Gitiles
Code Review Sign In
realtimeroboticsgroup.org / RealtimeRoboticsGroup / test / 9a24b370cbc07492fa6c205f1027367fa55b6742 / . / kernel / avx2 / kernel_dgemm_8x8_lib4.S
blob: 954c96d1620f852133e172d0ff45c765927b6818 [file] [log] [blame]
/**************************************************************************************************
* *
* This file is part of BLASFEO. *
* *
* BLASFEO -- BLAS For Embedded Optimization. *
* Copyright (C) 2016-2017 by Gianluca Frison. *
* Developed at IMTEK (University of Freiburg) under the supervision of Moritz Diehl. *
* All rights reserved. *
* *
* HPMPC is free software; you can redistribute it and/or *
* modify it under the terms of the GNU Lesser General Public *
* License as published by the Free Software Foundation; either *
* version 2.1 of the License, or (at your option) any later version. *
* *
* HPMPC is distributed in the hope that it will be useful, *
* but WITHOUT ANY WARRANTY; without even the implied warranty of *
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. *
* See the GNU Lesser General Public License for more details. *
* *
* You should have received a copy of the GNU Lesser General Public *
* License along with HPMPC; if not, write to the Free Software *
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA *
* *
* Author: Gianluca Frison, giaf (at) dtu.dk *
* gianluca.frison (at) imtek.uni-freiburg.de *
* *
**************************************************************************************************/
#if defined(OS_LINUX) | defined(OS_MAC)
//#define STACKSIZE 96
#define STACKSIZE 64
#define ARG1 %rdi
#define ARG2 %rsi
#define ARG3 %rdx
#define ARG4 %rcx
#define ARG5 %r8
#define ARG6 %r9
#define ARG7 STACKSIZE + 8(%rsp)
#define ARG8 STACKSIZE + 16(%rsp)
#define ARG9 STACKSIZE + 24(%rsp)
#define ARG10 STACKSIZE + 32(%rsp)
#define ARG11 STACKSIZE + 40(%rsp)
#define ARG12 STACKSIZE + 48(%rsp)
#define ARG13 STACKSIZE + 56(%rsp)
#define ARG14 STACKSIZE + 64(%rsp)
#define ARG15 STACKSIZE + 72(%rsp)
#define ARG16 STACKSIZE + 80(%rsp)
#define ARG17 STACKSIZE + 88(%rsp)
#define ARG18 STACKSIZE + 96(%rsp)
#define ARG19 STACKSIZE + 104(%rsp)
#define PROLOGUE \
subq $STACKSIZE, %rsp; \
movq %rbx, (%rsp); \
movq %rbp, 8(%rsp); \
movq %r12, 16(%rsp); \
movq %r13, 24(%rsp); \
movq %r14, 32(%rsp); \
movq %r15, 40(%rsp); \
vzeroupper;
#define EPILOGUE \
vzeroupper; \
movq (%rsp), %rbx; \
movq 8(%rsp), %rbp; \
movq 16(%rsp), %r12; \
movq 24(%rsp), %r13; \
movq 32(%rsp), %r14; \
movq 40(%rsp), %r15; \
addq $STACKSIZE, %rsp;
#elif defined(OS_WINDOWS)
#define STACKSIZE 256
#define ARG1 %rcx
#define ARG2 %rdx
#define ARG3 %r8
#define ARG4 %r9
#define ARG5 STACKSIZE + 40(%rsp)
#define ARG6 STACKSIZE + 48(%rsp)
#define ARG7 STACKSIZE + 56(%rsp)
#define ARG8 STACKSIZE + 64(%rsp)
#define ARG9 STACKSIZE + 72(%rsp)
#define ARG10 STACKSIZE + 80(%rsp)
#define ARG11 STACKSIZE + 88(%rsp)
#define ARG12 STACKSIZE + 96(%rsp)
#define ARG13 STACKSIZE + 104(%rsp)
#define ARG14 STACKSIZE + 112(%rsp)
#define ARG15 STACKSIZE + 120(%rsp)
#define ARG16 STACKSIZE + 128(%rsp)
#define ARG17 STACKSIZE + 136(%rsp)
#define ARG18 STACKSIZE + 144(%rsp)
#define ARG19 STACKSIZE + 152(%rsp)
#define PROLOGUE \
subq $STACKSIZE, %rsp; \
movq %rbx, (%rsp); \
movq %rbp, 8(%rsp); \
movq %r12, 16(%rsp); \
movq %r13, 24(%rsp); \
movq %r14, 32(%rsp); \
movq %r15, 40(%rsp); \
movq %rdi, 48(%rsp); \
movq %rsi, 56(%rsp); \
vmovups %xmm6, 64(%rsp); \
vmovups %xmm7, 80(%rsp); \
vmovups %xmm8, 96(%rsp); \
vmovups %xmm9, 112(%rsp); \
vmovups %xmm10, 128(%rsp); \
vmovups %xmm11, 144(%rsp); \
vmovups %xmm12, 160(%rsp); \
vmovups %xmm13, 176(%rsp); \
vmovups %xmm14, 192(%rsp); \
vmovups %xmm15, 208(%rsp); \
vzeroupper;
#define EPILOGUE \
vzeroupper; \
movq (%rsp), %rbx; \
movq 8(%rsp), %rbp; \
movq 16(%rsp), %r12; \
movq 24(%rsp), %r13; \
movq 32(%rsp), %r14; \
movq 40(%rsp), %r15; \
movq 48(%rsp), %rdi; \
movq 56(%rsp), %rsi; \
vmovups 64(%rsp), %xmm6; \
vmovups 80(%rsp), %xmm7; \
vmovups 96(%rsp), %xmm8; \
vmovups 112(%rsp), %xmm9; \
vmovups 128(%rsp), %xmm10; \
vmovups 144(%rsp), %xmm11; \
vmovups 160(%rsp), %xmm12; \
vmovups 176(%rsp), %xmm13; \
vmovups 192(%rsp), %xmm14; \
vmovups 208(%rsp), %xmm15; \
addq $STACKSIZE, %rsp;
#else
#error wrong OS
#endif
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.text
#elif defined(OS_MAC)
.section __TEXT,__text,regular,pure_instructions
#endif
// common inner routine with file scope
//
// input arguments:
// r10d <- k
// r11 <- A
// r12 <- 4*sda*sizeof(double)
// r13 <- B
// r14 <- 4*sdb*sizeof(double)
// r15 <- dirty
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- dirty
// ymm9 <- dirty
// ymm10 <- dirty
// ymm11 <- dirty
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10d <- 0
// r11 <- A+4*k*sizeof(double)
// r12 <- 4*sda*sizeof(double)
// r13 <- B+4*k*sizeof(double)
// r14 <- 4*sdb*sizeof(double)
// r15 <- dirty
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- dirty
// ymm9 <- dirty
// ymm10 <- dirty
// ymm11 <- dirty
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=2
.macro INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_kernel_dgemm_add_nt_8x8_lib4, @function
inner_kernel_dgemm_add_nt_8x8_lib4:
#elif defined(OS_MAC)
_inner_kernel_dgemm_add_nt_8x8_lib4:
#elif defined(OS_WINDOWS)
.def inner_kernel_dgemm_add_nt_8x8_lib4; .scl 2; .type 32; .endef
inner_kernel_dgemm_add_nt_8x8_lib4:
#endif
#endif
cmpl $0, %r10d
jle 2f // return
// preload
vmovapd 0(%r11), %ymm12
vmovapd 0(%r11, %r12, 1), %ymm13
vbroadcastsd 0(%r13), %ymm14
vbroadcastsd 0(%r13, %r14, 1), %ymm15
cmpl $4, %r10d
jle 0f // consider clean-up loop
// main loop
.p2align 3
1: // main loop
// unroll 0
vfmadd231pd %ymm12, %ymm14, %ymm0
subl $4, %r10d
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 8(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 8(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 16(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 16(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 24(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 24(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 32(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 32(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 32(%r11, %r12, 1), %ymm13
vbroadcastsd 32(%r13, %r14, 1), %ymm15
// unroll 1
vfmadd231pd %ymm12, %ymm14, %ymm0
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 40(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 40(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 48(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 48(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 56(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 56(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 64(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 64(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 64(%r11, %r12, 1), %ymm13
vbroadcastsd 64(%r13, %r14, 1), %ymm15
// unroll 2
vfmadd231pd %ymm12, %ymm14, %ymm0
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 72(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 72(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 80(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 80(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 88(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 88(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 96(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 96(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 96(%r11, %r12, 1), %ymm13
vbroadcastsd 96(%r13, %r14, 1), %ymm15
// unroll 3
vfmadd231pd %ymm12, %ymm14, %ymm0
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 104(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 104(%r13, %r14, 1), %ymm15
addq $128, %r11
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 112(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 112(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 120(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 120(%r13, %r14, 1), %ymm15
addq $128, %r13
vfmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 0(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 0(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 0(%r11, %r12, 1), %ymm13
vbroadcastsd 0(%r13, %r14, 1), %ymm15
cmpl $4, %r10d
jg 1b // main loop
0: // consider clean4-up
cmpl $3, %r10d
jle 4f // clean1
// unroll 0
vfmadd231pd %ymm12, %ymm14, %ymm0
subl $4, %r10d
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 8(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 8(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 16(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 16(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 24(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 24(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 32(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 32(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 32(%r11, %r12, 1), %ymm13
vbroadcastsd 32(%r13, %r14, 1), %ymm15
// unroll 1
vfmadd231pd %ymm12, %ymm14, %ymm0
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 40(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 40(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 48(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 48(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 56(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 56(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 64(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 64(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 64(%r11, %r12, 1), %ymm13
vbroadcastsd 64(%r13, %r14, 1), %ymm15
// unroll 2
vfmadd231pd %ymm12, %ymm14, %ymm0
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 72(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 72(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 80(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 80(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 88(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 88(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 96(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 96(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 96(%r11, %r12, 1), %ymm13
vbroadcastsd 96(%r13, %r14, 1), %ymm15
// unroll 3
vfmadd231pd %ymm12, %ymm14, %ymm0
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 104(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 104(%r13, %r14, 1), %ymm15
addq $128, %r11
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 112(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 112(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 120(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 120(%r13, %r14, 1), %ymm15
addq $128, %r13
vfmadd231pd %ymm12, %ymm14, %ymm3
// vmovapd 0(%r11), %ymm12
vfmadd231pd %ymm13, %ymm14, %ymm7
// vbroadcastsd 0(%r13), %ymm14
vfmadd231pd %ymm13, %ymm15, %ymm11
// vmovapd 0(%r11, %r12, 1), %ymm13
// vbroadcastsd 0(%r13, %r14, 1), %ymm15
jmp 2f
4: // consider clean1-up loop
cmpl $0, %r10d
jle 2f // return
// clean-up loop
3: // clean up loop
// unroll 0
vmovapd 0(%r11), %ymm12
vmovapd 0(%r11, %r12, 1), %ymm13
vbroadcastsd 0(%r13), %ymm14
vfmadd231pd %ymm12, %ymm14, %ymm0
vfmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 0(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm13, %ymm15, %ymm8
subl $1, %r10d
vbroadcastsd 8(%r13), %ymm14
vfmadd231pd %ymm12, %ymm14, %ymm1
vfmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 8(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm13, %ymm15, %ymm9
addq $32, %r11
vbroadcastsd 16(%r13), %ymm14
vfmadd231pd %ymm12, %ymm14, %ymm2
vfmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 16(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm13, %ymm15, %ymm10
addq $32, %r13
vbroadcastsd -8(%r13), %ymm14
vfmadd231pd %ymm12, %ymm14, %ymm3
vfmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd -8(%r13, %r14, 1), %ymm15
vfmadd231pd %ymm13, %ymm15, %ymm11
cmpl $0, %r10d
jg 3b // clean up loop
2: // return
#if MACRO_LEVEL>=2
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_kernel_dgemm_add_nt_8x8_lib4, .-inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// common inner routine with file scope
//
