalar>, const Derived>
-acos() const
-{
-  return derived();
-}
-
-/** \returns an expression of the coefficient-wise arc sine of *this.
-  *
-  * Example: \include Cwise_asin.cpp
-  * Output: \verbinclude Cwise_asin.out
-  *
-  * \sa sin(), acos()
-  */
-inline const CwiseUnaryOp<internal::scalar_asin_op<Scalar>, const Derived>
-asin() const
-{
-  return derived();
+  return SinReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise tan of *this.
@@ -123,29 +227,98 @@
   * Example: \include Cwise_tan.cpp
   * Output: \verbinclude Cwise_tan.out
   *
-  * \sa cos(), sin()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_tan">Math functions</a>, cos(), sin()
   */
-inline const CwiseUnaryOp<internal::scalar_tan_op<Scalar>, Derived>
+EIGEN_DEVICE_FUNC
+inline const TanReturnType
 tan() const
 {
-  return derived();
+  return TanReturnType(derived());
 }
 
-
-/** \returns an expression of the coefficient-wise power of *this to the given exponent.
+/** \returns an expression of the coefficient-wise arc tan of *this.
   *
-  * Example: \include Cwise_pow.cpp
-  * Output: \verbinclude Cwise_pow.out
+  * Example: \include Cwise_atan.cpp
+  * Output: \verbinclude Cwise_atan.out
   *
-  * \sa exp(), log()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_atan">Math functions</a>, tan(), asin(), acos()
   */
-inline const CwiseUnaryOp<internal::scalar_pow_op<Scalar>, const Derived>
-pow(const Scalar& exponent) const
+EIGEN_DEVICE_FUNC
+inline const AtanReturnType
+atan() const
 {
-  return CwiseUnaryOp<internal::scalar_pow_op<Scalar>, const Derived>
-          (derived(), internal::scalar_pow_op<Scalar>(exponent));
+  return AtanReturnType(derived());
 }
 
+/** \returns an expression of the coefficient-wise arc cosine of *this.
+  *
+  * Example: \include Cwise_acos.cpp
+  * Output: \verbinclude Cwise_acos.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_acos">Math functions</a>, cos(), asin()
+  */
+EIGEN_DEVICE_FUNC
+inline const AcosReturnType
+acos() const
+{
+  return AcosReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise arc sine of *this.
+  *
+  * Example: \include Cwise_asin.cpp
+  * Output: \verbinclude Cwise_asin.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_asin">Math functions</a>, sin(), acos()
+  */
+EIGEN_DEVICE_FUNC
+inline const AsinReturnType
+asin() const
+{
+  return AsinReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise hyperbolic tan of *this.
+  *
+  * Example: \include Cwise_tanh.cpp
+  * Output: \verbinclude Cwise_tanh.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_tanh">Math functions</a>, tan(), sinh(), cosh()
+  */
+EIGEN_DEVICE_FUNC
+inline const TanhReturnType
+tanh() const
+{
+  return TanhReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise hyperbolic sin of *this.
+  *
+  * Example: \include Cwise_sinh.cpp
+  * Output: \verbinclude Cwise_sinh.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_sinh">Math functions</a>, sin(), tanh(), cosh()
+  */
+EIGEN_DEVICE_FUNC
+inline const SinhReturnType
+sinh() const
+{
+  return SinhReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise hyperbolic cos of *this.
+  *
+  * Example: \include Cwise_cosh.cpp
+  * Output: \verbinclude Cwise_cosh.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_cosh">Math functions</a>, tan(), sinh(), cosh()
+  */
+EIGEN_DEVICE_FUNC
+inline const CoshReturnType
+cosh() const
+{
+  return CoshReturnType(derived());
+}
 
 /** \returns an expression of the coefficient-wise inverse of *this.
   *
@@ -154,10 +327,11 @@
   *
   * \sa operator/(), operator*()
   */
-inline const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const InverseReturnType
 inverse() const
 {
-  return derived();
+  return InverseReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise square of *this.
@@ -165,12 +339,13 @@
   * Example: \include Cwise_square.cpp
   * Output: \verbinclude Cwise_square.out
   *
-  * \sa operator/(), operator*(), abs2()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_squareE">Math functions</a>, abs2(), cube(), pow()
   */
-inline const CwiseUnaryOp<internal::scalar_square_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const SquareReturnType
 square() const
 {
-  return derived();
+  return SquareReturnType(derived());
 }
 
 /** \returns an expression of the coefficient-wise cube of *this.
@@ -178,10 +353,200 @@
   * Example: \include Cwise_cube.cpp
   * Output: \verbinclude Cwise_cube.out
   *
-  * \sa square(), pow()
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_cube">Math functions</a>, square(), pow()
   */
-inline const CwiseUnaryOp<internal::scalar_cube_op<Scalar>, const Derived>
+EIGEN_DEVICE_FUNC
+inline const CubeReturnType
 cube() const
 {
-  return derived();
+  return CubeReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise round of *this.
+  *
+  * Example: \include Cwise_round.cpp
+  * Output: \verbinclude Cwise_round.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_round">Math functions</a>, ceil(), floor()
+  */
+EIGEN_DEVICE_FUNC
+inline const RoundReturnType
+round() const
+{
+  return RoundReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise floor of *this.
+  *
+  * Example: \include Cwise_floor.cpp
+  * Output: \verbinclude Cwise_floor.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_floor">Math functions</a>, ceil(), round()
+  */
+EIGEN_DEVICE_FUNC
+inline const FloorReturnType
+floor() const
+{
+  return FloorReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise ceil of *this.
+  *
+  * Example: \include Cwise_ceil.cpp
+  * Output: \verbinclude Cwise_ceil.out
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_ceil">Math functions</a>, floor(), round()
+  */
+EIGEN_DEVICE_FUNC
+inline const CeilReturnType
+ceil() const
+{
+  return CeilReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise isnan of *this.
+  *
+  * Example: \include Cwise_isNaN.cpp
+  * Output: \verbinclude Cwise_isNaN.out
+  *
+  * \sa isfinite(), isinf()
+  */
+EIGEN_DEVICE_FUNC
+inline const IsNaNReturnType
+isNaN() const
+{
+  return IsNaNReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise isinf of *this.
+  *
+  * Example: \include Cwise_isInf.cpp
+  * Output: \verbinclude Cwise_isInf.out
+  *
+  * \sa isnan(), isfinite()
+  */
+EIGEN_DEVICE_FUNC
+inline const IsInfReturnType
+isInf() const
+{
+  return IsInfReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise isfinite of *this.
+  *
+  * Example: \include Cwise_isFinite.cpp
+  * Output: \verbinclude Cwise_isFinite.out
+  *
+  * \sa isnan(), isinf()
+  */
+EIGEN_DEVICE_FUNC
+inline const IsFiniteReturnType
+isFinite() const
+{
+  return IsFiniteReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise ! operator of *this
+  *
+  * \warning this operator is for expression of bool only.
+  *
+  * Example: \include Cwise_boolean_not.cpp
+  * Output: \verbinclude Cwise_boolean_not.out
+  *
+  * \sa operator!=()
+  */
+EIGEN_DEVICE_FUNC
+inline const BooleanNotReturnType
+operator!() const
+{
+  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value),
+                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
+  return BooleanNotReturnType(derived());
+}
+
+
+// --- SpecialFunctions module ---
+
+typedef CwiseUnaryOp<internal::scalar_lgamma_op<Scalar>, const Derived> LgammaReturnType;
+typedef CwiseUnaryOp<internal::scalar_digamma_op<Scalar>, const Derived> DigammaReturnType;
+typedef CwiseUnaryOp<internal::scalar_erf_op<Scalar>, const Derived> ErfReturnType;
+typedef CwiseUnaryOp<internal::scalar_erfc_op<Scalar>, const Derived> ErfcReturnType;
+
+/** \cpp11 \returns an expression of the coefficient-wise ln(|gamma(*this)|).
+  *
+  * \specialfunctions_module
+  *
+  * Example: \include Cwise_lgamma.cpp
+  * Output: \verbinclude Cwise_lgamma.out
+  *
+  * \note This function supports only float and double scalar types in c++11 mode. To support other scalar types,
+  * or float/double in non c++11 mode, the user has to provide implementations of lgamma(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_lgamma">Math functions</a>, digamma()
+  */
+EIGEN_DEVICE_FUNC
+inline const LgammaReturnType
+lgamma() const
+{
+  return LgammaReturnType(derived());
+}
+
+/** \returns an expression of the coefficient-wise digamma (psi, derivative of lgamma).
+  *
+  * \specialfunctions_module
+  *
+  * \note This function supports only float and double scalar types. To support other scalar types,
+  * the user has to provide implementations of digamma(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_digamma">Math functions</a>, Eigen::digamma(), Eigen::polygamma(), lgamma()
+  */
+EIGEN_DEVICE_FUNC
+inline const DigammaReturnType
+digamma() const
+{
+  return DigammaReturnType(derived());
+}
+
+/** \cpp11 \returns an expression of the coefficient-wise Gauss error
+  * function of *this.
+  *
+  * \specialfunctions_module
+  *
+  * Example: \include Cwise_erf.cpp
+  * Output: \verbinclude Cwise_erf.out
+  *
+  * \note This function supports only float and double scalar types in c++11 mode. To support other scalar types,
+  * or float/double in non c++11 mode, the user has to provide implementations of erf(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_erf">Math functions</a>, erfc()
+  */
+EIGEN_DEVICE_FUNC
+inline const ErfReturnType
+erf() const
+{
+  return ErfReturnType(derived());
+}
+
+/** \cpp11 \returns an expression of the coefficient-wise Complementary error
+  * function of *this.
+  *
+  * \specialfunctions_module
+  *
+  * Example: \include Cwise_erfc.cpp
+  * Output: \verbinclude Cwise_erfc.out
+  *
+  * \note This function supports only float and double scalar types in c++11 mode. To support other scalar types,
+  * or float/double in non c++11 mode, the user has to provide implementations of erfc(T) for any scalar
+  * type T to be supported.
+  *
+  * \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_erfc">Math functions</a>, erf()
+  */
+EIGEN_DEVICE_FUNC
+inline const ErfcReturnType
+erfc() const
+{
+  return ErfcReturnType(derived());
 }
diff --git a/Eigen/src/plugins/BlockMethods.h b/Eigen/src/plugins/BlockMethods.h
index 2788251..ac35a00 100644
--- a/Eigen/src/plugins/BlockMethods.h
+++ b/Eigen/src/plugins/BlockMethods.h
@@ -8,27 +8,32 @@
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 
-/** \internal expression type of a column */
+/// \internal expression type of a column */
 typedef Block<Derived, internal::traits<Derived>::RowsAtCompileTime, 1, !IsRowMajor> ColXpr;
 typedef const Block<const Derived, internal::traits<Derived>::RowsAtCompileTime, 1, !IsRowMajor> ConstColXpr;
-/** \internal expression type of a row */
+/// \internal expression type of a row */
 typedef Block<Derived, 1, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> RowXpr;
 typedef const Block<const Derived, 1, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> ConstRowXpr;
-/** \internal expression type of a block of whole columns */
+/// \internal expression type of a block of whole columns */
 typedef Block<Derived, internal::traits<Derived>::RowsAtCompileTime, Dynamic, !IsRowMajor> ColsBlockXpr;
 typedef const Block<const Derived, internal::traits<Derived>::RowsAtCompileTime, Dynamic, !IsRowMajor> ConstColsBlockXpr;
-/** \internal expression type of a block of whole rows */
+/// \internal expression type of a block of whole rows */
 typedef Block<Derived, Dynamic, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> RowsBlockXpr;
 typedef const Block<const Derived, Dynamic, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> ConstRowsBlockXpr;
-/** \internal expression type of a block of whole columns */
+/// \internal expression type of a block of whole columns */
 template<int N> struct NColsBlockXpr { typedef Block<Derived, internal::traits<Derived>::RowsAtCompileTime, N, !IsRowMajor> Type; };
 template<int N> struct ConstNColsBlockXpr { typedef const Block<const Derived, internal::traits<Derived>::RowsAtCompileTime, N, !IsRowMajor> Type; };
-/** \internal expression type of a block of whole rows */
+/// \internal expression type of a block of whole rows */
 template<int N> struct NRowsBlockXpr { typedef Block<Derived, N, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> Type; };
 template<int N> struct ConstNRowsBlockXpr { typedef const Block<const Derived, N, internal::traits<Derived>::ColsAtCompileTime, IsRowMajor> Type; };
+/// \internal expression of a block */
+typedef Block<Derived> BlockXpr;
+typedef const Block<const Derived> ConstBlockXpr;
+/// \internal expression of a block of fixed sizes */
+template<int Rows, int Cols> struct FixedBlockXpr { typedef Block<Derived,Rows,Cols> Type; };
+template<int Rows, int Cols> struct ConstFixedBlockXpr { typedef Block<const Derived,Rows,Cols> Type; };
 