// input arguments:
// r10d <- k
// r11 <- A
// r12 <- 4*sda*sizeof(double)
// r13 <- B
// r14 <- 4*sdb*sizeof(double)
// r15 <- dirty
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- dirty
// ymm9 <- dirty
// ymm10 <- dirty
// ymm11 <- dirty
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10d <- 0
// r11 <- A+4*k*sizeof(double)
// r12 <- 4*sda*sizeof(double)
// r13 <- B+4*k*sizeof(double)
// r14 <- 4*sdb*sizeof(double)
// r15 <- dirty
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- dirty
// ymm9 <- dirty
// ymm10 <- dirty
// ymm11 <- dirty
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=2
.macro INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_kernel_dgemm_sub_nt_8x8_lib4, @function
inner_kernel_dgemm_sub_nt_8x8_lib4:
#elif defined(OS_MAC)
_inner_kernel_dgemm_sub_nt_8x8_lib4:
#elif defined(OS_WINDOWS)
.def inner_kernel_dgemm_sub_nt_8x8_lib4; .scl 2; .type 32; .endef
inner_kernel_dgemm_sub_nt_8x8_lib4:
#endif
#endif
cmpl $0, %r10d
jle 2f // return
// preload
vmovapd 0(%r11), %ymm12
vmovapd 0(%r11, %r12, 1), %ymm13
vbroadcastsd 0(%r13), %ymm14
vbroadcastsd 0(%r13, %r14, 1), %ymm15
cmpl $4, %r10d
jle 0f // consider clean-up loop
// main loop
.p2align 3
1: // main loop
// unroll 0
vfnmadd231pd %ymm12, %ymm14, %ymm0
subl $4, %r10d
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 8(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 8(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 16(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 16(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 24(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 24(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 32(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 32(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 32(%r11, %r12, 1), %ymm13
vbroadcastsd 32(%r13, %r14, 1), %ymm15
// unroll 1
vfnmadd231pd %ymm12, %ymm14, %ymm0
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 40(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 40(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 48(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 48(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 56(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 56(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 64(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 64(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 64(%r11, %r12, 1), %ymm13
vbroadcastsd 64(%r13, %r14, 1), %ymm15
// unroll 2
vfnmadd231pd %ymm12, %ymm14, %ymm0
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 72(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 72(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 80(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 80(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 88(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 88(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 96(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 96(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 96(%r11, %r12, 1), %ymm13
vbroadcastsd 96(%r13, %r14, 1), %ymm15
// unroll 3
vfnmadd231pd %ymm12, %ymm14, %ymm0
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 104(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 104(%r13, %r14, 1), %ymm15
addq $128, %r11
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 112(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 112(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 120(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 120(%r13, %r14, 1), %ymm15
addq $128, %r13
vfnmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 0(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 0(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 0(%r11, %r12, 1), %ymm13
vbroadcastsd 0(%r13, %r14, 1), %ymm15
cmpl $4, %r10d
jg 1b // main loop
0: // consider clean4-up
cmpl $3, %r10d
jle 4f // clean1
// unroll 0
vfnmadd231pd %ymm12, %ymm14, %ymm0
subl $4, %r10d
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 8(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 8(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 16(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 16(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 24(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 24(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 32(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 32(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 32(%r11, %r12, 1), %ymm13
vbroadcastsd 32(%r13, %r14, 1), %ymm15
// unroll 1
vfnmadd231pd %ymm12, %ymm14, %ymm0
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 40(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 40(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 48(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 48(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 56(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 56(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 64(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 64(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 64(%r11, %r12, 1), %ymm13
vbroadcastsd 64(%r13, %r14, 1), %ymm15
// unroll 2
vfnmadd231pd %ymm12, %ymm14, %ymm0
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 72(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 72(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 80(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 80(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 88(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 88(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm3
vmovapd 96(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd 96(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
vmovapd 96(%r11, %r12, 1), %ymm13
vbroadcastsd 96(%r13, %r14, 1), %ymm15
// unroll 3
vfnmadd231pd %ymm12, %ymm14, %ymm0
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 104(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm8
vbroadcastsd 104(%r13, %r14, 1), %ymm15
addq $128, %r11
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 112(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm9
vbroadcastsd 112(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 120(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm10
vbroadcastsd 120(%r13, %r14, 1), %ymm15
addq $128, %r13
vfnmadd231pd %ymm12, %ymm14, %ymm3
// vmovapd 0(%r11), %ymm12
vfnmadd231pd %ymm13, %ymm14, %ymm7
// vbroadcastsd 0(%r13), %ymm14
vfnmadd231pd %ymm13, %ymm15, %ymm11
// vmovapd 0(%r11, %r12, 1), %ymm13
// vbroadcastsd 0(%r13, %r14, 1), %ymm15
jmp 2f
4: // consider clean1-up loop
cmpl $0, %r10d
jle 2f // return
// clean-up loop
3: // clean up loop
// unroll 0
vmovapd 0(%r11), %ymm12
vmovapd 0(%r11, %r12, 1), %ymm13
vbroadcastsd 0(%r13), %ymm14
vfnmadd231pd %ymm12, %ymm14, %ymm0
vfnmadd231pd %ymm13, %ymm14, %ymm4
vbroadcastsd 0(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm13, %ymm15, %ymm8
subl $1, %r10d
vbroadcastsd 8(%r13), %ymm14
vfnmadd231pd %ymm12, %ymm14, %ymm1
vfnmadd231pd %ymm13, %ymm14, %ymm5
vbroadcastsd 8(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm13, %ymm15, %ymm9
addq $32, %r11
vbroadcastsd 16(%r13), %ymm14
vfnmadd231pd %ymm12, %ymm14, %ymm2
vfnmadd231pd %ymm13, %ymm14, %ymm6
vbroadcastsd 16(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm13, %ymm15, %ymm10
addq $32, %r13
vbroadcastsd -8(%r13), %ymm14
vfnmadd231pd %ymm12, %ymm14, %ymm3
vfnmadd231pd %ymm13, %ymm14, %ymm7
vbroadcastsd -8(%r13, %r14, 1), %ymm15
vfnmadd231pd %ymm13, %ymm15, %ymm11
cmpl $0, %r10d
jg 3b // clean up loop
2: // return
#if MACRO_LEVEL>=2
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_kernel_dgemm_sub_nt_8x8_lib4, .-inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// common inner routine with file scope
//
// scale for generic alpha and beta
//
// input arguments:
// r10 <- &alpha
// r11 <- &beta
// r12 <- C
// r13 <- 4*sdc*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- [d80 d91 da2 db3]
// ymm9 <- [d81 d90 da3 db2]
// ymm10 <- [d83 d92 da1 db0]
// ymm11 <- [d82 d93 da0 db1]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10 <- &alpha
// r11 <- &beta
// r12 <- C
// r13 <- 4*sdc*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_SCALE_AB_8X8_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_scale_ab_8x8_lib4, @function
inner_scale_ab_8x8_lib4:
#elif defined(OS_MAC)
_inner_scale_ab_8x8_lib4:
#elif defined(OS_WINDOWS)
.def inner_scale_ab_8x8_lib4; .scl 2; .type 32; .endef
inner_scale_ab_8x8_lib4:
#endif
#endif
vbroadcastsd 0(%r10), %ymm15 // alpha
vmulpd %ymm0, %ymm15, %ymm0
vmulpd %ymm1, %ymm15, %ymm1
vmulpd %ymm2, %ymm15, %ymm2
vmulpd %ymm3, %ymm15, %ymm3
vmulpd %ymm4, %ymm15, %ymm4
vmulpd %ymm5, %ymm15, %ymm5
vmulpd %ymm6, %ymm15, %ymm6
vmulpd %ymm7, %ymm15, %ymm7
vmulpd %ymm8, %ymm15, %ymm8
vmulpd %ymm9, %ymm15, %ymm9
vmulpd %ymm10, %ymm15, %ymm10
vmulpd %ymm11, %ymm15, %ymm11
vbroadcastsd 0(%r11), %ymm14 // beta
vxorpd %ymm15, %ymm15, %ymm15 // 0.0
vucomisd %xmm15, %xmm14 // beta==0.0 ?
je 0f // end
vmovapd 0(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm0
vmovapd 32(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm1
vmovapd 64(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm2
vmovapd 96(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm3
vmovapd 0(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm4
vmovapd 32(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm5
vmovapd 64(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm6
vmovapd 96(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm7
vmovapd 128(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm8
vmovapd 160(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm9
vmovapd 192(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm10
vmovapd 224(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm11
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_scale_ab_8x8_lib4, .-inner_scale_ab_8x8_lib4
#endif
#endif
// common inner routine with file scope
//
// scale for generic alpha and beta
//
// input arguments:
// r10 <- &alpha
// r11 <- &beta
// r12 <- C
// r13 <- 4*sdc*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- [d80 d91 da2 db3]
// ymm9 <- [d81 d90 da3 db2]
// ymm10 <- [d83 d92 da1 db0]
// ymm11 <- [d82 d93 da0 db1]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10 <- &alpha
// r11 <- &beta
// r12 <- C
// r13 <- 4*sdc*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_TRAN_SCALE_AB_8X8_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_tran_scale_ab_8x8_lib4, @function
inner_tran_scale_ab_8x8_lib4:
#elif defined(OS_MAC)
_inner_tran_scale_ab_8x8_lib4:
#elif defined(OS_WINDOWS)
.def inner_tran_scale_ab_8x8_lib4; .scl 2; .type 32; .endef
inner_tran_scale_ab_8x8_lib4:
#endif
#endif
vunpcklpd %ymm1, %ymm0, %ymm12
vunpckhpd %ymm1, %ymm0, %ymm13
vunpcklpd %ymm3, %ymm2, %ymm14
vunpckhpd %ymm3, %ymm2, %ymm15
vperm2f128 $0x20, %ymm14, %ymm12, %ymm0
vperm2f128 $0x31, %ymm14, %ymm12, %ymm2
vperm2f128 $0x20, %ymm15, %ymm13, %ymm1
vperm2f128 $0x31, %ymm15, %ymm13, %ymm3
vbroadcastsd 0(%r10), %ymm15 // alpha
vmulpd %ymm0, %ymm15, %ymm0
vmulpd %ymm1, %ymm15, %ymm1
vmulpd %ymm2, %ymm15, %ymm2
vmulpd %ymm3, %ymm15, %ymm3
vunpcklpd %ymm5, %ymm4, %ymm12
vunpckhpd %ymm5, %ymm4, %ymm13
vunpcklpd %ymm7, %ymm6, %ymm14
vunpckhpd %ymm7, %ymm6, %ymm15
vperm2f128 $0x20, %ymm14, %ymm12, %ymm4
vperm2f128 $0x31, %ymm14, %ymm12, %ymm6
vperm2f128 $0x20, %ymm15, %ymm13, %ymm5
vperm2f128 $0x31, %ymm15, %ymm13, %ymm7
vbroadcastsd 0(%r10), %ymm15 // alpha
vmulpd %ymm4, %ymm15, %ymm4
vmulpd %ymm5, %ymm15, %ymm5
vmulpd %ymm6, %ymm15, %ymm6
vmulpd %ymm7, %ymm15, %ymm7
vunpcklpd %ymm9, %ymm8, %ymm12
vunpckhpd %ymm9, %ymm8, %ymm13
vunpcklpd %ymm11, %ymm10, %ymm14
vunpckhpd %ymm11, %ymm10, %ymm15
vperm2f128 $0x20, %ymm14, %ymm12, %ymm8
vperm2f128 $0x31, %ymm14, %ymm12, %ymm10
vperm2f128 $0x20, %ymm15, %ymm13, %ymm9
vperm2f128 $0x31, %ymm15, %ymm13, %ymm11
vbroadcastsd 0(%r10), %ymm15 // alpha
vmulpd %ymm8, %ymm15, %ymm8
vmulpd %ymm9, %ymm15, %ymm9
vmulpd %ymm10, %ymm15, %ymm10
vmulpd %ymm11, %ymm15, %ymm11
vbroadcastsd 0(%r11), %ymm14 // beta
vxorpd %ymm15, %ymm15, %ymm15 // 0.0
vucomisd %xmm15, %xmm14 // beta==0.0 ?
je 0f // end
vmovapd 0(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm0
vmovapd 32(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm1
vmovapd 64(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm2
vmovapd 96(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm3
vmovapd 128(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm4
vmovapd 160(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm5
vmovapd 192(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm6
vmovapd 224(%r12), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm7
vmovapd 128(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm8
vmovapd 160(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm9
vmovapd 192(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm10
vmovapd 224(%r12, %r13, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm11
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_tran_scale_ab_8x8_lib4, .-inner_tran_scale_ab_8x8_lib4
#endif
#endif
// common inner routine with file scope
//
// scale for alpha=1.0 and beta=1.0
//
// input arguments:
// r10 <- C
// r11 <- 4*sdc*sizeof(double)
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- [d80 d91 da2 db3]
// ymm9 <- [d81 d90 da3 db2]
// ymm10 <- [d83 d92 da1 db0]
// ymm11 <- [d82 d93 da0 db1]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10 <- C
// r11 <- 4*sdc*sizeof(double)
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_SCALE_11_8X8_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_scale_11_8x8_lib4, @function
inner_scale_11_8x8_lib4:
#elif defined(OS_MAC)
_inner_scale_11_8x8_lib4:
#elif defined(OS_WINDOWS)
.def inner_scale_11_8x8_lib4; .scl 2; .type 32; .endef
inner_scale_11_8x8_lib4:
#endif
#endif
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vmovapd .LC04(%rip), %ymm14 // beta=1.0
#else
vmovapd LC04(%rip), %ymm14 // beta=1.0
#endif
vmovapd 0(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm0
vmovapd 32(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm1
vmovapd 64(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm2
vmovapd 96(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm3
vmovapd 0(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm4
vmovapd 32(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm5
vmovapd 64(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm6
vmovapd 96(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm7
vmovapd 128(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm8
vmovapd 160(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm9
vmovapd 192(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm10
vmovapd 224(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm11
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_scale_11_8x8_lib4, .-inner_scale_11_8x8_lib4
#endif
#endif
// common inner routine with file scope
//
// scale for alpha=1.0 and beta=1.0
//
// input arguments:
// r10 <- C
// r11 <- 4*sdc*sizeof(double)
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm8 <- [d80 d91 da2 db3]
// ymm9 <- [d81 d90 da3 db2]
// ymm10 <- [d83 d92 da1 db0]
// ymm11 <- [d82 d93 da0 db1]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10 <- C
// r11 <- 4*sdc*sizeof(double)
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_TRAN_SCALE_11_8X8_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_tran_scale_11_8x8_lib4, @function
inner_tran_scale_11_8x8_lib4:
#elif defined(OS_MAC)
_inner_tran_scale_11_8x8_lib4:
#elif defined(OS_WINDOWS)
.def inner_tran_scale_11_8x8_lib4; .scl 2; .type 32; .endef
inner_tran_scale_11_8x8_lib4:
#endif
#endif
vunpcklpd %ymm1, %ymm0, %ymm12
vunpckhpd %ymm1, %ymm0, %ymm13
vunpcklpd %ymm3, %ymm2, %ymm14
vunpckhpd %ymm3, %ymm2, %ymm15
vperm2f128 $0x20, %ymm14, %ymm12, %ymm0
vperm2f128 $0x31, %ymm14, %ymm12, %ymm2
vperm2f128 $0x20, %ymm15, %ymm13, %ymm1
vperm2f128 $0x31, %ymm15, %ymm13, %ymm3
vunpcklpd %ymm5, %ymm4, %ymm12
vunpckhpd %ymm5, %ymm4, %ymm13
vunpcklpd %ymm7, %ymm6, %ymm14
vunpckhpd %ymm7, %ymm6, %ymm15
vperm2f128 $0x20, %ymm14, %ymm12, %ymm4
vperm2f128 $0x31, %ymm14, %ymm12, %ymm6
vperm2f128 $0x20, %ymm15, %ymm13, %ymm5
vperm2f128 $0x31, %ymm15, %ymm13, %ymm7
vunpcklpd %ymm9, %ymm8, %ymm12
vunpckhpd %ymm9, %ymm8, %ymm13
vunpcklpd %ymm11, %ymm10, %ymm14
vunpckhpd %ymm11, %ymm10, %ymm15
vperm2f128 $0x20, %ymm14, %ymm12, %ymm8
vperm2f128 $0x31, %ymm14, %ymm12, %ymm10
vperm2f128 $0x20, %ymm15, %ymm13, %ymm9
vperm2f128 $0x31, %ymm15, %ymm13, %ymm11
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vmovapd .LC04(%rip), %ymm14 // beta=1.0
#else
vmovapd LC04(%rip), %ymm14 // beta=1.0
#endif
vmovapd 0(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm0
vmovapd 32(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm1
vmovapd 64(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm2
vmovapd 96(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm3
vmovapd 128(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm4
vmovapd 160(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm5
vmovapd 192(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm6
vmovapd 224(%r10), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm7
vmovapd 128(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm8
vmovapd 160(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm9
vmovapd 192(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm10
vmovapd 224(%r10, %r11, 1), %ymm15
vfmadd231pd %ymm14, %ymm15, %ymm11
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_tran_scale_11_8x8_lib4, .-inner_tran_scale_11_8x8_lib4
#endif
#endif
// common inner routine with file scope
//
// cholesky factorization
//
// input arguments:
// r10 <- inv_diag_E
// r11d <- kn
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10 <- inv_diag_E
// r11d <- kn
// ymm12 <- dirty
// ymm13 <- dirty
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_EDGE_DPOTRF_8X8_VS_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_edge_dpotrf_8x8_vs_lib4, @function
inner_edge_dpotrf_8x8_vs_lib4:
#elif defined(OS_MAC)
_inner_edge_dpotrf_8x8_vs_lib4:
#elif defined(OS_WINDOWS)
.def inner_edge_dpotrf_8x8_vs_lib4; .scl 2; .type 32; .endef
inner_edge_dpotrf_8x8_vs_lib4:
#endif
#endif
vxorpd %ymm15, %ymm15, %ymm15 // 0.0
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vmovsd .LC04(%rip), %xmm14 // 1.0
#elif defined(OS_MAC)
vmovsd LC04(%rip), %xmm14 // 1.0
#endif
vmovsd %xmm0, %xmm0, %xmm13
vucomisd %xmm15, %xmm13 // d_00 > 0.0 ?
jbe 1f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
2:
vmovsd %xmm13, 0(%r10)
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm0, %ymm13, %ymm0
vmulpd %ymm4, %ymm13, %ymm4
vpermpd $0x55, %ymm0, %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm1
vfnmadd231pd %ymm4, %ymm13, %ymm5
vperm2f128 $0x11, %ymm0, %ymm0, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm2
vfnmadd231pd %ymm4, %ymm13, %ymm6
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm3
vfnmadd231pd %ymm4, %ymm13, %ymm7
vperm2f128 $0x00, %ymm4, %ymm4, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm8
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm9
vperm2f128 $0x11, %ymm4, %ymm4, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm10
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm11
vpermilpd $0x3, %xmm1, %xmm13
vucomisd %xmm15, %xmm13 // d_11 > 0.0 ?
jbe 3f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
4:
vmovsd %xmm13, 8(%r10)
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm1, %ymm13, %ymm1
vmulpd %ymm5, %ymm13, %ymm5
vperm2f128 $0x11, %ymm1, %ymm1, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm2
vfnmadd231pd %ymm5, %ymm13, %ymm6
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm3
vfnmadd231pd %ymm5, %ymm13, %ymm7
vperm2f128 $0x00, %ymm5, %ymm5, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm8
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm9
vperm2f128 $0x11, %ymm5, %ymm5, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm10
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm11
vextractf128 $0x1, %ymm2, %xmm13
vucomisd %xmm15, %xmm13 // d_22 > 0.0 ?
jbe 5f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
6:
vmovsd %xmm13, 16(%r10)
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm2, %ymm13, %ymm2
vmulpd %ymm6, %ymm13, %ymm6
vpermpd $0xff, %ymm2, %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm3
vfnmadd231pd %ymm6, %ymm13, %ymm7
vperm2f128 $0x00, %ymm6, %ymm6, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm8
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm9
vperm2f128 $0x11, %ymm6, %ymm6, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm10
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm11
vpermpd $0xff, %ymm3, %ymm13
vucomisd %xmm15, %xmm13 // d_33 > 0.0 ?
jbe 7f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
8:
vmovsd %xmm13, 24(%r10)
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm3, %ymm13, %ymm3
vmulpd %ymm7, %ymm13, %ymm7
vperm2f128 $0x00, %ymm7, %ymm7, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm8
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm9
vperm2f128 $0x11, %ymm7, %ymm7, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm10
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm11
vmovsd %xmm8, %xmm8, %xmm13
vucomisd %xmm15, %xmm13 // d_00 > 0.0 ?
jbe 9f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
10:
vmovsd %xmm13, 32(%r10)
// vmovddup %xmm13, %xmm13
// vperm2f128 $0x00, %ymm13, %ymm13, %ymm13
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm8, %ymm13, %ymm8
cmpl $6, %r11d
jl 0f // ret
// vperm2f128 $0x00, %ymm8, %ymm8, %ymm12
// vpermilpd $0xf, %ymm12, %ymm13
vpermpd $0x55, %ymm8, %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm9
vperm2f128 $0x11, %ymm8, %ymm8, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm10
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm11
vpermilpd $0x3, %xmm9, %xmm13
vucomisd %xmm15, %xmm13 // d_11 > 0.0 ?
jbe 11f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
12:
vmovsd %xmm13, 40(%r10)
// vmovddup %xmm13, %xmm13
// vperm2f128 $0x00, %ymm13, %ymm13, %ymm13
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm9, %ymm13, %ymm9
cmpl $7, %r11d
jl 0f // ret
vperm2f128 $0x11, %ymm9, %ymm9, %ymm12
vpermilpd $0x0, %ymm12, %ymm13
vfnmadd231pd %ymm9, %ymm13, %ymm10
vpermilpd $0xf, %ymm12, %ymm13
vfnmadd231pd %ymm9, %ymm13, %ymm11
vextractf128 $0x1, %ymm10, %xmm13
vucomisd %xmm15, %xmm13 // d_22 > 0.0 ?
jbe 13f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
14:
vmovsd %xmm13, 48(%r10)
// vmovddup %xmm13, %xmm13
// vperm2f128 $0x00, %ymm13, %ymm13, %ymm13
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm10, %ymm13, %ymm10
cmpl $8, %r11d
jl 0f // ret
// vperm2f128 $0x11, %ymm10, %ymm10, %ymm12
// vpermilpd $0xf, %ymm12, %ymm13
vpermpd $0xff, %ymm10, %ymm13
vfnmadd231pd %ymm10, %ymm13, %ymm11
// vextractf128 $0x1, %ymm11, %xmm13
// vpermilpd $0x3, %xmm13, %xmm13
vpermpd $0xff, %ymm11, %ymm13
vucomisd %xmm15, %xmm13 // d_33 > 0.0 ?
jbe 15f
vsqrtsd %xmm13, %xmm13, %xmm13
vdivsd %xmm13, %xmm14, %xmm13
16:
vmovsd %xmm13, 56(%r10)
// vmovddup %xmm13, %xmm13
// vperm2f128 $0x00, %ymm13, %ymm13, %ymm13
vpermpd $0x00, %ymm13, %ymm13
vmulpd %ymm11, %ymm13, %ymm11
jmp 0f
1:
vxorpd %ymm13, %ymm13, %ymm13
jmp 2b
3:
vxorpd %ymm13, %ymm13, %ymm13
jmp 4b
5:
vxorpd %ymm13, %ymm13, %ymm13
jmp 6b
7:
vxorpd %ymm13, %ymm13, %ymm13
jmp 8b
9:
vxorpd %ymm13, %ymm13, %ymm13
jmp 10b
11:
vxorpd %ymm13, %ymm13, %ymm13
jmp 12b
13:
vxorpd %ymm13, %ymm13, %ymm13
jmp 14b
15:
vxorpd %ymm13, %ymm13, %ymm13
jmp 16b
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_edge_dpotrf_8x8_vs_lib4, .-inner_edge_dpotrf_8x8_vs_lib4
#endif
#endif
// common inner routine with file scope
//
// triangular substitution for cholesky factorization
//
// input arguments:
// r10 <- E
// r11 <- inv_diag_E
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
//
// output arguments:
// r10 <- E
// r11 <- inv_diag_E
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_EDGE_DTRSM_RLT_INV_8X8L_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_edge_dtrsm_rlt_inv_8x8l_lib4, @function
inner_edge_dtrsm_rlt_inv_8x8l_lib4:
#elif defined(OS_MAC)
_inner_edge_dtrsm_rlt_inv_8x8l_lib4:
#elif defined(OS_WINDOWS)
.def inner_edge_dtrsm_rlt_inv_8x8l_lib4; .scl 2; .type 32; .endef