 typedef VectorBlock<Derived> SegmentReturnType;
 typedef const VectorBlock<const Derived> ConstSegmentReturnType;
@@ -37,378 +42,430 @@
 
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
-/** \returns a dynamic-size expression of a block in *this.
-  *
-  * \param startRow the first row in the block
-  * \param startCol the first column in the block
-  * \param blockRows the number of rows in the block
-  * \param blockCols the number of columns in the block
-  *
-  * Example: \include MatrixBase_block_int_int_int_int.cpp
-  * Output: \verbinclude MatrixBase_block_int_int_int_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size matrix, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, block(Index,Index)
-  */
-inline Block<Derived> block(Index startRow, Index startCol, Index blockRows, Index blockCols)
+/// \returns a dynamic-size expression of a block in *this.
+///
+/// \param startRow the first row in the block
+/// \param startCol the first column in the block
+/// \param blockRows the number of rows in the block
+/// \param blockCols the number of columns in the block
+///
+/// Example: \include MatrixBase_block_int_int_int_int.cpp
+/// Output: \verbinclude MatrixBase_block_int_int_int_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size matrix, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr block(Index startRow, Index startCol, Index blockRows, Index blockCols)
 {
-  return Block<Derived>(derived(), startRow, startCol, blockRows, blockCols);
+  return BlockXpr(derived(), startRow, startCol, blockRows, blockCols);
 }
 
-/** This is the const version of block(Index,Index,Index,Index). */
-inline const Block<const Derived> block(Index startRow, Index startCol, Index blockRows, Index blockCols) const
+/// This is the const version of block(Index,Index,Index,Index). */
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr block(Index startRow, Index startCol, Index blockRows, Index blockCols) const
 {
-  return Block<const Derived>(derived(), startRow, startCol, blockRows, blockCols);
+  return ConstBlockXpr(derived(), startRow, startCol, blockRows, blockCols);
 }
 
 
 
 
-/** \returns a dynamic-size expression of a top-right corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_topRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_topRightCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> topRightCorner(Index cRows, Index cCols)
+/// \returns a dynamic-size expression of a top-right corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_topRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_topRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr topRightCorner(Index cRows, Index cCols)
 {
-  return Block<Derived>(derived(), 0, cols() - cCols, cRows, cCols);
+  return BlockXpr(derived(), 0, cols() - cCols, cRows, cCols);
 }
 
-/** This is the const version of topRightCorner(Index, Index).*/
-inline const Block<const Derived> topRightCorner(Index cRows, Index cCols) const
+/// This is the const version of topRightCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr topRightCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived>(derived(), 0, cols() - cCols, cRows, cCols);
+  return ConstBlockXpr(derived(), 0, cols() - cCols, cRows, cCols);
 }
 
-/** \returns an expression of a fixed-size top-right corner of *this.
-  *
-  * \tparam CRows the number of rows in the corner
-  * \tparam CCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_template_int_int_topRightCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topRightCorner.out
-  *
-  * \sa class Block, block<int,int>(Index,Index)
-  */
+/// \returns an expression of a fixed-size top-right corner of *this.
+///
+/// \tparam CRows the number of rows in the corner
+/// \tparam CCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_template_int_int_topRightCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topRightCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block<int,int>(Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topRightCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type topRightCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, cols() - CCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - CCols);
 }
 
-/** This is the const version of topRightCorner<int, int>().*/
+/// This is the const version of topRightCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topRightCorner() const
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topRightCorner() const
 {
-  return Block<const Derived, CRows, CCols>(derived(), 0, cols() - CCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - CCols);
 }
 
-/** \returns an expression of a top-right corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_topRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topRightCorner_int_int.out
-  *
-  * \sa class Block
-  */
+/// \returns an expression of a top-right corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_topRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topRightCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type topRightCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, cols() - cCols, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - cCols, cRows, cCols);
 }
 
-/** This is the const version of topRightCorner<int, int>(Index, Index).*/
+/// This is the const version of topRightCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topRightCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topRightCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), 0, cols() - cCols, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, cols() - cCols, cRows, cCols);
 }
 
 
 
-/** \returns a dynamic-size expression of a top-left corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_topLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_topLeftCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> topLeftCorner(Index cRows, Index cCols)
+/// \returns a dynamic-size expression of a top-left corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_topLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_topLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr topLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived>(derived(), 0, 0, cRows, cCols);
+  return BlockXpr(derived(), 0, 0, cRows, cCols);
 }
 
-/** This is the const version of topLeftCorner(Index, Index).*/
-inline const Block<const Derived> topLeftCorner(Index cRows, Index cCols) const
+/// This is the const version of topLeftCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr topLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived>(derived(), 0, 0, cRows, cCols);
+  return ConstBlockXpr(derived(), 0, 0, cRows, cCols);
 }
 
-/** \returns an expression of a fixed-size top-left corner of *this.
-  *
-  * The template parameters CRows and CCols are the number of rows and columns in the corner.
-  *
-  * Example: \include MatrixBase_template_int_int_topLeftCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topLeftCorner.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns an expression of a fixed-size top-left corner of *this.
+///
+/// The template parameters CRows and CCols are the number of rows and columns in the corner.
+///
+/// Example: \include MatrixBase_template_int_int_topLeftCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topLeftCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topLeftCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type topLeftCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, 0);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0);
 }
 
-/** This is the const version of topLeftCorner<int, int>().*/
+/// This is the const version of topLeftCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topLeftCorner() const
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topLeftCorner() const
 {
-  return Block<const Derived, CRows, CCols>(derived(), 0, 0);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0);
 }
 
-/** \returns an expression of a top-left corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_topLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_topLeftCorner_int_int.out
-  *
-  * \sa class Block
-  */
+/// \returns an expression of a top-left corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_topLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_topLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> topLeftCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type topLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), 0, 0, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0, cRows, cCols);
 }
 
-/** This is the const version of topLeftCorner<int, int>(Index, Index).*/
+/// This is the const version of topLeftCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> topLeftCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type topLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), 0, 0, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), 0, 0, cRows, cCols);
 }
 
 
 
-/** \returns a dynamic-size expression of a bottom-right corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_bottomRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_bottomRightCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> bottomRightCorner(Index cRows, Index cCols)
+/// \returns a dynamic-size expression of a bottom-right corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_bottomRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_bottomRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr bottomRightCorner(Index cRows, Index cCols)
 {
-  return Block<Derived>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return BlockXpr(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
-/** This is the const version of bottomRightCorner(Index, Index).*/
-inline const Block<const Derived> bottomRightCorner(Index cRows, Index cCols) const
+/// This is the const version of bottomRightCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr bottomRightCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return ConstBlockXpr(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
-/** \returns an expression of a fixed-size bottom-right corner of *this.
-  *
-  * The template parameters CRows and CCols are the number of rows and columns in the corner.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomRightCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns an expression of a fixed-size bottom-right corner of *this.
+///
+/// The template parameters CRows and CCols are the number of rows and columns in the corner.
+///
+/// Example: \include MatrixBase_template_int_int_bottomRightCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomRightCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomRightCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - CRows, cols() - CCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, cols() - CCols);
 }
 
-/** This is the const version of bottomRightCorner<int, int>().*/
+/// This is the const version of bottomRightCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomRightCorner() const
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomRightCorner() const
 {
-  return Block<const Derived, CRows, CCols>(derived(), rows() - CRows, cols() - CCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, cols() - CCols);
 }
 
-/** \returns an expression of a bottom-right corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomRightCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner_int_int.out
-  *
-  * \sa class Block
-  */
+/// \returns an expression of a bottom-right corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_bottomRightCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomRightCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomRightCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomRightCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
-/** This is the const version of bottomRightCorner<int, int>(Index, Index).*/
+/// This is the const version of bottomRightCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomRightCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomRightCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, cols() - cCols, cRows, cCols);
 }
 
 
 
-/** \returns a dynamic-size expression of a bottom-left corner of *this.
-  *
-  * \param cRows the number of rows in the corner
-  * \param cCols the number of columns in the corner
-  *
-  * Example: \include MatrixBase_bottomLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_bottomLeftCorner_int_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-inline Block<Derived> bottomLeftCorner(Index cRows, Index cCols)
+/// \returns a dynamic-size expression of a bottom-left corner of *this.
+///
+/// \param cRows the number of rows in the corner
+/// \param cCols the number of columns in the corner
+///
+/// Example: \include MatrixBase_bottomLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_bottomLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
+inline BlockXpr bottomLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived>(derived(), rows() - cRows, 0, cRows, cCols);
+  return BlockXpr(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
-/** This is the const version of bottomLeftCorner(Index, Index).*/
-inline const Block<const Derived> bottomLeftCorner(Index cRows, Index cCols) const
+/// This is the const version of bottomLeftCorner(Index, Index).
+EIGEN_DEVICE_FUNC
+inline const ConstBlockXpr bottomLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived>(derived(), rows() - cRows, 0, cRows, cCols);
+  return ConstBlockXpr(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
-/** \returns an expression of a fixed-size bottom-left corner of *this.
-  *
-  * The template parameters CRows and CCols are the number of rows and columns in the corner.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomLeftCorner.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns an expression of a fixed-size bottom-left corner of *this.
+///
+/// The template parameters CRows and CCols are the number of rows and columns in the corner.
+///
+/// Example: \include MatrixBase_template_int_int_bottomLeftCorner.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomLeftCorner()
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomLeftCorner()
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - CRows, 0);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, 0);
 }
 
-/** This is the const version of bottomLeftCorner<int, int>().*/
+/// This is the const version of bottomLeftCorner<int, int>().
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomLeftCorner() const
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomLeftCorner() const
 {
-  return Block<const Derived, CRows, CCols>(derived(), rows() - CRows, 0);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - CRows, 0);
 }
 
-/** \returns an expression of a bottom-left corner of *this.
-  *
-  * \tparam CRows number of rows in corner as specified at compile-time
-  * \tparam CCols number of columns in corner as specified at compile-time
-  * \param  cRows number of rows in corner as specified at run-time
-  * \param  cCols number of columns in corner as specified at run-time
-  *
-  * This function is mainly useful for corners where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a cRows should equal \a CRows unless
-  * \a CRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_bottomLeftCorner_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner_int_int.out
-  *
-  * \sa class Block
-  */
+/// \returns an expression of a bottom-left corner of *this.
+///
+/// \tparam CRows number of rows in corner as specified at compile-time
+/// \tparam CCols number of columns in corner as specified at compile-time
+/// \param  cRows number of rows in corner as specified at run-time
+/// \param  cCols number of columns in corner as specified at run-time
+///
+/// This function is mainly useful for corners where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a cRows should equal \a CRows unless
+/// \a CRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_bottomLeftCorner_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_bottomLeftCorner_int_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block
+///
 template<int CRows, int CCols>
-inline Block<Derived, CRows, CCols> bottomLeftCorner(Index cRows, Index cCols)
+inline typename FixedBlockXpr<CRows,CCols>::Type bottomLeftCorner(Index cRows, Index cCols)
 {
-  return Block<Derived, CRows, CCols>(derived(), rows() - cRows, 0, cRows, cCols);
+  return typename FixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
-/** This is the const version of bottomLeftCorner<int, int>(Index, Index).*/
+/// This is the const version of bottomLeftCorner<int, int>(Index, Index).
 template<int CRows, int CCols>
-inline const Block<const Derived, CRows, CCols> bottomLeftCorner(Index cRows, Index cCols) const
+inline const typename ConstFixedBlockXpr<CRows,CCols>::Type bottomLeftCorner(Index cRows, Index cCols) const
 {
-  return Block<const Derived, CRows, CCols>(derived(), rows() - cRows, 0, cRows, cCols);
+  return typename ConstFixedBlockXpr<CRows,CCols>::Type(derived(), rows() - cRows, 0, cRows, cCols);
 }
 
 
 
-/** \returns a block consisting of the top rows of *this.
-  *
-  * \param n the number of rows in the block
-  *
-  * Example: \include MatrixBase_topRows_int.cpp
-  * Output: \verbinclude MatrixBase_topRows_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the top rows of *this.
+///
+/// \param n the number of rows in the block
+///
+/// Example: \include MatrixBase_topRows_int.cpp
+/// Output: \verbinclude MatrixBase_topRows_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline RowsBlockXpr topRows(Index n)
 {
   return RowsBlockXpr(derived(), 0, 0, n, cols());
 }
 
-/** This is the const version of topRows(Index).*/
+/// This is the const version of topRows(Index).
+EIGEN_DEVICE_FUNC
 inline ConstRowsBlockXpr topRows(Index n) const
 {
   return ConstRowsBlockXpr(derived(), 0, 0, n, cols());
 }
 
-/** \returns a block consisting of the top rows of *this.
-  *
-  * \tparam N the number of rows in the block as specified at compile-time
-  * \param n the number of rows in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_topRows.cpp
-  * Output: \verbinclude MatrixBase_template_int_topRows.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the top rows of *this.
+///
+/// \tparam N the number of rows in the block as specified at compile-time
+/// \param n the number of rows in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_topRows.cpp
+/// Output: \verbinclude MatrixBase_template_int_topRows.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NRowsBlockXpr<N>::Type topRows(Index n = N)
 {
   return typename NRowsBlockXpr<N>::Type(derived(), 0, 0, n, cols());
 }
 
-/** This is the const version of topRows<int>().*/
+/// This is the const version of topRows<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNRowsBlockXpr<N>::Type topRows(Index n = N) const
 {
   return typename ConstNRowsBlockXpr<N>::Type(derived(), 0, 0, n, cols());
@@ -416,47 +473,55 @@
 
 
 
-/** \returns a block consisting of the bottom rows of *this.
-  *
-  * \param n the number of rows in the block
-  *
-  * Example: \include MatrixBase_bottomRows_int.cpp
-  * Output: \verbinclude MatrixBase_bottomRows_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the bottom rows of *this.
+///
+/// \param n the number of rows in the block
+///
+/// Example: \include MatrixBase_bottomRows_int.cpp
+/// Output: \verbinclude MatrixBase_bottomRows_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline RowsBlockXpr bottomRows(Index n)
 {
   return RowsBlockXpr(derived(), rows() - n, 0, n, cols());
 }
 
-/** This is the const version of bottomRows(Index).*/
+/// This is the const version of bottomRows(Index).
+EIGEN_DEVICE_FUNC
 inline ConstRowsBlockXpr bottomRows(Index n) const
 {
   return ConstRowsBlockXpr(derived(), rows() - n, 0, n, cols());
 }
 
-/** \returns a block consisting of the bottom rows of *this.
-  *
-  * \tparam N the number of rows in the block as specified at compile-time
-  * \param n the number of rows in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_bottomRows.cpp
-  * Output: \verbinclude MatrixBase_template_int_bottomRows.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the bottom rows of *this.
+///
+/// \tparam N the number of rows in the block as specified at compile-time
+/// \param n the number of rows in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_bottomRows.cpp
+/// Output: \verbinclude MatrixBase_template_int_bottomRows.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NRowsBlockXpr<N>::Type bottomRows(Index n = N)
 {
   return typename NRowsBlockXpr<N>::Type(derived(), rows() - n, 0, n, cols());
 }
 