inner_edge_dtrsm_rlt_inv_8x8l_lib4:
#endif
#endif
vbroadcastsd 0(%r12), %ymm13
vmulpd %ymm0, %ymm13, %ymm0
vmulpd %ymm4, %ymm13, %ymm4
vbroadcastsd 8(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm1
vfnmadd231pd %ymm4, %ymm13, %ymm5
vbroadcastsd 16(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm2
vfnmadd231pd %ymm4, %ymm13, %ymm6
vbroadcastsd 24(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm3
vfnmadd231pd %ymm4, %ymm13, %ymm7
vbroadcastsd 0(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm8
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm11
vbroadcastsd 8(%r12), %ymm13
vmulpd %ymm1, %ymm13, %ymm1
vmulpd %ymm5, %ymm13, %ymm5
vbroadcastsd 48(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm2
vfnmadd231pd %ymm5, %ymm13, %ymm6
vbroadcastsd 56(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm3
vfnmadd231pd %ymm5, %ymm13, %ymm7
vbroadcastsd 32(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm8
vbroadcastsd 40(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm9
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm11
vbroadcastsd 16(%r12), %ymm13
vmulpd %ymm2, %ymm13, %ymm2
vmulpd %ymm6, %ymm13, %ymm6
vbroadcastsd 88(%r10), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm3
vfnmadd231pd %ymm6, %ymm13, %ymm7
vbroadcastsd 64(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm8
vbroadcastsd 72(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm9
vbroadcastsd 80(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm10
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm11
vbroadcastsd 24(%r12), %ymm13
vmulpd %ymm3, %ymm13, %ymm3
vmulpd %ymm7, %ymm13, %ymm7
vbroadcastsd 96(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm8
vbroadcastsd 104(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm9
vbroadcastsd 112(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm10
vbroadcastsd 120(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm11
addq $128, %r10
vbroadcastsd 32(%r12), %ymm13
vmulpd %ymm8, %ymm13, %ymm8
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm11
vbroadcastsd 40(%r12), %ymm13
vmulpd %ymm9, %ymm13, %ymm9
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm9, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm9, %ymm13, %ymm11
vbroadcastsd 48(%r12), %ymm13
vmulpd %ymm10, %ymm13, %ymm10
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm10, %ymm13, %ymm11
vbroadcastsd 56(%r12), %ymm13
vmulpd %ymm11, %ymm13, %ymm11
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_edge_dtrsm_rlt_inv_8x8l_lib4, .-inner_edge_dtrsm_rlt_inv_8x8l_lib4
#endif
#endif
// common inner routine with file scope
//
// triangular substitution for cholesky factorization
//
// input arguments:
// r10 <- E
// r11 <- sde
// r12 <- inv_diag_E
// r13 <- D
// r14 <- sdd
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
//
// output arguments:
// r10 <- E
// r11 <- sde
// r12 <- inv_diag_E
// r13 <- D
// r14 <- sdd
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_EDGE_DTRSM_RLT_INV_8X8U_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_edge_dtrsm_rlt_inv_8x8u_lib4, @function
inner_edge_dtrsm_rlt_inv_8x8u_lib4:
#elif defined(OS_MAC)
_inner_edge_dtrsm_rlt_inv_8x8u_lib4:
#elif defined(OS_WINDOWS)
.def inner_edge_dtrsm_rlt_inv_8x8u_lib4; .scl 2; .type 32; .endef
inner_edge_dtrsm_rlt_inv_8x8u_lib4:
#endif
#endif
vbroadcastsd 0(%r12), %ymm13
vmulpd %ymm0, %ymm13, %ymm0
vbroadcastsd 8(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm1
vbroadcastsd 16(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm2
vbroadcastsd 24(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm3
vmovapd 0(%r13, %r14, 1), %ymm12
vbroadcastsd 0(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
vbroadcastsd 8(%r12), %ymm13
vmulpd %ymm1, %ymm13, %ymm1
vbroadcastsd 48(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm2
vbroadcastsd 56(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm3
vmovapd 32(%r13, %r14, 1), %ymm12
vbroadcastsd 32(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 40(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
vbroadcastsd 16(%r12), %ymm13
vmulpd %ymm2, %ymm13, %ymm2
vbroadcastsd 88(%r10), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm3
vmovapd 64(%r13, %r14, 1), %ymm12
vbroadcastsd 64(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 72(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 80(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
vbroadcastsd 24(%r12), %ymm13
vmulpd %ymm3, %ymm13, %ymm3
vmovapd 96(%r13, %r14, 1), %ymm12
vbroadcastsd 96(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 104(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 112(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 120(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
addq $128, %r10
vbroadcastsd 32(%r12), %ymm13
vmulpd %ymm4, %ymm13, %ymm4
vmulpd %ymm8, %ymm13, %ymm8
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm5
vfnmadd231pd %ymm8, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm6
vfnmadd231pd %ymm8, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm7
vfnmadd231pd %ymm8, %ymm13, %ymm11
vbroadcastsd 40(%r12), %ymm13
vmulpd %ymm5, %ymm13, %ymm5
vmulpd %ymm9, %ymm13, %ymm9
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm6
vfnmadd231pd %ymm9, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm7
vfnmadd231pd %ymm9, %ymm13, %ymm11
vbroadcastsd 48(%r12), %ymm13
vmulpd %ymm6, %ymm13, %ymm6
vmulpd %ymm10, %ymm13, %ymm10
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm7
vfnmadd231pd %ymm10, %ymm13, %ymm11
vbroadcastsd 56(%r12), %ymm13
vmulpd %ymm7, %ymm13, %ymm7
vmulpd %ymm11, %ymm13, %ymm11
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_edge_dtrsm_rlt_inv_8x8u_lib4, .-inner_edge_dtrsm_rlt_inv_8x8u_lib4
#endif
#endif
// common inner routine with file scope
//
// triangular substitution for cholesky factorization
//
// input arguments:
// r10 <- D
// r11 <- inv_diag_D
// r12d <- kn
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
//
// output arguments:
// r10 <- D
// r11 <- inv_diag_D
// r12d <- kn
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_EDGE_DTRSM_RLT_INV_8X8L_VS_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4, @function
inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4:
#elif defined(OS_MAC)
_inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4:
#elif defined(OS_WINDOWS)
.def inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4; .scl 2; .type 32; .endef
inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4:
#endif
#endif
vbroadcastsd 0(%r12), %ymm13
vmulpd %ymm0, %ymm13, %ymm0
vmulpd %ymm4, %ymm13, %ymm4
vbroadcastsd 8(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm1
vfnmadd231pd %ymm4, %ymm13, %ymm5
vbroadcastsd 16(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm2
vfnmadd231pd %ymm4, %ymm13, %ymm6
vbroadcastsd 24(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm3
vfnmadd231pd %ymm4, %ymm13, %ymm7
vbroadcastsd 0(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm8
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm11
vbroadcastsd 8(%r12), %ymm13
vmulpd %ymm1, %ymm13, %ymm1
vmulpd %ymm5, %ymm13, %ymm5
vbroadcastsd 48(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm2
vfnmadd231pd %ymm5, %ymm13, %ymm6
vbroadcastsd 56(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm3
vfnmadd231pd %ymm5, %ymm13, %ymm7
vbroadcastsd 32(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm8
vbroadcastsd 40(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm9
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm11
vbroadcastsd 16(%r12), %ymm13
vmulpd %ymm2, %ymm13, %ymm2
vmulpd %ymm6, %ymm13, %ymm6
vbroadcastsd 88(%r10), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm3
vfnmadd231pd %ymm6, %ymm13, %ymm7
vbroadcastsd 64(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm8
vbroadcastsd 72(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm9
vbroadcastsd 80(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm10
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm11
vbroadcastsd 24(%r12), %ymm13
vmulpd %ymm3, %ymm13, %ymm3
vmulpd %ymm7, %ymm13, %ymm7
vbroadcastsd 96(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm8
vbroadcastsd 104(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm9
vbroadcastsd 112(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm10
vbroadcastsd 120(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm7, %ymm13, %ymm11
addq $128, %r10
vbroadcastsd 32(%r12), %ymm13
vmulpd %ymm8, %ymm13, %ymm8
cmpl $6, %r13d
jl 0f // ret
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm8, %ymm13, %ymm11
vbroadcastsd 40(%r12), %ymm13
vmulpd %ymm9, %ymm13, %ymm9
cmpl $7, %r13d
jl 0f // ret
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm9, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm9, %ymm13, %ymm11
vbroadcastsd 48(%r12), %ymm13
vmulpd %ymm10, %ymm13, %ymm10
cmpl $8, %r13d
jl 0f // ret
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm10, %ymm13, %ymm11
vbroadcastsd 56(%r12), %ymm13
vmulpd %ymm11, %ymm13, %ymm11
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4, .-inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4
#endif
#endif
// common inner routine with file scope
//
// triangular substitution for cholesky factorization
//
// input arguments:
// r10 <- E
// r11 <- sde
// r12 <- inv_diag_E
// r13 <- D
// r14 <- sdd
// r15d <- kn
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
//
// output arguments:
// r10 <- E
// r11 <- sde
// r12 <- inv_diag_E
// r13 <- D
// r14 <- sdd
// r15d <- kn
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d51 d62 d73]
// ymm5 <- [d41 d50 d63 d72]
// ymm6 <- [d43 d52 d61 d70]
// ymm7 <- [d42 d53 d60 d71]
// ymm12 <- dirty
// ymm13 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_EDGE_DTRSM_RLT_INV_8X8U_VS_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4, @function
inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4:
#elif defined(OS_MAC)
_inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4:
#elif defined(OS_WINDOWS)
.def inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4; .scl 2; .type 32; .endef
inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4:
#endif
#endif
vbroadcastsd 0(%r12), %ymm13
vmulpd %ymm0, %ymm13, %ymm0
vbroadcastsd 8(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm1
vbroadcastsd 16(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm2
vbroadcastsd 24(%r10), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm3
vmovapd 0(%r13, %r14, 1), %ymm12
vbroadcastsd 0(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm0, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
vbroadcastsd 8(%r12), %ymm13
vmulpd %ymm1, %ymm13, %ymm1
vbroadcastsd 48(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm2
vbroadcastsd 56(%r10), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm3
vmovapd 32(%r13, %r14, 1), %ymm12
vbroadcastsd 32(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 40(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm1, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
vbroadcastsd 16(%r12), %ymm13
vmulpd %ymm2, %ymm13, %ymm2
vbroadcastsd 88(%r10), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm3
vmovapd 64(%r13, %r14, 1), %ymm12
vbroadcastsd 64(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 72(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 80(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm2, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
vbroadcastsd 24(%r12), %ymm13
vmulpd %ymm3, %ymm13, %ymm3
vmovapd 96(%r13, %r14, 1), %ymm12
vbroadcastsd 96(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm4
vfnmadd231pd %ymm12, %ymm13, %ymm8
vbroadcastsd 104(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm5
vfnmadd231pd %ymm12, %ymm13, %ymm9
vbroadcastsd 112(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm6
vfnmadd231pd %ymm12, %ymm13, %ymm10
vbroadcastsd 120(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm3, %ymm13, %ymm7
vfnmadd231pd %ymm12, %ymm13, %ymm11
addq $128, %r10
vbroadcastsd 32(%r12), %ymm13
vmulpd %ymm4, %ymm13, %ymm4
vmulpd %ymm8, %ymm13, %ymm8
cmpl $6, %r15d
jl 0f // ret
vbroadcastsd 8(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm5
vfnmadd231pd %ymm8, %ymm13, %ymm9
vbroadcastsd 16(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm6
vfnmadd231pd %ymm8, %ymm13, %ymm10
vbroadcastsd 24(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm4, %ymm13, %ymm7
vfnmadd231pd %ymm8, %ymm13, %ymm11
vbroadcastsd 40(%r12), %ymm13
vmulpd %ymm5, %ymm13, %ymm5
vmulpd %ymm9, %ymm13, %ymm9
cmpl $7, %r15d
jl 0f // ret
vbroadcastsd 48(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm6
vfnmadd231pd %ymm9, %ymm13, %ymm10
vbroadcastsd 56(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm5, %ymm13, %ymm7
vfnmadd231pd %ymm9, %ymm13, %ymm11
vbroadcastsd 48(%r12), %ymm13
vmulpd %ymm6, %ymm13, %ymm6
vmulpd %ymm10, %ymm13, %ymm10
cmpl $8, %r15d
jl 0f // ret
vbroadcastsd 88(%r10, %r11, 1), %ymm13
vfnmadd231pd %ymm6, %ymm13, %ymm7
vfnmadd231pd %ymm10, %ymm13, %ymm11
vbroadcastsd 56(%r12), %ymm13
vmulpd %ymm7, %ymm13, %ymm7
vmulpd %ymm11, %ymm13, %ymm11
// subq $128, %r10
// vmovapd 0(%r10, %r11, 1), %ymm4
// vmovapd 32(%r10, %r11, 1), %ymm5
// vmovapd 64(%r10, %r11, 1), %ymm6
// vmovapd 96(%r10, %r11, 1), %ymm7
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4, .-inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4
#endif
#endif
// common inner routine with file scope
//
// store n
//
// input arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
//
// output arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
#if MACRO_LEVEL>=1
.macro INNER_STORE_8X8L_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_store_8x8l_lib4, @function
inner_store_8x8l_lib4:
#elif defined(OS_MAC)
_inner_store_8x8l_lib4:
#elif defined(OS_WINDOWS)
.def inner_store_8x8l_lib4; .scl 2; .type 32; .endef
inner_store_8x8l_lib4:
#endif
#endif
vmovapd %ymm0, 0(%r10)
vmovapd %ymm1, 32(%r10)
vmovapd %ymm2, 64(%r10)
vmovapd %ymm3, 96(%r10)
vmovapd %ymm4, 0(%r10, %r11, 1)
vmovapd %ymm5, 32(%r10, %r11, 1)
vmovapd %ymm6, 64(%r10, %r11, 1)
vmovapd %ymm7, 96(%r10, %r11, 1)
vmovapd %ymm8, 128(%r10, %r11, 1)
vmovapd %ymm9, 160(%r10, %r11, 1)
vmovapd %ymm10, 192(%r10, %r11, 1)
vmovapd %ymm11, 224(%r10, %r11, 1)
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_store_8x8l_lib4, .-inner_store_8x8l_lib4
#endif
#endif
// common inner routine with file scope
//
// store n
//
// input arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
//
// output arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
#if MACRO_LEVEL>=1
.macro INNER_STORE_8X8U_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_store_8x8u_lib4, @function
inner_store_8x8u_lib4:
#elif defined(OS_MAC)
_inner_store_8x8u_lib4:
#elif defined(OS_WINDOWS)
.def inner_store_8x8u_lib4; .scl 2; .type 32; .endef
inner_store_8x8u_lib4:
#endif
#endif
vmovapd %ymm0, 0(%r10)
vmovapd %ymm1, 32(%r10)
vmovapd %ymm2, 64(%r10)
vmovapd %ymm3, 96(%r10)
vmovapd %ymm4, 128(%r10)
vmovapd %ymm5, 160(%r10)
vmovapd %ymm6, 192(%r10)
vmovapd %ymm7, 224(%r10)
vmovapd %ymm8, 128(%r10, %r11, 1)
vmovapd %ymm9, 160(%r10, %r11, 1)
vmovapd %ymm10, 192(%r10, %r11, 1)
vmovapd %ymm11, 224(%r10, %r11, 1)
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_store_8x8u_lib4, .-inner_store_8x8u_lib4
#endif
#endif
// common inner routine with file scope
//
// store n
//
// input arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r12d <- km
// r13d <- kn
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
//
// output arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r12d <- km
// r13d <- kn
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
#if MACRO_LEVEL>=1
.macro INNER_STORE_8X8L_VS_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_store_8x8l_vs_lib4, @function
inner_store_8x8l_vs_lib4:
#elif defined(OS_MAC)
_inner_store_8x8l_vs_lib4:
#elif defined(OS_WINDOWS)
.def inner_store_8x8l_vs_lib4; .scl 2; .type 32; .endef
inner_store_8x8l_vs_lib4:
#endif
#endif
vcvtsi2sd %r12d, %xmm15, %xmm15
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vmovupd .LC03(%rip), %ymm14
#elif defined(OS_MAC)
vmovupd LC03(%rip), %ymm14
#endif
vmovddup %xmm15, %xmm15
vinsertf128 $1, %xmm15, %ymm15, %ymm15
vsubpd %ymm15, %ymm14, %ymm15
vmovapd %ymm0, 0(%r10)
vmovapd %ymm1, 32(%r10)
vmovapd %ymm2, 64(%r10)
vmovapd %ymm3, 96(%r10)
vmaskmovpd %ymm4, %ymm15, 0(%r10, %r11, 1)
vmaskmovpd %ymm5, %ymm15, 32(%r10, %r11, 1)
vmaskmovpd %ymm6, %ymm15, 64(%r10, %r11, 1)
vmaskmovpd %ymm7, %ymm15, 96(%r10, %r11, 1)
vmaskmovpd %ymm8, %ymm15, 128(%r10, %r11, 1)
cmpl $6, %r13d
jl 0f // end
vmaskmovpd %ymm9, %ymm15, 160(%r10, %r11, 1)
cmpl $7, %r13d
jl 0f // end
vmaskmovpd %ymm10, %ymm15, 192(%r10, %r11, 1)
je 0f // end
vmaskmovpd %ymm11, %ymm15, 224(%r10, %r11, 1)
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_store_8x8l_vs_lib4, .-inner_store_8x8l_vs_lib4
#endif
#endif
// common inner routine with file scope
//
// store n
//
// input arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r12d <- km
// r13d <- kn
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
//
// output arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r12d <- km
// r13d <- kn
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d91 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
#if MACRO_LEVEL>=1
.macro INNER_STORE_8X8U_VS_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_store_8x8u_vs_lib4, @function
inner_store_8x8u_vs_lib4:
#elif defined(OS_MAC)
_inner_store_8x8u_vs_lib4:
#elif defined(OS_WINDOWS)
.def inner_store_8x8u_vs_lib4; .scl 2; .type 32; .endef
inner_store_8x8u_vs_lib4:
#endif
#endif
vcvtsi2sd %r12d, %xmm15, %xmm15
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vmovupd .LC03(%rip), %ymm14
#elif defined(OS_MAC)
vmovupd LC03(%rip), %ymm14
#endif
vmovddup %xmm15, %xmm15
vinsertf128 $1, %xmm15, %ymm15, %ymm15
vsubpd %ymm15, %ymm14, %ymm15
vmovapd %ymm0, 0(%r10)
vmovapd %ymm1, 32(%r10)
vmovapd %ymm2, 64(%r10)
vmovapd %ymm3, 96(%r10)
vmovapd %ymm4, 128(%r10)
vmaskmovpd %ymm8, %ymm15, 128(%r10, %r11, 1)
cmpl $6, %r13d
jl 0f // end
vmovapd %ymm5, 160(%r10)
vmaskmovpd %ymm9, %ymm15, 160(%r10, %r11, 1)
cmpl $7, %r13d
jl 0f // end
vmovapd %ymm6, 192(%r10)
vmaskmovpd %ymm10, %ymm15, 192(%r10, %r11, 1)
je 0f // end
vmovapd %ymm7, 224(%r10)
vmaskmovpd %ymm11, %ymm15, 224(%r10, %r11, 1)
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_store_8x8u_vs_lib4, .-inner_store_8x8u_vs_lib4
#endif
#endif
// common inner routine with file scope
//
// store lower n
//
// input arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d50 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d90 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d90 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_STORE_L_8X8_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_store_l_8x8_lib4, @function
inner_store_l_8x8_lib4:
#elif defined(OS_MAC)
_inner_store_l_8x8_lib4:
#elif defined(OS_WINDOWS)
.def inner_store_l_8x8_lib4; .scl 2; .type 32; .endef
inner_store_l_8x8_lib4:
#endif
#endif
vmovapd %ymm0, 0(%r10)
vmovapd 32(%r10), %ymm14
vblendpd $0x1, %ymm14, %ymm1, %ymm1
vmovapd %ymm1, 32(%r10)
vmovapd 64(%r10), %ymm14
vblendpd $0x3, %ymm14, %ymm2, %ymm2
vmovapd %ymm2, 64(%r10)
vmovapd 96(%r10), %ymm14
vblendpd $0x7, %ymm14, %ymm3, %ymm3
vmovapd %ymm3, 96(%r10)
vmovapd %ymm4, 0(%r10, %r11, 1)
vmovapd %ymm5, 32(%r10, %r11, 1)
vmovapd %ymm6, 64(%r10, %r11, 1)
vmovapd %ymm7, 96(%r10, %r11, 1)
vmovapd %ymm8, 128(%r10, %r11, 1)
vmovapd 160(%r10, %r11, 1), %ymm14
vblendpd $0x1, %ymm14, %ymm9, %ymm9
vmovapd %ymm9, 160(%r10, %r11, 1)
vmovapd 192(%r10, %r11, 1), %ymm14
vblendpd $0x3, %ymm14, %ymm10, %ymm10
vmovapd %ymm10, 192(%r10, %r11, 1)
vmovapd 224(%r10, %r11, 1), %ymm14
vblendpd $0x7, %ymm14, %ymm11, %ymm11
vmovapd %ymm11, 224(%r10, %r11, 1)
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_store_l_8x8_lib4, .-inner_store_l_8x8_lib4
#endif
#endif
// common inner routine with file scope
//
// store lower n
//
// input arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r12d <- km
// r13d <- kn
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d50 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d90 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm14 <- dirty
// ymm15 <- dirty
//
// output arguments:
// r10 <- D
// r11 <- 4*sdd*sizeof(double)
// r12d <- km
// r13d <- kn
// r14 <- dirty
// r15 <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
// ymm8 <- [d80 d90 da0 db0]
// ymm9 <- [d81 d90 da1 db1]
// ymm10 <- [d82 d92 da2 db2]
// ymm11 <- [d83 d93 da3 db3]
// ymm14 <- dirty
// ymm15 <- dirty
#if MACRO_LEVEL>=1
.macro INNER_STORE_L_8X8_VS_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_store_l_8x8_vs_lib4, @function
inner_store_l_8x8_vs_lib4:
#elif defined(OS_MAC)
_inner_store_l_8x8_vs_lib4:
#elif defined(OS_WINDOWS)
.def inner_store_l_8x8_vs_lib4; .scl 2; .type 32; .endef
inner_store_l_8x8_vs_lib4:
#endif
#endif
vcvtsi2sd %r12d, %xmm15, %xmm15
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vmovupd .LC03(%rip), %ymm14
#elif defined(OS_MAC)
vmovupd LC03(%rip), %ymm14
#endif
vmovddup %xmm15, %xmm15
vinsertf128 $1, %xmm15, %ymm15, %ymm15
vsubpd %ymm15, %ymm14, %ymm15
vmovapd %ymm0, 0(%r10)
vmovapd 32(%r10), %ymm14
vblendpd $0x1, %ymm14, %ymm1, %ymm1
vmovapd %ymm1, 32(%r10)
vmovapd 64(%r10), %ymm14
vblendpd $0x3, %ymm14, %ymm2, %ymm2
vmovapd %ymm2, 64(%r10)
vmovapd 96(%r10), %ymm14
vblendpd $0x7, %ymm14, %ymm3, %ymm3
vmovapd %ymm3, 96(%r10)
vmaskmovpd %ymm4, %ymm15, 0(%r10, %r11, 1)
vmaskmovpd %ymm5, %ymm15, 32(%r10, %r11, 1)
vmaskmovpd %ymm6, %ymm15, 64(%r10, %r11, 1)
vmaskmovpd %ymm7, %ymm15, 96(%r10, %r11, 1)
vmaskmovpd %ymm8, %ymm15, 128(%r10, %r11, 1)
cmpl $6, %r13d
jl 0f // end
vmovapd 160(%r10, %r11, 1), %ymm14
vblendpd $0x1, %ymm14, %ymm9, %ymm9
vmaskmovpd %ymm9, %ymm15, 160(%r10, %r11, 1)
cmpl $7, %r13d
jl 0f // end
vmovapd 192(%r10, %r11, 1), %ymm14
vblendpd $0x3, %ymm14, %ymm10, %ymm10
vmaskmovpd %ymm10, %ymm15, 192(%r10, %r11, 1)
je 0f // end
vmovapd 224(%r10, %r11, 1), %ymm14
vblendpd $0x7, %ymm14, %ymm11, %ymm11
vmaskmovpd %ymm11, %ymm15, 224(%r10, %r11, 1)
0:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_store_l_8x8_vs_lib4, .-inner_store_l_8x8_vs_lib4
#endif
#endif
// common inner routine with file scope
//
// store n generalized
//
// input arguments:
// r10 <- offset
// r11 <- D
// r12 <- 4*sdd*sizeof(double)
// r13 <- m0 // row index: start from (inc)
// r14 <- m1 // row index: up to (exc)
// r15 <- n0 // col index: start from (inc)
// rax <- n1 // col index: up to (exc)
// rbx <- dirty
// rbp <- dirty
// ymm0 <- [d00 d11 d22 d33]
// ymm1 <- [d01 d10 d23 d32]
// ymm2 <- [d03 d12 d21 d30]
// ymm3 <- [d02 d13 d20 d31]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
//
// output arguments:
// r10 <- offset
// r11 <- D
// r12 <- 4*sdd*sizeof(double)
// r13 <- m0 // row index: start from (inc)
// r14 <- m1 // row index: up to (exc)
// r15 <- n1-n0
// rax <- n1-n0
// rbx <- dirty
// rbp <- dirty
// ymm0 <- [d00 d10 d20 d30]
// ymm1 <- [d01 d11 d21 d31]
// ymm2 <- [d02 d12 d22 d32]
// ymm3 <- [d03 d13 d23 d33]
// ymm4 <- [d40 d50 d60 d70]
// ymm5 <- [d41 d51 d61 d71]
// ymm6 <- [d42 d52 d62 d72]
// ymm7 <- [d43 d53 d63 d73]
#if MACRO_LEVEL>=1
.macro INNER_STORE_8X8_GEN_LIB4
#else
.p2align 4,,15
#if defined(OS_LINUX)
.type inner_store_8x8_gen_lib4, @function
inner_store_8x8_gen_lib4:
#elif defined(OS_MAC)
_inner_store_8x8_gen_lib4:
#elif defined(OS_WINDOWS)
.def inner_store_8x8_gen_lib4; .scl 2; .type 32; .endef
inner_store_8x8_gen_lib4:
#endif
#endif
// compute mask for rows
vcvtsi2sd %r13d, %xmm14, %xmm14
vcvtsi2sd %r14d, %xmm15, %xmm15
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vmovupd .LC02(%rip), %ymm12
vmovupd .LC03(%rip), %ymm13
#elif defined(OS_MAC)
vmovupd LC02(%rip), %ymm12
vmovupd LC03(%rip), %ymm13
#endif
vmovddup %xmm14, %xmm14
vmovddup %xmm15, %xmm15
vinsertf128 $1, %xmm14, %ymm14, %ymm14
vinsertf128 $1, %xmm15, %ymm15, %ymm15
vsubpd %ymm12, %ymm14, %ymm14
vsubpd %ymm15, %ymm13, %ymm15
// shift D and sol for cols
cmpl $0, %r15d
jle 0f
vmovapd %ymm1, %ymm0
vmovapd %ymm5, %ymm4
vmovapd %ymm2, %ymm1
vmovapd %ymm6, %ymm5
vmovapd %ymm3, %ymm2
vmovapd %ymm7, %ymm6
vmovapd %ymm8, %ymm7
vmovapd %ymm9, %ymm8
vmovapd %ymm10, %ymm9
vmovapd %ymm11, %ymm10
addq $32, %r11
cmpl $1, %r15d
jle 0f
vmovapd %ymm1, %ymm0
vmovapd %ymm5, %ymm4