-/** This is the const version of bottomRows<int>().*/
+/// This is the const version of bottomRows<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNRowsBlockXpr<N>::Type bottomRows(Index n = N) const
 {
   return typename ConstNRowsBlockXpr<N>::Type(derived(), rows() - n, 0, n, cols());
@@ -464,49 +529,57 @@
 
 
 
-/** \returns a block consisting of a range of rows of *this.
-  *
-  * \param startRow the index of the first row in the block
-  * \param n the number of rows in the block
-  *
-  * Example: \include DenseBase_middleRows_int.cpp
-  * Output: \verbinclude DenseBase_middleRows_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of rows of *this.
+///
+/// \param startRow the index of the first row in the block
+/// \param n the number of rows in the block
+///
+/// Example: \include DenseBase_middleRows_int.cpp
+/// Output: \verbinclude DenseBase_middleRows_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline RowsBlockXpr middleRows(Index startRow, Index n)
 {
   return RowsBlockXpr(derived(), startRow, 0, n, cols());
 }
 
-/** This is the const version of middleRows(Index,Index).*/
+/// This is the const version of middleRows(Index,Index).
+EIGEN_DEVICE_FUNC
 inline ConstRowsBlockXpr middleRows(Index startRow, Index n) const
 {
   return ConstRowsBlockXpr(derived(), startRow, 0, n, cols());
 }
 
-/** \returns a block consisting of a range of rows of *this.
-  *
-  * \tparam N the number of rows in the block as specified at compile-time
-  * \param startRow the index of the first row in the block
-  * \param n the number of rows in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include DenseBase_template_int_middleRows.cpp
-  * Output: \verbinclude DenseBase_template_int_middleRows.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of rows of *this.
+///
+/// \tparam N the number of rows in the block as specified at compile-time
+/// \param startRow the index of the first row in the block
+/// \param n the number of rows in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include DenseBase_template_int_middleRows.cpp
+/// Output: \verbinclude DenseBase_template_int_middleRows.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NRowsBlockXpr<N>::Type middleRows(Index startRow, Index n = N)
 {
   return typename NRowsBlockXpr<N>::Type(derived(), startRow, 0, n, cols());
 }
 
-/** This is the const version of middleRows<int>().*/
+/// This is the const version of middleRows<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNRowsBlockXpr<N>::Type middleRows(Index startRow, Index n = N) const
 {
   return typename ConstNRowsBlockXpr<N>::Type(derived(), startRow, 0, n, cols());
@@ -514,47 +587,55 @@
 
 
 
-/** \returns a block consisting of the left columns of *this.
-  *
-  * \param n the number of columns in the block
-  *
-  * Example: \include MatrixBase_leftCols_int.cpp
-  * Output: \verbinclude MatrixBase_leftCols_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the left columns of *this.
+///
+/// \param n the number of columns in the block
+///
+/// Example: \include MatrixBase_leftCols_int.cpp
+/// Output: \verbinclude MatrixBase_leftCols_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline ColsBlockXpr leftCols(Index n)
 {
   return ColsBlockXpr(derived(), 0, 0, rows(), n);
 }
 
-/** This is the const version of leftCols(Index).*/
+/// This is the const version of leftCols(Index).
+EIGEN_DEVICE_FUNC
 inline ConstColsBlockXpr leftCols(Index n) const
 {
   return ConstColsBlockXpr(derived(), 0, 0, rows(), n);
 }
 
-/** \returns a block consisting of the left columns of *this.
-  *
-  * \tparam N the number of columns in the block as specified at compile-time
-  * \param n the number of columns in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_leftCols.cpp
-  * Output: \verbinclude MatrixBase_template_int_leftCols.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the left columns of *this.
+///
+/// \tparam N the number of columns in the block as specified at compile-time
+/// \param n the number of columns in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_leftCols.cpp
+/// Output: \verbinclude MatrixBase_template_int_leftCols.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NColsBlockXpr<N>::Type leftCols(Index n = N)
 {
   return typename NColsBlockXpr<N>::Type(derived(), 0, 0, rows(), n);
 }
 
-/** This is the const version of leftCols<int>().*/
+/// This is the const version of leftCols<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNColsBlockXpr<N>::Type leftCols(Index n = N) const
 {
   return typename ConstNColsBlockXpr<N>::Type(derived(), 0, 0, rows(), n);
@@ -562,47 +643,55 @@
 
 
 
-/** \returns a block consisting of the right columns of *this.
-  *
-  * \param n the number of columns in the block
-  *
-  * Example: \include MatrixBase_rightCols_int.cpp
-  * Output: \verbinclude MatrixBase_rightCols_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the right columns of *this.
+///
+/// \param n the number of columns in the block
+///
+/// Example: \include MatrixBase_rightCols_int.cpp
+/// Output: \verbinclude MatrixBase_rightCols_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline ColsBlockXpr rightCols(Index n)
 {
   return ColsBlockXpr(derived(), 0, cols() - n, rows(), n);
 }
 
-/** This is the const version of rightCols(Index).*/
+/// This is the const version of rightCols(Index).
+EIGEN_DEVICE_FUNC
 inline ConstColsBlockXpr rightCols(Index n) const
 {
   return ConstColsBlockXpr(derived(), 0, cols() - n, rows(), n);
 }
 
-/** \returns a block consisting of the right columns of *this.
-  *
-  * \tparam N the number of columns in the block as specified at compile-time
-  * \param n the number of columns in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_rightCols.cpp
-  * Output: \verbinclude MatrixBase_template_int_rightCols.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of the right columns of *this.
+///
+/// \tparam N the number of columns in the block as specified at compile-time
+/// \param n the number of columns in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_rightCols.cpp
+/// Output: \verbinclude MatrixBase_template_int_rightCols.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NColsBlockXpr<N>::Type rightCols(Index n = N)
 {
   return typename NColsBlockXpr<N>::Type(derived(), 0, cols() - n, rows(), n);
 }
 
-/** This is the const version of rightCols<int>().*/
+/// This is the const version of rightCols<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNColsBlockXpr<N>::Type rightCols(Index n = N) const
 {
   return typename ConstNColsBlockXpr<N>::Type(derived(), 0, cols() - n, rows(), n);
@@ -610,49 +699,57 @@
 
 
 
-/** \returns a block consisting of a range of columns of *this.
-  *
-  * \param startCol the index of the first column in the block
-  * \param numCols the number of columns in the block
-  *
-  * Example: \include DenseBase_middleCols_int.cpp
-  * Output: \verbinclude DenseBase_middleCols_int.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of columns of *this.
+///
+/// \param startCol the index of the first column in the block
+/// \param numCols the number of columns in the block
+///
+/// Example: \include DenseBase_middleCols_int.cpp
+/// Output: \verbinclude DenseBase_middleCols_int.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline ColsBlockXpr middleCols(Index startCol, Index numCols)
 {
   return ColsBlockXpr(derived(), 0, startCol, rows(), numCols);
 }
 
-/** This is the const version of middleCols(Index,Index).*/
+/// This is the const version of middleCols(Index,Index).
+EIGEN_DEVICE_FUNC
 inline ConstColsBlockXpr middleCols(Index startCol, Index numCols) const
 {
   return ConstColsBlockXpr(derived(), 0, startCol, rows(), numCols);
 }
 
-/** \returns a block consisting of a range of columns of *this.
-  *
-  * \tparam N the number of columns in the block as specified at compile-time
-  * \param startCol the index of the first column in the block
-  * \param n the number of columns in the block as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include DenseBase_template_int_middleCols.cpp
-  * Output: \verbinclude DenseBase_template_int_middleCols.out
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
+/// \returns a block consisting of a range of columns of *this.
+///
+/// \tparam N the number of columns in the block as specified at compile-time
+/// \param startCol the index of the first column in the block
+/// \param n the number of columns in the block as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include DenseBase_template_int_middleCols.cpp
+/// Output: \verbinclude DenseBase_template_int_middleCols.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename NColsBlockXpr<N>::Type middleCols(Index startCol, Index n = N)
 {
   return typename NColsBlockXpr<N>::Type(derived(), 0, startCol, rows(), n);
 }
 
-/** This is the const version of middleCols<int>().*/
+/// This is the const version of middleCols<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstNColsBlockXpr<N>::Type middleCols(Index startCol, Index n = N) const
 {
   return typename ConstNColsBlockXpr<N>::Type(derived(), 0, startCol, rows(), n);
@@ -660,119 +757,134 @@
 
 
 