vmovapd %ymm2, %ymm1
vmovapd %ymm6, %ymm5
vmovapd %ymm7, %ymm6
vmovapd %ymm8, %ymm7
vmovapd %ymm9, %ymm8
vmovapd %ymm10, %ymm9
addq $32, %r11
cmpl $2, %r15d
jle 0f
vmovapd %ymm1, %ymm0
vmovapd %ymm5, %ymm4
vmovapd %ymm6, %ymm5
vmovapd %ymm7, %ymm6
vmovapd %ymm8, %ymm7
vmovapd %ymm9, %ymm8
addq $32, %r11
0:
// compute number of cols
cmpl $8, %eax
jle 0f
movl $8, %eax
0:
subl %r15d, %eax
movl %eax, %r15d
cmpl $0, %r10d
jg 0f
// offset==0
vmaskmovpd %ymm0, %ymm14, 0(%r11)
vmaskmovpd %ymm1, %ymm14, 32(%r11)
vmaskmovpd %ymm2, %ymm14, 64(%r11)
vmaskmovpd %ymm3, %ymm14, 96(%r11)
vmaskmovpd %ymm4, %ymm15, 0(%r11, %r12, 1)
vmaskmovpd %ymm5, %ymm15, 32(%r11, %r12, 1)
vmaskmovpd %ymm6, %ymm15, 64(%r11, %r12, 1)
vmaskmovpd %ymm7, %ymm15, 96(%r11, %r12, 1)
vmaskmovpd %ymm8, %ymm15, 128(%r11, %r12, 1)
cmpl $6, %r15d
jl 4f // end
vmaskmovpd %ymm9, %ymm15, 160(%r11, %r12, 1)
cmpl $7, %r15d
jl 4f // end
vmaskmovpd %ymm10, %ymm15, 192(%r11, %r12, 1)
je 4f // end
vmaskmovpd %ymm11, %ymm15, 224(%r11, %r12, 1)
jmp 4f
0:
cmpl $1, %r10d
jg 1f
// offset==1
vmovapd %ymm0, %ymm13
vperm2f128 $0x03, %ymm4, %ymm0, %ymm12
vshufpd $0x5, %ymm0, %ymm12, %ymm0
vperm2f128 $0x03, %ymm13, %ymm4, %ymm12
vshufpd $0x5, %ymm4, %ymm12, %ymm4
vmovapd %ymm1, %ymm13
vperm2f128 $0x03, %ymm5, %ymm1, %ymm12
vshufpd $0x5, %ymm1, %ymm12, %ymm1
vperm2f128 $0x03, %ymm13, %ymm5, %ymm12
vshufpd $0x5, %ymm5, %ymm12, %ymm5
vmovapd %ymm2, %ymm13
vperm2f128 $0x03, %ymm6, %ymm2, %ymm12
vshufpd $0x5, %ymm2, %ymm12, %ymm2
vperm2f128 $0x03, %ymm13, %ymm6, %ymm12
vshufpd $0x5, %ymm6, %ymm12, %ymm6
vmovapd %ymm3, %ymm13
vperm2f128 $0x03, %ymm7, %ymm3, %ymm12
vshufpd $0x5, %ymm3, %ymm12, %ymm3
vperm2f128 $0x03, %ymm13, %ymm7, %ymm12
vshufpd $0x5, %ymm7, %ymm12, %ymm7
vperm2f128 $0x01, %ymm8, %ymm8, %ymm12
vshufpd $0x5, %ymm8, %ymm12, %ymm8
vperm2f128 $0x01, %ymm9, %ymm9, %ymm12
vshufpd $0x5, %ymm9, %ymm12, %ymm9
vperm2f128 $0x01, %ymm10, %ymm10, %ymm12
vshufpd $0x5, %ymm10, %ymm12, %ymm10
vperm2f128 $0x01, %ymm11, %ymm11, %ymm12
vshufpd $0x5, %ymm11, %ymm12, %ymm11
vperm2f128 $0x01, %ymm15, %ymm15, %ymm12
vshufpd $0x5, %ymm15, %ymm12, %ymm15
vperm2f128 $0x01, %ymm14, %ymm14, %ymm12
vshufpd $0x5, %ymm14, %ymm12, %ymm14
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vandpd .LC08(%rip), %ymm14, %ymm12
vandpd .LC05(%rip), %ymm15, %ymm13
#elif defined(OS_MAC)
vandpd LC08(%rip), %ymm14, %ymm12
vandpd LC05(%rip), %ymm15, %ymm13
#endif
vblendpd $0x1, %ymm14, %ymm15, %ymm14
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vandpd .LC08(%rip), %ymm15, %ymm15
#elif defined(OS_MAC)
vandpd LC08(%rip), %ymm15, %ymm15
#endif
jmp 3f
1:
cmpl $2, %r10d
jg 2f
// offset==2
vmovapd %ymm0, %ymm13
vperm2f128 $0x03, %ymm4, %ymm0, %ymm0
vperm2f128 $0x03, %ymm13, %ymm4, %ymm4
vmovapd %ymm1, %ymm13
vperm2f128 $0x03, %ymm5, %ymm1, %ymm1
vperm2f128 $0x03, %ymm13, %ymm5, %ymm5
vmovapd %ymm2, %ymm13
vperm2f128 $0x03, %ymm6, %ymm2, %ymm2
vperm2f128 $0x03, %ymm13, %ymm6, %ymm6
vmovapd %ymm3, %ymm13
vperm2f128 $0x03, %ymm7, %ymm3, %ymm3
vperm2f128 $0x03, %ymm13, %ymm7, %ymm7
vperm2f128 $0x01, %ymm8, %ymm8, %ymm8
vperm2f128 $0x01, %ymm9, %ymm9, %ymm9
vperm2f128 $0x01, %ymm10, %ymm10, %ymm10
vperm2f128 $0x01, %ymm11, %ymm11, %ymm11
vperm2f128 $0x01, %ymm14, %ymm14, %ymm14
vperm2f128 $0x01, %ymm15, %ymm15, %ymm15
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vandpd .LC09(%rip), %ymm14, %ymm12
vandpd .LC06(%rip), %ymm15, %ymm13
#elif defined(OS_MAC)
vandpd LC09(%rip), %ymm14, %ymm12
vandpd LC06(%rip), %ymm15, %ymm13
#endif
vblendpd $0x3, %ymm14, %ymm15, %ymm14
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vandpd .LC09(%rip), %ymm15, %ymm15
#elif defined(OS_MAC)
vandpd LC09(%rip), %ymm15, %ymm15
#endif
jmp 3f
2:
// offset==3
vmovapd %ymm0, %ymm13
vperm2f128 $0x21, %ymm0, %ymm4, %ymm12
vshufpd $0x5, %ymm12, %ymm4, %ymm0
vperm2f128 $0x21, %ymm4, %ymm13, %ymm12
vshufpd $0x5, %ymm12, %ymm13, %ymm4
vmovapd %ymm1, %ymm13
vperm2f128 $0x21, %ymm1, %ymm5, %ymm12
vshufpd $0x5, %ymm12, %ymm5, %ymm1
vperm2f128 $0x21, %ymm5, %ymm13, %ymm12
vshufpd $0x5, %ymm12, %ymm13, %ymm5
vmovapd %ymm2, %ymm13
vperm2f128 $0x21, %ymm2, %ymm6, %ymm12
vshufpd $0x5, %ymm12, %ymm6, %ymm2
vperm2f128 $0x21, %ymm6, %ymm13, %ymm12
vshufpd $0x5, %ymm12, %ymm13, %ymm6
vmovapd %ymm3, %ymm13
vperm2f128 $0x21, %ymm3, %ymm7, %ymm12
vshufpd $0x5, %ymm12, %ymm7, %ymm3
vperm2f128 $0x21, %ymm7, %ymm13, %ymm12
vshufpd $0x5, %ymm12, %ymm13, %ymm7
vperm2f128 $0x01, %ymm8, %ymm8, %ymm12
vshufpd $0x5, %ymm12, %ymm8, %ymm8
vperm2f128 $0x01, %ymm9, %ymm9, %ymm12
vshufpd $0x5, %ymm12, %ymm9, %ymm9
vperm2f128 $0x01, %ymm10, %ymm10, %ymm12
vshufpd $0x5, %ymm12, %ymm10, %ymm10
vperm2f128 $0x01, %ymm11, %ymm11, %ymm12
vshufpd $0x5, %ymm12, %ymm11, %ymm11
vperm2f128 $0x01, %ymm14, %ymm14, %ymm12
vshufpd $0x5, %ymm12, %ymm14, %ymm14
vperm2f128 $0x01, %ymm15, %ymm15, %ymm12
vshufpd $0x5, %ymm12, %ymm15, %ymm15
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vandpd .LC10(%rip), %ymm14, %ymm12
vandpd .LC07(%rip), %ymm15, %ymm13
#elif defined(OS_MAC)
vandpd LC10(%rip), %ymm14, %ymm12
vandpd LC07(%rip), %ymm15, %ymm13
#endif
vblendpd $0x7, %ymm14, %ymm15, %ymm14
#if defined(OS_LINUX) | defined(OS_WINDOWS)
vandpd .LC10(%rip), %ymm15, %ymm15
#elif defined(OS_MAC)
vandpd LC10(%rip), %ymm15, %ymm15
#endif
3:
vmaskmovpd %ymm0, %ymm12, 0(%r11)
vmaskmovpd %ymm4, %ymm14, 0(%r11, %r12, 1)
vmaskmovpd %ymm0, %ymm13, 0(%r11, %r12, 2)
vmaskmovpd %ymm1, %ymm12, 32(%r11)
vmaskmovpd %ymm5, %ymm14, 32(%r11, %r12, 1)
vmaskmovpd %ymm1, %ymm13, 32(%r11, %r12, 2)
vmaskmovpd %ymm2, %ymm12, 64(%r11)
vmaskmovpd %ymm6, %ymm14, 64(%r11, %r12, 1)
vmaskmovpd %ymm2, %ymm13, 64(%r11, %r12, 2)
vmaskmovpd %ymm3, %ymm12, 96(%r11)
vmaskmovpd %ymm7, %ymm14, 96(%r11, %r12, 1)
vmaskmovpd %ymm3, %ymm13, 96(%r11, %r12, 2)
vmaskmovpd %ymm8, %ymm15, 128(%r11, %r12, 1)
vmaskmovpd %ymm8, %ymm13, 128(%r11, %r12, 2)
cmpl $6, %r15d
jl 4f // end
vmaskmovpd %ymm9, %ymm15, 160(%r11, %r12, 1)
vmaskmovpd %ymm9, %ymm13, 160(%r11, %r12, 2)
cmpl $7, %r15d
jl 4f // end
vmaskmovpd %ymm10, %ymm15, 192(%r11, %r12, 1)
vmaskmovpd %ymm10, %ymm13, 192(%r11, %r12, 2)
je 4f // end
vmaskmovpd %ymm11, %ymm15, 224(%r11, %r12, 1)
vmaskmovpd %ymm11, %ymm13, 224(%r11, %r12, 2)
4:
#if MACRO_LEVEL>=1
.endm
#else
ret
#if defined(OS_LINUX)
.size inner_store_8x8_gen_lib4, .-inner_store_8x8_gen_lib4
#endif
#endif
// 1 2 3 4 5 6 7 8 9 10 11
// void kernel_dgemm_nt_8x8l_lib4(int k, double *alpha, double *A, int sda, double *B, int sdb, double *beta, double *C, int sdc, double *D, int sdd);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dgemm_nt_8x8l_lib4
.type kernel_dgemm_nt_8x8l_lib4, @function
kernel_dgemm_nt_8x8l_lib4:
#elif defined(OS_MAC)
.globl _kernel_dgemm_nt_8x8l_lib4
_kernel_dgemm_nt_8x8l_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dgemm_nt_8x8l_lib4
.def kernel_dgemm_nt_8x8l_lib4; .scl 2; .type 32; .endef
kernel_dgemm_nt_8x8l_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG3, %r11 // A
movq ARG4, %r12 // sda
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG5, %r13 // B
movq ARG6, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner blend scale
movq ARG2, %r10 // alpha
movq ARG7, %r11 // beta
movq ARG8, %r12 // C
movq ARG9, %r13 // sdc
sall $5, %r13d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_AB_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_ab_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_ab_8x8_lib4
#endif
#endif
// store n
movq ARG10, %r10 // D
movq ARG11, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_8X8L_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8l_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8l_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dgemm_nt_8x8l_lib4, .-kernel_dgemm_nt_8x8l_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11
// void kernel_dgemm_nt_8x8u_lib4(int k, double *alpha, double *A, int sda, double *B, int sdb, double *beta, double *C, int sdc, double *D, int sdd);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dgemm_nt_8x8u_lib4
.type kernel_dgemm_nt_8x8u_lib4, @function
kernel_dgemm_nt_8x8u_lib4:
#elif defined(OS_MAC)
.globl _kernel_dgemm_nt_8x8u_lib4
_kernel_dgemm_nt_8x8u_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dgemm_nt_8x8u_lib4
.def kernel_dgemm_nt_8x8u_lib4; .scl 2; .type 32; .endef
kernel_dgemm_nt_8x8u_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG5, %r11 // B
movq ARG6, %r12 // sdb
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG3, %r13 // A
movq ARG4, %r14 // sda
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner blend scale
movq ARG2, %r10 // alpha
movq ARG7, %r11 // beta
movq ARG8, %r12 // C
movq ARG9, %r13 // sdc
sall $5, %r13d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_TRAN_SCALE_AB_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_tran_scale_ab_8x8_lib4
#elif defined(OS_MAC)
callq _inner_tran_scale_ab_8x8_lib4
#endif
#endif
// store n
movq ARG10, %r10 // D
movq ARG11, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_8X8U_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8u_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8u_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dgemm_nt_8x8u_lib4, .-kernel_dgemm_nt_8x8u_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13
// void kernel_dgemm_nt_8x8l_vs_lib4(int k, double *alpha, double *A, int sda, double *B, int sdb, double *beta, double *C, int sdc, double *D, int sdd, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dgemm_nt_8x8l_vs_lib4
.type kernel_dgemm_nt_8x8l_vs_lib4, @function
kernel_dgemm_nt_8x8l_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dgemm_nt_8x8l_vs_lib4
_kernel_dgemm_nt_8x8l_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dgemm_nt_8x8l_vs_lib4
.def kernel_dgemm_nt_8x8l_vs_lib4; .scl 2; .type 32; .endef
kernel_dgemm_nt_8x8l_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG3, %r11 // A
movq ARG4, %r12 // sda
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG5, %r13 // B
movq ARG6, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner blend scale
movq ARG2, %r10 // alpha
movq ARG7, %r11 // beta
movq ARG8, %r12 // C
movq ARG9, %r13 // sdc
sall $5, %r13d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_AB_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_ab_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_ab_8x8_lib4
#endif
#endif
// store n
movq ARG10, %r10 // D
movq ARG11, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG12, %r12 // km
movq ARG13, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_8X8L_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8l_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8l_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dgemm_nt_8x8l_vs_lib4, .-kernel_dgemm_nt_8x8l_vs_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13
// void kernel_dgemm_nt_8x8u_vs_lib4(int k, double *alpha, double *A, int sda, double *B, int sdb, double *beta, double *C, int sdc, double *D, int sdd, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dgemm_nt_8x8u_vs_lib4
.type kernel_dgemm_nt_8x8u_vs_lib4, @function
kernel_dgemm_nt_8x8u_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dgemm_nt_8x8u_vs_lib4
_kernel_dgemm_nt_8x8u_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dgemm_nt_8x8u_vs_lib4
.def kernel_dgemm_nt_8x8u_vs_lib4; .scl 2; .type 32; .endef
kernel_dgemm_nt_8x8u_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG5, %r11 // B
movq ARG6, %r12 // sdb
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG3, %r13 // A
movq ARG4, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner blend scale
movq ARG2, %r10 // alpha
movq ARG7, %r11 // beta
movq ARG8, %r12 // C
movq ARG9, %r13 // sdc
sall $5, %r13d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_TRAN_SCALE_AB_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_tran_scale_ab_8x8_lib4
#elif defined(OS_MAC)
callq _inner_tran_scale_ab_8x8_lib4
#endif
#endif
// store n
movq ARG10, %r10 // D
movq ARG11, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG12, %r12 // km
movq ARG13, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_8X8U_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8u_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8u_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dgemm_nt_8x8u_vs_lib4, .-kernel_dgemm_nt_8x8u_vs_lib4
#endif
#if 0
// 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
// void kernel_dgemm_nt_8x8_gen_lib4(int k, double *alpha, double *A, int sda, double *B, int sdb, double *beta, int offC, double *C, int sdc, int offD, double *D, int sdd, int m0, int m1, int n0, int n1);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dgemm_nt_8x8_gen_lib4
.type kernel_dgemm_nt_8x8_gen_lib4, @function
kernel_dgemm_nt_8x8_gen_lib4:
#elif defined(OS_MAC)
.globl _kernel_dgemm_nt_8x8_gen_lib4
_kernel_dgemm_nt_8x8_gen_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dgemm_nt_8x8_gen_lib4
.def kernel_dgemm_nt_8x8_gen_lib4; .scl 2; .type 32; .endef
kernel_dgemm_nt_8x8_gen_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG3, %r11 // A
movq ARG4, %r12 // sda
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG5, %r13 // B
movq ARG6, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner blend scale
movq ARG2, %r10 // alpha
movq ARG7, %r11 // beta
movq ARG8, %r12 // C
movq ARG9, %r13 // sdc
sall $5, %r13d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_AB_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_ab_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_ab_8x8_lib4
#endif
#endif
// store n
movq ARG10, %r10 // D
movq ARG11, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dgemm_nt_8x8_lib4, .-kernel_dgemm_nt_8x8_lib4
#endif
#endif
// 1 2 3 4 5 6 7 8 9 10 11
// void kernel_dsyrk_nt_8x8_lib4(int k, double *alpha, double *A, int sda, double *B, int sdb, double *beta, double *C, int sdc, double *D, int sdd);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dsyrk_nt_l_8x8_lib4
.type kernel_dsyrk_nt_l_8x8_lib4, @function
kernel_dsyrk_nt_l_8x8_lib4:
#elif defined(OS_MAC)
.globl _kernel_dsyrk_nt_l_8x8_lib4
_kernel_dsyrk_nt_l_8x8_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dsyrk_nt_l_8x8_lib4
.def kernel_dsyrk_nt_l_8x8_lib4; .scl 2; .type 32; .endef
kernel_dsyrk_nt_l_8x8_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG3, %r11 // A
movq ARG4, %r12 // sda
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG5, %r13 // B
movq ARG6, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner blend scale
movq ARG2, %r10 // alpha
movq ARG7, %r11 // beta
movq ARG8, %r12 // C
movq ARG9, %r13 // sdc
sall $5, %r13d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_AB_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_ab_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_ab_8x8_lib4
#endif
#endif
// store n
movq ARG10, %r10 // D
movq ARG11, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_L_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_l_8x8_lib4
#elif defined(OS_MAC)
callq _inner_store_l_8x8_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dsyrk_nt_l_8x8_lib4, .-kernel_dsyrk_nt_l_8x8_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13
// void kernel_dsyrk_nt_8x8_vs_lib4(int k, double *alpha, double *A, int sda, double *B, int sdb, double *beta, double *C, int sdc, double *D, int sdd, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dsyrk_nt_l_8x8_vs_lib4
.type kernel_dsyrk_nt_l_8x8_vs_lib4, @function
kernel_dsyrk_nt_l_8x8_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dsyrk_nt_l_8x8_vs_lib4
_kernel_dsyrk_nt_l_8x8_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dsyrk_nt_l_8x8_vs_lib4
.def kernel_dsyrk_nt_l_8x8_vs_lib4; .scl 2; .type 32; .endef
kernel_dsyrk_nt_l_8x8_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG3, %r11 // A
movq ARG4, %r12 // sda
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG5, %r13 // B
movq ARG6, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner blend scale
movq ARG2, %r10 // alpha
movq ARG7, %r11 // beta
movq ARG8, %r12 // C
movq ARG9, %r13 // sdc
sall $5, %r13d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_AB_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_ab_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_ab_8x8_lib4
#endif
#endif
// store n
movq ARG10, %r10 // D
movq ARG11, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG12, %r12 // D
movq ARG13, %r13 // D
#if MACRO_LEVEL>=1
INNER_STORE_L_8X8_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_l_8x8_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_l_8x8_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dsyrk_nt_l_8x8_vs_lib4, .-kernel_dsyrk_nt_l_8x8_vs_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10
// void kernel_dpotrf_nt_l_8x8_lib4(int k, double *A, int sda, double *B, int sdb, double *C, int sdc, double *D, int sdd, double *inv_diag_D);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dpotrf_nt_l_8x8_lib4
.type kernel_dpotrf_nt_l_8x8_lib4, @function
kernel_dpotrf_nt_l_8x8_lib4:
#elif defined(OS_MAC)
.globl _kernel_dpotrf_nt_l_8x8_lib4
_kernel_dpotrf_nt_l_8x8_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dpotrf_nt_l_8x8_lib4
.def kernel_dpotrf_nt_l_8x8_lib4; .scl 2; .type 32; .endef
kernel_dpotrf_nt_l_8x8_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG2, %r11 // A
movq ARG3, %r12 // sda
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG4, %r13 // B
movq ARG5, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG6, %r10 // C
movq ARG7, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_11_8x8_lib4
#endif
#endif
// factorization
movq ARG10, %r10 // inv_diag_D
movl $8, %r11d
#if MACRO_LEVEL>=1
INNER_EDGE_DPOTRF_8X8_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dpotrf_8x8_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dpotrf_8x8_vs_lib4
#endif
#endif
// store n
movq ARG8, %r10 // store address D
movq ARG9, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_L_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_l_8x8_lib4
#elif defined(OS_MAC)
callq _inner_store_l_8x8_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dpotrf_nt_l_8x8_lib4, .-kernel_dpotrf_nt_l_8x8_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12
// void kernel_dpotrf_nt_l_8x8_vs_lib4(int k, double *A, int sda, double *B, int sdb, double *C, int sdc, double *D, int sdd, double *inv_diag_D, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dpotrf_nt_l_8x8_vs_lib4
.type kernel_dpotrf_nt_l_8x8_vs_lib4, @function
kernel_dpotrf_nt_l_8x8_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dpotrf_nt_l_8x8_vs_lib4
_kernel_dpotrf_nt_l_8x8_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dpotrf_nt_l_8x8_vs_lib4
.def kernel_dpotrf_nt_l_8x8_vs_lib4; .scl 2; .type 32; .endef
kernel_dpotrf_nt_l_8x8_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10 // k
movq ARG2, %r11 // A
movq ARG3, %r12 // sda
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG4, %r13 // B
movq ARG5, %r14 // sdb
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG6, %r10 // C
movq ARG7, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_11_8x8_lib4
#endif
#endif
// factorization
movq ARG10, %r10 // inv_diag_D
movq ARG12, %r11 // kn
#if MACRO_LEVEL>=1
INNER_EDGE_DPOTRF_8X8_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dpotrf_8x8_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dpotrf_8x8_vs_lib4
#endif
#endif
// store n
movq ARG8, %r10 // store address D
movq ARG9, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG11, %r12 // km
movq ARG12, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_L_8X8_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_l_8x8_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_l_8x8_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dpotrf_nt_l_8x8_vs_lib4, .-kernel_dpotrf_nt_l_8x8_vs_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
// void kernel_dsyrk_dpotrf_nt_l_8x8_lib4(int kp, double *Ap, int sdap, double *Bp, int sdbp, int km, double *Am, int sdam, double *Bm, int sdbm, double *C, int sdc, double *D, int sdd, double *inv_diag_D);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dsyrk_dpotrf_nt_l_8x8_lib4
.type kernel_dsyrk_dpotrf_nt_l_8x8_lib4, @function
kernel_dsyrk_dpotrf_nt_l_8x8_lib4:
#elif defined(OS_MAC)
.globl _kernel_dsyrk_dpotrf_nt_l_8x8_lib4
_kernel_dsyrk_dpotrf_nt_l_8x8_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dsyrk_dpotrf_nt_l_8x8_lib4
.def kernel_dsyrk_dpotrf_nt_l_8x8_lib4; .scl 2; .type 32; .endef
kernel_dsyrk_dpotrf_nt_l_8x8_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt add
movq ARG1, %r10 // kp
movq ARG2, %r11 // Ap
movq ARG3, %r12 // sdap
sall $5, %r12d // 4*sdap*sizeof(double)
movq ARG4, %r13 // Bp
movq ARG5, %r14 // sdbp
sall $5, %r14d // 4*sdbp*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
movq ARG6, %r10 // km
movq ARG7, %r11 // Am
movq ARG8, %r12 // sdam
sall $5, %r12d // 4*sdam*sizeof(double)
movq ARG9, %r13 // Bm
movq ARG10, %r14 // sdbm
sall $5, %r14d // 4*sdbm*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG11, %r10 // C
movq ARG12, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_11_8x8_lib4
#endif
#endif
// factorization
movq ARG15, %r10 // inv_diag_D
movl $8, %r11d
#if MACRO_LEVEL>=1
INNER_EDGE_DPOTRF_8X8_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dpotrf_8x8_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dpotrf_8x8_vs_lib4
#endif
#endif
// store n
movq ARG13, %r10 // store address D
movq ARG14, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_L_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_l_8x8_lib4
#elif defined(OS_MAC)
callq _inner_store_l_8x8_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dsyrk_dpotrf_nt_l_8x8_lib4, .-kernel_dsyrk_dpotrf_nt_l_8x8_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
// void kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4(int kp, double *Ap, int sdap, double *Bp, int sdbp, int km, double *Am, int sdam, double *Bm, int sdbm, double *C, int sdc, double *D, int sdd, double *inv_diag_D, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4
.type kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4, @function
kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4
_kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4
.def kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4; .scl 2; .type 32; .endef
kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt add
movq ARG1, %r10 // kp
movq ARG2, %r11 // Ap
movq ARG3, %r12 // sdap
sall $5, %r12d // 4*sdap*sizeof(double)
movq ARG4, %r13 // Bp
movq ARG5, %r14 // sdbp
sall $5, %r14d // 4*sdbp*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
// call inner dgemm kernel nt sub
movq ARG6, %r10 // km
movq ARG7, %r11 // Am
movq ARG8, %r12 // sdam
sall $5, %r12d // 4*sdam*sizeof(double)
movq ARG9, %r13 // Bm
movq ARG10, %r14 // sdbm
sall $5, %r14d // 4*sdbm*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG11, %r10 // C
movq ARG12, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_11_8x8_lib4
#endif
#endif
// factorization
movq ARG15, %r10 // inv_diag_D
movq ARG17, %r11 // kn
#if MACRO_LEVEL>=1
INNER_EDGE_DPOTRF_8X8_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dpotrf_8x8_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dpotrf_8x8_vs_lib4
#endif
#endif
// store n
movq ARG13, %r10 // store address D
movq ARG14, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG16, %r12 // km
movq ARG17, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_L_8X8_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_l_8x8_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_l_8x8_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4, .-kernel_dsyrk_dpotrf_nt_l_8x8_vs_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12
// void kernel_dtrsm_nt_rl_inv_8x8l_lib4(int k, double *A, int sda, double *B, int sdb, double *C, int sdc, double *D, int sdd, double *E, int sde, double *inv_diag_E);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dtrsm_nt_rl_inv_8x8l_lib4
.type kernel_dtrsm_nt_rl_inv_8x8l_lib4, @function
kernel_dtrsm_nt_rl_inv_8x8l_lib4:
#elif defined(OS_MAC)
.globl _kernel_dtrsm_nt_rl_inv_8x8l_lib4
_kernel_dtrsm_nt_rl_inv_8x8l_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dtrsm_nt_rl_inv_8x8l_lib4
.def kernel_dtrsm_nt_rl_inv_8x8l_lib4; .scl 2; .type 32; .endef
kernel_dtrsm_nt_rl_inv_8x8l_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10
movq ARG2, %r11
movq ARG3, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG4, %r13