-/** \returns a fixed-size expression of a block in *this.
-  *
-  * The template parameters \a BlockRows and \a BlockCols are the number of
-  * rows and columns in the block.
-  *
-  * \param startRow the first row in the block
-  * \param startCol the first column in the block
-  *
-  * Example: \include MatrixBase_block_int_int.cpp
-  * Output: \verbinclude MatrixBase_block_int_int.out
-  *
-  * \note since block is a templated member, the keyword template has to be used
-  * if the matrix type is also a template parameter: \code m.template block<3,3>(1,1); \endcode
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-template<int BlockRows, int BlockCols>
-inline Block<Derived, BlockRows, BlockCols> block(Index startRow, Index startCol)
+/// \returns a fixed-size expression of a block in *this.
+///
+/// The template parameters \a NRows and \a NCols are the number of
+/// rows and columns in the block.
+///
+/// \param startRow the first row in the block
+/// \param startCol the first column in the block
+///
+/// Example: \include MatrixBase_block_int_int.cpp
+/// Output: \verbinclude MatrixBase_block_int_int.out
+///
+/// \note since block is a templated member, the keyword template has to be used
+/// if the matrix type is also a template parameter: \code m.template block<3,3>(1,1); \endcode
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+template<int NRows, int NCols>
+EIGEN_DEVICE_FUNC
+inline typename FixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol)
 {
-  return Block<Derived, BlockRows, BlockCols>(derived(), startRow, startCol);
+  return typename FixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol);
 }
 
-/** This is the const version of block<>(Index, Index). */
-template<int BlockRows, int BlockCols>
-inline const Block<const Derived, BlockRows, BlockCols> block(Index startRow, Index startCol) const
+/// This is the const version of block<>(Index, Index). */
+template<int NRows, int NCols>
+EIGEN_DEVICE_FUNC
+inline const typename ConstFixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol) const
 {
-  return Block<const Derived, BlockRows, BlockCols>(derived(), startRow, startCol);
+  return typename ConstFixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol);
 }
 
-/** \returns an expression of a block in *this.
-  *
-  * \tparam BlockRows number of rows in block as specified at compile-time
-  * \tparam BlockCols number of columns in block as specified at compile-time
-  * \param  startRow  the first row in the block
-  * \param  startCol  the first column in the block
-  * \param  blockRows number of rows in block as specified at run-time
-  * \param  blockCols number of columns in block as specified at run-time
-  *
-  * This function is mainly useful for blocks where the number of rows is specified at compile-time
-  * and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
-  * information should not contradict. In other words, \a blockRows should equal \a BlockRows unless
-  * \a BlockRows is \a Dynamic, and the same for the number of columns.
-  *
-  * Example: \include MatrixBase_template_int_int_block_int_int_int_int.cpp
-  * Output: \verbinclude MatrixBase_template_int_int_block_int_int_int_int.cpp
-  *
-  * \sa class Block, block(Index,Index,Index,Index)
-  */
-template<int BlockRows, int BlockCols>
-inline Block<Derived, BlockRows, BlockCols> block(Index startRow, Index startCol, 
+/// \returns an expression of a block in *this.
+///
+/// \tparam NRows number of rows in block as specified at compile-time
+/// \tparam NCols number of columns in block as specified at compile-time
+/// \param  startRow  the first row in the block
+/// \param  startCol  the first column in the block
+/// \param  blockRows number of rows in block as specified at run-time
+/// \param  blockCols number of columns in block as specified at run-time
+///
+/// This function is mainly useful for blocks where the number of rows is specified at compile-time
+/// and the number of columns is specified at run-time, or vice versa. The compile-time and run-time
+/// information should not contradict. In other words, \a blockRows should equal \a NRows unless
+/// \a NRows is \a Dynamic, and the same for the number of columns.
+///
+/// Example: \include MatrixBase_template_int_int_block_int_int_int_int.cpp
+/// Output: \verbinclude MatrixBase_template_int_int_block_int_int_int_int.cpp
+///
+EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL
+///
+/// \sa class Block, block(Index,Index,Index,Index)
+///
+template<int NRows, int NCols>
+inline typename FixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol,
                                                   Index blockRows, Index blockCols)
 {
-  return Block<Derived, BlockRows, BlockCols>(derived(), startRow, startCol, blockRows, blockCols);
+  return typename FixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol, blockRows, blockCols);
 }
 
-/** This is the const version of block<>(Index, Index, Index, Index). */
-template<int BlockRows, int BlockCols>
-inline const Block<const Derived, BlockRows, BlockCols> block(Index startRow, Index startCol,
+/// This is the const version of block<>(Index, Index, Index, Index).
+template<int NRows, int NCols>
+inline const typename ConstFixedBlockXpr<NRows,NCols>::Type block(Index startRow, Index startCol,
                                                               Index blockRows, Index blockCols) const
 {
-  return Block<const Derived, BlockRows, BlockCols>(derived(), startRow, startCol, blockRows, blockCols);
+  return typename ConstFixedBlockXpr<NRows,NCols>::Type(derived(), startRow, startCol, blockRows, blockCols);
 }
 
-/** \returns an expression of the \a i-th column of *this. Note that the numbering starts at 0.
-  *
-  * Example: \include MatrixBase_col.cpp
-  * Output: \verbinclude MatrixBase_col.out
-  *
+/// \returns an expression of the \a i-th column of *this. Note that the numbering starts at 0.
+///
+/// Example: \include MatrixBase_col.cpp
+/// Output: \verbinclude MatrixBase_col.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(column-major)
+/**
   * \sa row(), class Block */
+EIGEN_DEVICE_FUNC
 inline ColXpr col(Index i)
 {
   return ColXpr(derived(), i);
 }
 
-/** This is the const version of col(). */
+/// This is the const version of col().
+EIGEN_DEVICE_FUNC
 inline ConstColXpr col(Index i) const
 {
   return ConstColXpr(derived(), i);
 }
 
-/** \returns an expression of the \a i-th row of *this. Note that the numbering starts at 0.
-  *
-  * Example: \include MatrixBase_row.cpp
-  * Output: \verbinclude MatrixBase_row.out
-  *
+/// \returns an expression of the \a i-th row of *this. Note that the numbering starts at 0.
+///
+/// Example: \include MatrixBase_row.cpp
+/// Output: \verbinclude MatrixBase_row.out
+///
+EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(row-major)
+/**
   * \sa col(), class Block */
+EIGEN_DEVICE_FUNC
 inline RowXpr row(Index i)
 {
   return RowXpr(derived(), i);
 }
 
-/** This is the const version of row(). */
+/// This is the const version of row(). */
+EIGEN_DEVICE_FUNC
 inline ConstRowXpr row(Index i) const
 {
   return ConstRowXpr(derived(), i);
 }
 
-/** \returns a dynamic-size expression of a segment (i.e. a vector block) in *this.
-  *
-  * \only_for_vectors
-  *
-  * \param start the first coefficient in the segment
-  * \param n the number of coefficients in the segment
-  *
-  * Example: \include MatrixBase_segment_int_int.cpp
-  * Output: \verbinclude MatrixBase_segment_int_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size vector, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, segment(Index)
-  */
+/// \returns a dynamic-size expression of a segment (i.e. a vector block) in *this.
+///
+/// \only_for_vectors
+///
+/// \param start the first coefficient in the segment
+/// \param n the number of coefficients in the segment
+///
+/// Example: \include MatrixBase_segment_int_int.cpp
+/// Output: \verbinclude MatrixBase_segment_int_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size vector, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+/// \sa class Block, segment(Index)
+///
+EIGEN_DEVICE_FUNC
 inline SegmentReturnType segment(Index start, Index n)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
@@ -780,154 +892,165 @@
 }
 
 
-/** This is the const version of segment(Index,Index).*/
+/// This is the const version of segment(Index,Index).
+EIGEN_DEVICE_FUNC
 inline ConstSegmentReturnType segment(Index start, Index n) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return ConstSegmentReturnType(derived(), start, n);
 }
 
-/** \returns a dynamic-size expression of the first coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \param n the number of coefficients in the segment
-  *
-  * Example: \include MatrixBase_start_int.cpp
-  * Output: \verbinclude MatrixBase_start_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size vector, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, block(Index,Index)
-  */
+/// \returns a dynamic-size expression of the first coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \param n the number of coefficients in the segment
+///
+/// Example: \include MatrixBase_start_int.cpp
+/// Output: \verbinclude MatrixBase_start_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size vector, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+/// \sa class Block, block(Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline SegmentReturnType head(Index n)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return SegmentReturnType(derived(), 0, n);
 }
 
-/** This is the const version of head(Index).*/
+/// This is the const version of head(Index).
+EIGEN_DEVICE_FUNC
 inline ConstSegmentReturnType head(Index n) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return ConstSegmentReturnType(derived(), 0, n);
 }
 
-/** \returns a dynamic-size expression of the last coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \param n the number of coefficients in the segment
-  *
-  * Example: \include MatrixBase_end_int.cpp
-  * Output: \verbinclude MatrixBase_end_int.out
-  *
-  * \note Even though the returned expression has dynamic size, in the case
-  * when it is applied to a fixed-size vector, it inherits a fixed maximal size,
-  * which means that evaluating it does not cause a dynamic memory allocation.
-  *
-  * \sa class Block, block(Index,Index)
-  */
+/// \returns a dynamic-size expression of the last coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \param n the number of coefficients in the segment
+///
+/// Example: \include MatrixBase_end_int.cpp
+/// Output: \verbinclude MatrixBase_end_int.out
+///
+/// \note Even though the returned expression has dynamic size, in the case
+/// when it is applied to a fixed-size vector, it inherits a fixed maximal size,
+/// which means that evaluating it does not cause a dynamic memory allocation.
+///
+/// \sa class Block, block(Index,Index)
+///
+EIGEN_DEVICE_FUNC
 inline SegmentReturnType tail(Index n)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return SegmentReturnType(derived(), this->size() - n, n);
 }
 
-/** This is the const version of tail(Index).*/
+/// This is the const version of tail(Index).
+EIGEN_DEVICE_FUNC
 inline ConstSegmentReturnType tail(Index n) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return ConstSegmentReturnType(derived(), this->size() - n, n);
 }
 
-/** \returns a fixed-size expression of a segment (i.e. a vector block) in \c *this
-  *
-  * \only_for_vectors
-  *
-  * \tparam N the number of coefficients in the segment as specified at compile-time
-  * \param start the index of the first element in the segment
-  * \param n the number of coefficients in the segment as specified at compile-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_segment.cpp
-  * Output: \verbinclude MatrixBase_template_int_segment.out
-  *
-  * \sa class Block
-  */
+/// \returns a fixed-size expression of a segment (i.e. a vector block) in \c *this
+///
+/// \only_for_vectors
+///
+/// \tparam N the number of coefficients in the segment as specified at compile-time
+/// \param start the index of the first element in the segment
+/// \param n the number of coefficients in the segment as specified at compile-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_segment.cpp
+/// Output: \verbinclude MatrixBase_template_int_segment.out
+///
+/// \sa class Block
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename FixedSegmentReturnType<N>::Type segment(Index start, Index n = N)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename FixedSegmentReturnType<N>::Type(derived(), start, n);
 }
 
-/** This is the const version of segment<int>(Index).*/
+/// This is the const version of segment<int>(Index).
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstFixedSegmentReturnType<N>::Type segment(Index start, Index n = N) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename ConstFixedSegmentReturnType<N>::Type(derived(), start, n);
 }
 
-/** \returns a fixed-size expression of the first coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \tparam N the number of coefficients in the segment as specified at compile-time
-  * \param  n the number of coefficients in the segment as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_start.cpp
-  * Output: \verbinclude MatrixBase_template_int_start.out
-  *
-  * \sa class Block
-  */
+/// \returns a fixed-size expression of the first coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \tparam N the number of coefficients in the segment as specified at compile-time
+/// \param  n the number of coefficients in the segment as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_start.cpp
+/// Output: \verbinclude MatrixBase_template_int_start.out
+///
+/// \sa class Block
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename FixedSegmentReturnType<N>::Type head(Index n = N)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename FixedSegmentReturnType<N>::Type(derived(), 0, n);
 }
 
-/** This is the const version of head<int>().*/
+/// This is the const version of head<int>().
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstFixedSegmentReturnType<N>::Type head(Index n = N) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename ConstFixedSegmentReturnType<N>::Type(derived(), 0, n);
 }
 
-/** \returns a fixed-size expression of the last coefficients of *this.
-  *
-  * \only_for_vectors
-  *
-  * \tparam N the number of coefficients in the segment as specified at compile-time
-  * \param  n the number of coefficients in the segment as specified at run-time
-  *
-  * The compile-time and run-time information should not contradict. In other words,
-  * \a n should equal \a N unless \a N is \a Dynamic.
-  *
-  * Example: \include MatrixBase_template_int_end.cpp
-  * Output: \verbinclude MatrixBase_template_int_end.out
-  *
-  * \sa class Block
-  */
+/// \returns a fixed-size expression of the last coefficients of *this.
+///
+/// \only_for_vectors
+///
+/// \tparam N the number of coefficients in the segment as specified at compile-time
+/// \param  n the number of coefficients in the segment as specified at run-time
+///
+/// The compile-time and run-time information should not contradict. In other words,
+/// \a n should equal \a N unless \a N is \a Dynamic.
+///
+/// Example: \include MatrixBase_template_int_end.cpp
+/// Output: \verbinclude MatrixBase_template_int_end.out
+///
+/// \sa class Block
+///
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename FixedSegmentReturnType<N>::Type tail(Index n = N)
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
   return typename FixedSegmentReturnType<N>::Type(derived(), size() - n);
 }
 
-/** This is the const version of tail<int>.*/
+/// This is the const version of tail<int>.
 template<int N>
+EIGEN_DEVICE_FUNC
 inline typename ConstFixedSegmentReturnType<N>::Type tail(Index n = N) const
 {
   EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived)
diff --git a/Eigen/src/plugins/CMakeLists.txt b/Eigen/src/plugins/CMakeLists.txt
deleted file mode 100644
index 1a1d3ff..0000000
--- a/Eigen/src/plugins/CMakeLists.txt
+++ /dev/null
@@ -1,6 +0,0 @@
-FILE(GLOB Eigen_plugins_SRCS "*.h")
-
-INSTALL(FILES
-  ${Eigen_plugins_SRCS}
-  DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/plugins COMPONENT Devel
-  )
diff --git a/Eigen/src/plugins/CommonCwiseBinaryOps.h b/Eigen/src/plugins/CommonCwiseBinaryOps.h
index 688d224..8b6730e 100644
--- a/Eigen/src/plugins/CommonCwiseBinaryOps.h
+++ b/Eigen/src/plugins/CommonCwiseBinaryOps.h
@@ -1,7 +1,7 @@
 // This file is part of Eigen, a lightweight C++ template library
 // for linear algebra.
 //
-// Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr>
+// Copyright (C) 2008-2016 Gael Guennebaud <gael.guennebaud@inria.fr>
 // Copyright (C) 2006-2008 Benoit Jacob <jacob.benoit.1@gmail.com>
 //
 // This Source Code Form is subject to the terms of the Mozilla
@@ -16,7 +16,7 @@
   *
   * \sa class CwiseBinaryOp, operator-=()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(operator-,internal::scalar_difference_op)
+EIGEN_MAKE_CWISE_BINARY_OP(operator-,difference)
 
 /** \returns an expression of the sum of \c *this and \a other
   *
@@ -24,7 +24,7 @@
   *
   * \sa class CwiseBinaryOp, operator+=()
   */
-EIGEN_MAKE_CWISE_BINARY_OP(operator+,internal::scalar_sum_op)
+EIGEN_MAKE_CWISE_BINARY_OP(operator+,sum)
 
 /** \returns an expression of a custom coefficient-wise operator \a func of *this and \a other
   *
@@ -38,9 +38,78 @@
   * \sa class CwiseBinaryOp, operator+(), operator-(), cwiseProduct()
   */
 template<typename CustomBinaryOp, typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE const CwiseBinaryOp<CustomBinaryOp, const Derived, const OtherDerived>
 binaryExpr(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other, const CustomBinaryOp& func = CustomBinaryOp()) const
 {
   return CwiseBinaryOp<CustomBinaryOp, const Derived, const OtherDerived>(derived(), other.derived(), func);
 }
 
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP(operator*,product)
+#else
+/** \returns an expression of \c *this scaled by the scalar factor \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T>
+const CwiseBinaryOp<internal::scalar_product_op<Scalar,T>,Derived,Constant<T> > operator*(const T& scalar) const;
+/** \returns an expression of \a expr scaled by the scalar factor \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T> friend
+const CwiseBinaryOp<internal::scalar_product_op<T,Scalar>,Constant<T>,Derived> operator*(const T& scalar, const StorageBaseType& expr);
+#endif
+
+
+
+#ifndef EIGEN_PARSED_BY_DOXYGEN
+EIGEN_MAKE_SCALAR_BINARY_OP_ONTHERIGHT(operator/,quotient)
+#else
+/** \returns an expression of \c *this divided by the scalar value \a scalar
+  *
+  * \tparam T is the scalar type of \a scalar. It must be compatible with the scalar type of the given expression.
+  */
+template<typename T>
+const CwiseBinaryOp<internal::scalar_quotient_op<Scalar,T>,Derived,Constant<T> > operator/(const T& scalar) const;
+#endif
+
+/** \returns an expression of the coefficient-wise boolean \b and operator of \c *this and \a other
+  *
+  * \warning this operator is for expression of bool only.
+  *
+  * Example: \include Cwise_boolean_and.cpp
+  * Output: \verbinclude Cwise_boolean_and.out
+  *
+  * \sa operator||(), select()
+  */
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+inline const CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>
+operator&&(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+{
+  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
+                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
+  return CwiseBinaryOp<internal::scalar_boolean_and_op, const Derived, const OtherDerived>(derived(),other.derived());
+}
+
+/** \returns an expression of the coefficient-wise boolean \b or operator of \c *this and \a other
+  *
+  * \warning this operator is for expression of bool only.
+  *
+  * Example: \include Cwise_boolean_or.cpp
+  * Output: \verbinclude Cwise_boolean_or.out
+  *
+  * \sa operator&&(), select()
+  */
+template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
+inline const CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>
+operator||(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
+{
+  EIGEN_STATIC_ASSERT((internal::is_same<bool,Scalar>::value && internal::is_same<bool,typename OtherDerived::Scalar>::value),
+                      THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_OF_BOOL);
+  return CwiseBinaryOp<internal::scalar_boolean_or_op, const Derived, const OtherDerived>(derived(),other.derived());
+}
diff --git a/Eigen/src/plugins/CommonCwiseUnaryOps.h b/Eigen/src/plugins/CommonCwiseUnaryOps.h
index 08e931a..89f4faa 100644
--- a/Eigen/src/plugins/CommonCwiseUnaryOps.h
+++ b/Eigen/src/plugins/CommonCwiseUnaryOps.h
@@ -12,10 +12,6 @@
 
 #ifndef EIGEN_PARSED_BY_DOXYGEN
 
-/** \internal Represents a scalar multiple of an expression */
-typedef CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const Derived> ScalarMultipleReturnType;
-/** \internal Represents a quotient of an expression by a scalar*/
-typedef CwiseUnaryOp<internal::scalar_quotient1_op<Scalar>, const Derived> ScalarQuotient1ReturnType;
 /** \internal the return type of conjugate() */
 typedef typename internal::conditional<NumTraits<Scalar>::IsComplex,
                     const CwiseUnaryOp<internal::scalar_conjugate_op<Scalar>, const Derived>,
@@ -36,137 +32,132 @@
 /** \internal the return type of imag() */
 typedef CwiseUnaryView<internal::scalar_imag_ref_op<Scalar>, Derived> NonConstImagReturnType;
 
+typedef CwiseUnaryOp<internal::scalar_opposite_op<Scalar>, const Derived> NegativeReturnType;
+
 #endif // not EIGEN_PARSED_BY_DOXYGEN
 
-/** \returns an expression of the opposite of \c *this
-  */
-inline const CwiseUnaryOp<internal::scalar_opposite_op<typename internal::traits<Derived>::Scalar>, const Derived>
-operator-() const { return derived(); }
+/// \returns an expression of the opposite of \c *this
+///
+EIGEN_DOC_UNARY_ADDONS(operator-,opposite)
+///
+EIGEN_DEVICE_FUNC
+inline const NegativeReturnType
+operator-() const { return NegativeReturnType(derived()); }
 
 
-/** \returns an expression of \c *this scaled by the scalar factor \a scalar */
-inline const ScalarMultipleReturnType
-operator*(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_multiple_op<Scalar>, const Derived>
-    (derived(), internal::scalar_multiple_op<Scalar>(scalar));
-}
+template<class NewType> struct CastXpr { typedef typename internal::cast_return_type<Derived,const CwiseUnaryOp<internal::scalar_cast_op<Scalar, NewType>, const Derived> >::type Type; };
 
-#ifdef EIGEN_PARSED_BY_DOXYGEN
-const ScalarMultipleReturnType operator*(const RealScalar& scalar) const;
-#endif
-
-/** \returns an expression of \c *this divided by the scalar value \a scalar */
-inline const CwiseUnaryOp<internal::scalar_quotient1_op<typename internal::traits<Derived>::Scalar>, const Derived>
-operator/(const Scalar& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_quotient1_op<Scalar>, const Derived>
-    (derived(), internal::scalar_quotient1_op<Scalar>(scalar));
-}
-
-/** Overloaded for efficient real matrix times complex scalar value */
-inline const CwiseUnaryOp<internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >, const Derived>
-operator*(const std::complex<Scalar>& scalar) const
-{
-  return CwiseUnaryOp<internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >, const Derived>
-    (*static_cast<const Derived*>(this), internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >(scalar));
-}
-
-inline friend const ScalarMultipleReturnType
-operator*(const Scalar& scalar, const StorageBaseType& matrix)
-{ return matrix*scalar; }
-
-inline friend const CwiseUnaryOp<internal::scalar_multiple2_op<Scalar,std::complex<Scalar> >, const Derived>
-operator*(const std::complex<Scalar>& scalar, const StorageBaseType& matrix)
-{ return matrix*scalar; }
-
-/** \returns an expression of *this with the \a Scalar type casted to
-  * \a NewScalar.
-  *
-  * The template parameter \a NewScalar is the type we are casting the scalars to.
-  *
-  * \sa class CwiseUnaryOp
-  */
+/// \returns an expression of \c *this with the \a Scalar type casted to
+/// \a NewScalar.
+///
+/// The template parameter \a NewScalar is the type we are casting the scalars to.
+///
+EIGEN_DOC_UNARY_ADDONS(cast,conversion function)
+///
+/// \sa class CwiseUnaryOp
+///
 template<typename NewType>
-typename internal::cast_return_type<Derived,const CwiseUnaryOp<internal::scalar_cast_op<typename internal::traits<Derived>::Scalar, NewType>, const Derived> >::type
+EIGEN_DEVICE_FUNC
+typename CastXpr<NewType>::Type
 cast() const
 {
-  return derived();
+  return typename CastXpr<NewType>::Type(derived());
 }
 
-/** \returns an expression of the complex conjugate of \c *this.
-  *
-  * \sa adjoint() */
+/// \returns an expression of the complex conjugate of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(conjugate,complex conjugate)
+///
+/// \sa <a href="group__CoeffwiseMathFunctions.html#cwisetable_conj">Math functions</a>, MatrixBase::adjoint()
+EIGEN_DEVICE_FUNC
 inline ConjugateReturnType
 conjugate() const
 {
   return ConjugateReturnType(derived());
 }
 
-/** \returns a read-only expression of the real part of \c *this.
-  *
-  * \sa imag() */
+/// \returns a read-only expression of the real part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(real,real part function)
+///
+/// \sa imag()
+EIGEN_DEVICE_FUNC
 inline RealReturnType
-real() const { return derived(); }
+real() const { return RealReturnType(derived()); }
 
-/** \returns an read-only expression of the imaginary part of \c *this.
-  *
-  * \sa real() */
+/// \returns an read-only expression of the imaginary part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(imag,imaginary part function)
+///
+/// \sa real()
+EIGEN_DEVICE_FUNC
 inline const ImagReturnType
-imag() const { return derived(); }
+imag() const { return ImagReturnType(derived()); }
 
-/** \brief Apply a unary operator coefficient-wise
-  * \param[in]  func  Functor implementing the unary operator
-  * \tparam  CustomUnaryOp Type of \a func  
-  * \returns An expression of a custom coefficient-wise unary operator \a func of *this
-  *
-  * The function \c ptr_fun() from the C++ standard library can be used to make functors out of normal functions.
-  *
-  * Example:
-  * \include class_CwiseUnaryOp_ptrfun.cpp
-  * Output: \verbinclude class_CwiseUnaryOp_ptrfun.out
-  *
-  * Genuine functors allow for more possibilities, for instance it may contain a state.
-  *
-  * Example:
-  * \include class_CwiseUnaryOp.cpp
-  * Output: \verbinclude class_CwiseUnaryOp.out
-  *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp
-  */
+/// \brief Apply a unary operator coefficient-wise
+/// \param[in]  func  Functor implementing the unary operator
+/// \tparam  CustomUnaryOp Type of \a func
+/// \returns An expression of a custom coefficient-wise unary operator \a func of *this
+///
+/// The function \c ptr_fun() from the C++ standard library can be used to make functors out of normal functions.
+///
+/// Example:
+/// \include class_CwiseUnaryOp_ptrfun.cpp
+/// Output: \verbinclude class_CwiseUnaryOp_ptrfun.out
+///
+/// Genuine functors allow for more possibilities, for instance it may contain a state.
+///
+/// Example:
+/// \include class_CwiseUnaryOp.cpp
+/// Output: \verbinclude class_CwiseUnaryOp.out
+///
+EIGEN_DOC_UNARY_ADDONS(unaryExpr,unary function)
+///
+/// \sa unaryViewExpr, binaryExpr, class CwiseUnaryOp
+///
 template<typename CustomUnaryOp>
+EIGEN_DEVICE_FUNC
 inline const CwiseUnaryOp<CustomUnaryOp, const Derived>
 unaryExpr(const CustomUnaryOp& func = CustomUnaryOp()) const
 {
   return CwiseUnaryOp<CustomUnaryOp, const Derived>(derived(), func);
 }
 
-/** \returns an expression of a custom coefficient-wise unary operator \a func of *this
-  *
-  * The template parameter \a CustomUnaryOp is the type of the functor
-  * of the custom unary operator.
-  *
-  * Example:
-  * \include class_CwiseUnaryOp.cpp
-  * Output: \verbinclude class_CwiseUnaryOp.out
-  *
-  * \sa class CwiseUnaryOp, class CwiseBinaryOp
-  */
+/// \returns an expression of a custom coefficient-wise unary operator \a func of *this
+///
+/// The template parameter \a CustomUnaryOp is the type of the functor
+/// of the custom unary operator.
+///
+/// Example:
+/// \include class_CwiseUnaryOp.cpp
+/// Output: \verbinclude class_CwiseUnaryOp.out
+///
+EIGEN_DOC_UNARY_ADDONS(unaryViewExpr,unary function)
+///
+/// \sa unaryExpr, binaryExpr class CwiseUnaryOp
+///
 template<typename CustomViewOp>
+EIGEN_DEVICE_FUNC
 inline const CwiseUnaryView<CustomViewOp, const Derived>
 unaryViewExpr(const CustomViewOp& func = CustomViewOp()) const
 {
   return CwiseUnaryView<CustomViewOp, const Derived>(derived(), func);
 }
 
-/** \returns a non const expression of the real part of \c *this.
-  *
-  * \sa imag() */
+/// \returns a non const expression of the real part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(real,real part function)
+///
+/// \sa imag()
+EIGEN_DEVICE_FUNC
 inline NonConstRealReturnType
-real() { return derived(); }
+real() { return NonConstRealReturnType(derived()); }
 
-/** \returns a non const expression of the imaginary part of \c *this.
-  *
-  * \sa real() */
+/// \returns a non const expression of the imaginary part of \c *this.
+///
+EIGEN_DOC_UNARY_ADDONS(imag,imaginary part function)
+///
+/// \sa real()
+EIGEN_DEVICE_FUNC
 inline NonConstImagReturnType
-imag() { return derived(); }
+imag() { return NonConstImagReturnType(derived()); }
diff --git a/Eigen/src/plugins/MatrixCwiseBinaryOps.h b/Eigen/src/plugins/MatrixCwiseBinaryOps.h
index c4a042b..f1084ab 100644
--- a/Eigen/src/plugins/MatrixCwiseBinaryOps.h
+++ b/Eigen/src/plugins/MatrixCwiseBinaryOps.h
@@ -18,10 +18,11 @@
   * \sa class CwiseBinaryOp, cwiseAbs2
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)
 cwiseProduct(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return EIGEN_CWISE_PRODUCT_RETURN_TYPE(Derived,OtherDerived)(derived(), other.derived());
+  return EIGEN_CWISE_BINARY_RETURN_TYPE(Derived,OtherDerived,product)(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise == operator of *this and \a other
@@ -37,6 +38,7 @@
   * \sa cwiseNotEqual(), isApprox(), isMuchSmallerThan()
   */
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 inline const CwiseBinaryOp<std::equal_to<Scalar>, const Derived, const OtherDerived>
 cwiseEqual(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
@@ -56,6 +58,7 @@
   * \sa cwiseEqual(), isApprox(), isMuchSmallerThan()
   */
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 inline const CwiseBinaryOp<std::not_equal_to<Scalar>, const Derived, const OtherDerived>
 cwiseNotEqual(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
@@ -70,17 +73,19 @@
   * \sa class CwiseBinaryOp, max()
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived, const OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived, const OtherDerived>
 cwiseMin(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
+  return CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise min of *this and scalar \a other
   *
   * \sa class CwiseBinaryOp, min()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar>, const Derived, const ConstantReturnType>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_min_op<Scalar,Scalar>, const Derived, const ConstantReturnType>
 cwiseMin(const Scalar &other) const
 {
   return cwiseMin(Derived::Constant(rows(), cols(), other));
@@ -94,17 +99,19 @@
   * \sa class CwiseBinaryOp, min()
   */
 template<typename OtherDerived>
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived, const OtherDerived>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived, const OtherDerived>
 cwiseMax(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
-  return CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
+  return CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
 /** \returns an expression of the coefficient-wise max of *this and scalar \a other
   *
   * \sa class CwiseBinaryOp, min()
   */
-EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar>, const Derived, const ConstantReturnType>
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_max_op<Scalar,Scalar>, const Derived, const ConstantReturnType>
 cwiseMax(const Scalar &other) const
 {
   return cwiseMax(Derived::Constant(rows(), cols(), other));
@@ -119,13 +126,14 @@
   * \sa class CwiseBinaryOp, cwiseProduct(), cwiseInverse()
   */
 template<typename OtherDerived>
+EIGEN_DEVICE_FUNC
 EIGEN_STRONG_INLINE const CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>
 cwiseQuotient(const EIGEN_CURRENT_STORAGE_BASE_CLASS<OtherDerived> &other) const
 {
   return CwiseBinaryOp<internal::scalar_quotient_op<Scalar>, const Derived, const OtherDerived>(derived(), other.derived());
 }
 
-typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,internal::cmp_EQ>, const Derived, const ConstantReturnType> CwiseScalarEqualReturnType;
+typedef CwiseBinaryOp<internal::scalar_cmp_op<Scalar,Scalar,internal::cmp_EQ>, const Derived, const ConstantReturnType> CwiseScalarEqualReturnType;
 
 /** \returns an expression of the coefficient-wise == operator of \c *this and a scalar \a s
   *
@@ -136,8 +144,9 @@
   *
   * \sa cwiseEqual(const MatrixBase<OtherDerived> &) const
   */
+EIGEN_DEVICE_FUNC
 inline const CwiseScalarEqualReturnType
 cwiseEqual(const Scalar& s) const
 {
-  return CwiseScalarEqualReturnType(derived(), Derived::Constant(rows(), cols(), s), internal::scalar_cmp_op<Scalar,internal::cmp_EQ>());
+  return CwiseScalarEqualReturnType(derived(), Derived::Constant(rows(), cols(), s), internal::scalar_cmp_op<Scalar,Scalar,internal::cmp_EQ>());
 }
diff --git a/Eigen/src/plugins/MatrixCwiseUnaryOps.h b/Eigen/src/plugins/MatrixCwiseUnaryOps.h
index 8de1093..b1be3d5 100644
--- a/Eigen/src/plugins/MatrixCwiseUnaryOps.h
+++ b/Eigen/src/plugins/MatrixCwiseUnaryOps.h
@@ -8,45 +8,78 @@
 // Public License v. 2.0. If a copy of the MPL was not distributed
 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
 
-// This file is a base class plugin containing matrix specifics coefficient wise functions.
+// This file is included into the body of the base classes supporting matrix specific coefficient-wise functions.
+// This include MatrixBase and SparseMatrixBase.
 
-/** \returns an expression of the coefficient-wise absolute value of \c *this
-  *
-  * Example: \include MatrixBase_cwiseAbs.cpp
-  * Output: \verbinclude MatrixBase_cwiseAbs.out
-  *
-  * \sa cwiseAbs2()
-  */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived>
-cwiseAbs() const { return derived(); }
 
-/** \returns an expression of the coefficient-wise squared absolute value of \c *this
-  *
-  * Example: \include MatrixBase_cwiseAbs2.cpp
-  * Output: \verbinclude MatrixBase_cwiseAbs2.out
-  *
-  * \sa cwiseAbs()
-  */
-EIGEN_STRONG_INLINE const CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived>
-cwiseAbs2() const { return derived(); }
+typedef CwiseUnaryOp<internal::scalar_abs_op<Scalar>, const Derived> CwiseAbsReturnType;
+typedef CwiseUnaryOp<internal::scalar_abs2_op<Scalar>, const Derived> CwiseAbs2ReturnType;
+typedef CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived> CwiseSqrtReturnType;
+typedef CwiseUnaryOp<internal::scalar_sign_op<Scalar>, const Derived> CwiseSignReturnType;
+typedef CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived> CwiseInverseReturnType;
 
-/** \returns an expression of the coefficient-wise square root of *this.
-  *
-  * Example: \include MatrixBase_cwiseSqrt.cpp
-  * Output: \verbinclude MatrixBase_cwiseSqrt.out
-  *
-  * \sa cwisePow(), cwiseSquare()
-  */
-inline const CwiseUnaryOp<internal::scalar_sqrt_op<Scalar>, const Derived>
-cwiseSqrt() const { return derived(); }
+/// \returns an expression of the coefficient-wise absolute value of \c *this
+///
+/// Example: \include MatrixBase_cwiseAbs.cpp
+/// Output: \verbinclude MatrixBase_cwiseAbs.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseAbs,absolute value)
+///
+/// \sa cwiseAbs2()
+///
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseAbsReturnType
+cwiseAbs() const { return CwiseAbsReturnType(derived()); }
 
-/** \returns an expression of the coefficient-wise inverse of *this.
-  *
-  * Example: \include MatrixBase_cwiseInverse.cpp
-  * Output: \verbinclude MatrixBase_cwiseInverse.out
-  *
-  * \sa cwiseProduct()
-  */
-inline const CwiseUnaryOp<internal::scalar_inverse_op<Scalar>, const Derived>
-cwiseInverse() const { return derived(); }
+/// \returns an expression of the coefficient-wise squared absolute value of \c *this
+///
+/// Example: \include MatrixBase_cwiseAbs2.cpp
+/// Output: \verbinclude MatrixBase_cwiseAbs2.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseAbs2,squared absolute value)
+///
+/// \sa cwiseAbs()
+///
+EIGEN_DEVICE_FUNC
+EIGEN_STRONG_INLINE const CwiseAbs2ReturnType
+cwiseAbs2() const { return CwiseAbs2ReturnType(derived()); }
+
+/// \returns an expression of the coefficient-wise square root of *this.
+///
+/// Example: \include MatrixBase_cwiseSqrt.cpp
+/// Output: \verbinclude MatrixBase_cwiseSqrt.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseSqrt,square-root)
+///
+/// \sa cwisePow(), cwiseSquare()
+///
+EIGEN_DEVICE_FUNC
+inline const CwiseSqrtReturnType
+cwiseSqrt() const { return CwiseSqrtReturnType(derived()); }
+
+/// \returns an expression of the coefficient-wise signum of *this.
+///
+/// Example: \include MatrixBase_cwiseSign.cpp
+/// Output: \verbinclude MatrixBase_cwiseSign.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseSign,sign function)
+///
+EIGEN_DEVICE_FUNC
+inline const CwiseSignReturnType
+cwiseSign() const { return CwiseSignReturnType(derived()); }
+
+
+/// \returns an expression of the coefficient-wise inverse of *this.
+///
+/// Example: \include MatrixBase_cwiseInverse.cpp
+/// Output: \verbinclude MatrixBase_cwiseInverse.out
+///
+EIGEN_DOC_UNARY_ADDONS(cwiseInverse,inverse)
+///
+/// \sa cwiseProduct()
+///
+EIGEN_DEVICE_FUNC
+inline const CwiseInverseReturnType
+cwiseInverse() const { return CwiseInverseReturnType(derived()); }
+