movq ARG5, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG6, %r10
movq ARG7, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_11_8x8_lib4
#endif
#endif
// solve
movq ARG10, %r10 // E
movq ARG11, %r11 // sde
sall $5, %r11d // 4*sde*sizeof(double)
movq ARG12, %r12 // inv_diag_E
#if MACRO_LEVEL>=1
INNER_EDGE_DTRSM_RLT_INV_8X8L_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dtrsm_rlt_inv_8x8l_lib4
#elif defined(OS_MAC)
callq _inner_edge_dtrsm_rlt_inv_8x8l_lib4
#endif
#endif
// store n
movq ARG8, %r10 // store address D
movq ARG9, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_8X8L_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8l_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8l_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dtrsm_nt_rl_inv_8x8l_lib4, .-kernel_dtrsm_nt_rl_inv_8x8l_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12
// void kernel_dtrsm_nt_rl_inv_8x8u_lib4(int k, double *A, int sda, double *B, int sdb, double *C, int sdc, double *D, int sdd, double *E, int sde, double *inv_diag_E);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dtrsm_nt_rl_inv_8x8u_lib4
.type kernel_dtrsm_nt_rl_inv_8x8u_lib4, @function
kernel_dtrsm_nt_rl_inv_8x8u_lib4:
#elif defined(OS_MAC)
.globl _kernel_dtrsm_nt_rl_inv_8x8u_lib4
_kernel_dtrsm_nt_rl_inv_8x8u_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dtrsm_nt_rl_inv_8x8u_lib4
.def kernel_dtrsm_nt_rl_inv_8x8u_lib4; .scl 2; .type 32; .endef
kernel_dtrsm_nt_rl_inv_8x8u_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10
movq ARG4, %r11
movq ARG5, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG2, %r13
movq ARG3, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG6, %r10
movq ARG7, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_TRAN_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_tran_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_tran_scale_11_8x8_lib4
#endif
#endif
// solve
movq ARG10, %r10 // E
movq ARG11, %r11 // sde
sall $5, %r11d // 4*sde*sizeof(double)
movq ARG12, %r12 // inv_diag_E
movq ARG8, %r13 // D
movq ARG9, %r14 // sdd
#if MACRO_LEVEL>=1
INNER_EDGE_DTRSM_RLT_INV_8X8U_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dtrsm_rlt_inv_8x8u_lib4
#elif defined(OS_MAC)
callq _inner_edge_dtrsm_rlt_inv_8x8u_lib4
#endif
#endif
// store n
movq ARG8, %r10 // store address D
movq ARG9, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
#if MACRO_LEVEL>=1
INNER_STORE_8X8U_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8u_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8u_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dtrsm_nt_rl_inv_8x8u_lib4, .-kernel_dtrsm_nt_rl_inv_8x8u_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13 14
// void kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4(int k, double *A, int sda, double *B, int sdb, double *C, int sdc, double *D, int sdd, double *E, int sde, double *inv_diag_E, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4
.type kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4, @function
kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4
_kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4
.def kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4; .scl 2; .type 32; .endef
kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10
movq ARG2, %r11
movq ARG3, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG4, %r13
movq ARG5, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG6, %r10
movq ARG7, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_11_8x8_lib4
#endif
#endif
// solve
movq ARG10, %r10 // E
movq ARG11, %r11 // sde
sall $5, %r11d // 4*sde*sizeof(double)
movq ARG12, %r12 // inv_diag_E
movq ARG14, %r13 // kn
#if MACRO_LEVEL>=1
INNER_EDGE_DTRSM_RLT_INV_8X8L_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4
#endif
#endif
// store n
movq ARG8, %r10 // store address D
movq ARG9, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG13, %r12 // km
movq ARG14, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_8X8L_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8l_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8l_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4, .-kernel_dtrsm_nt_rl_inv_8x8l_vs_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13 14
// void kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4(int k, double *A, int sda, double *B, int sdb, double *C, int sdc, double *D, int sdd, double *E, int sde, double *inv_diag_E, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4
.type kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4, @function
kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4
_kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4
.def kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4; .scl 2; .type 32; .endef
kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10
movq ARG4, %r11
movq ARG5, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG2, %r13
movq ARG3, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG6, %r10 // C
movq ARG7, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_TRAN_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_tran_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_tran_scale_11_8x8_lib4
#endif
#endif
// solve
movq ARG10, %r10 // E
movq ARG11, %r11 // sde
sall $5, %r11d // 4*sde*sizeof(double)
movq ARG12, %r12 // inv_diag_E
movq ARG8, %r13 // D
movq ARG9, %r14 // sdd
sall $5, %r14d // 4*sdc*sizeof(double)
movq ARG14, %r15 // kn
#if MACRO_LEVEL>=1
INNER_EDGE_DTRSM_RLT_INV_8X8U_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4
#endif
#endif
// store n
movq ARG8, %r10 // store address D
movq ARG9, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG13, %r12 // km
movq ARG14, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_8X8U_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8u_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8u_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4, .-kernel_dtrsm_nt_rl_inv_8x8u_vs_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
// void kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4(int kp, double *Ap, int sdap, double *Bp, int sdbp, int km, double *Am, int sdam, double *Bm, int sdbm, double *C, int sdc, double *D, int sdd, double *E, int sde, double *inv_diag_E, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4
.type kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4, @function
kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4
_kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4
.def kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4; .scl 2; .type 32; .endef
kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10
movq ARG2, %r11
movq ARG3, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG4, %r13
movq ARG5, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
movq ARG6, %r10
movq ARG7, %r11
movq ARG8, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG9, %r13
movq ARG10, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG11, %r10
movq ARG12, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_scale_11_8x8_lib4
#endif
#endif
// solve
movq ARG15, %r10 // E
movq ARG16, %r11 // sde
sall $5, %r11d // 4*sde*sizeof(double)
movq ARG17, %r12 // inv_diag_E
movq ARG19, %r13 // kn
#if MACRO_LEVEL>=1
INNER_EDGE_DTRSM_RLT_INV_8X8L_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dtrsm_rlt_inv_8x8l_vs_lib4
#endif
#endif
// store n
movq ARG13, %r10 // store address D
movq ARG14, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG18, %r12 // km
movq ARG19, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_8X8L_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8l_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8l_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4, .-kernel_dgemm_dtrsm_nt_rl_inv_8x8l_vs_lib4
#endif
// 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
// void kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4(int kp, double *Ap, int sdap, double *Bp, int sdbp, int km, double *Am, int sdam, double *Bm, int sdbm, double *C, int sdc, double *D, int sdd, double *E, int sde, double *inv_diag_E, int km, int kn);
.p2align 4,,15
#if defined(OS_LINUX)
.globl kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4
.type kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4, @function
kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4:
#elif defined(OS_MAC)
.globl _kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4
_kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4:
#elif defined(OS_WINDOWS)
.globl kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4
.def kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4; .scl 2; .type 32; .endef
kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4:
#endif
PROLOGUE
// zero accumulation registers
vxorpd %ymm0, %ymm0, %ymm0
vmovapd %ymm0, %ymm1
vmovapd %ymm0, %ymm2
vmovapd %ymm0, %ymm3
vmovapd %ymm0, %ymm4
vmovapd %ymm0, %ymm5
vmovapd %ymm0, %ymm6
vmovapd %ymm0, %ymm7
vmovapd %ymm0, %ymm8
vmovapd %ymm0, %ymm9
vmovapd %ymm0, %ymm10
vmovapd %ymm0, %ymm11
// call inner dgemm kernel nt
movq ARG1, %r10
movq ARG4, %r11
movq ARG5, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG2, %r13
movq ARG3, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_ADD_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_add_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_add_nt_8x8_lib4
#endif
#endif
movq ARG6, %r10
movq ARG9, %r11
movq ARG10, %r12
sall $5, %r12d // 4*sda*sizeof(double)
movq ARG7, %r13
movq ARG8, %r14
sall $5, %r14d // 4*sdb*sizeof(double)
#if MACRO_LEVEL>=2
INNER_KERNEL_DGEMM_SUB_NT_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_kernel_dgemm_sub_nt_8x8_lib4
#elif defined(OS_MAC)
callq _inner_kernel_dgemm_sub_nt_8x8_lib4
#endif
#endif
// call inner blender nn
movq ARG11, %r10 // C
movq ARG12, %r11 // sdc
sall $5, %r11d // 4*sdc*sizeof(double)
#if MACRO_LEVEL>=1
INNER_TRAN_SCALE_11_8X8_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_tran_scale_11_8x8_lib4
#elif defined(OS_MAC)
callq _inner_tran_scale_11_8x8_lib4
#endif
#endif
// solve
movq ARG15, %r10 // E
movq ARG16, %r11 // sde
sall $5, %r11d // 4*sde*sizeof(double)
movq ARG17, %r12 // inv_diag_E
movq ARG13, %r13 // D
movq ARG14, %r14 // sdd
sall $5, %r14d // 4*sdc*sizeof(double)
movq ARG19, %r15 // kn
#if MACRO_LEVEL>=1
INNER_EDGE_DTRSM_RLT_INV_8X8U_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4
#elif defined(OS_MAC)
callq _inner_edge_dtrsm_rlt_inv_8x8u_vs_lib4
#endif
#endif
// store n
movq ARG13, %r10 // store address D
movq ARG14, %r11 // sdd
sall $5, %r11d // 4*sdd*sizeof(double)
movq ARG18, %r12 // km
movq ARG19, %r13 // kn
#if MACRO_LEVEL>=1
INNER_STORE_8X8U_VS_LIB4
#else
#if defined(OS_LINUX) | defined(OS_WINDOWS)
call inner_store_8x8u_vs_lib4
#elif defined(OS_MAC)
callq _inner_store_8x8u_vs_lib4
#endif
#endif
EPILOGUE
ret
#if defined(OS_LINUX)
.size kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4, .-kernel_dgemm_dtrsm_nt_rl_inv_8x8u_vs_lib4
#endif
// read-only data
#if defined(OS_LINUX)
.section .rodata.cst32,"aM",@progbits,32
#elif defined(OS_MAC)
.section __TEXT,__const
#elif defined(OS_WINDOWS)
.section .rdata,"dr"
#endif
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC00: // { -1 -1 -1 1 }
#elif defined(OS_MAC)
LC00: // { -1 -1 -1 1 }
.align 5
#endif
.quad -1
.quad -1
.quad -1
.quad 1
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC01: // { -1 -1 -1 -1 }
#elif defined(OS_MAC)
LC01: // { -1 -1 -1 -1 }
.align 5
#endif
.quad -1
.quad -1
.quad -1
.quad -1
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC02: // { 3.5 2.5 1.5 0.5 }
#elif defined(OS_MAC)
LC02: // { 3.5 2.5 1.5 0.5 }
.align 5
#endif
.long 0
.long 1071644672
.long 0
.long 1073217536
.long 0
.long 1074003968
.long 0
.long 1074528256
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC03: // { 7.5 6.5 5.5 4.5 }
#elif defined(OS_MAC)
LC03: // { 7.5 6.5 5.5 4.5 }
.align 5
#endif
.long 0
.long 1074921472
.long 0
.long 1075183616
.long 0
.long 1075445760
.long 0
.long 1075707904
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC04: // { 1.0 1.0 1.0 1.0 }
#elif defined(OS_MAC)
LC04: // { 1.0 1.0 1.0 1.0 }
.align 5
#endif
.long 0
.long 1072693248
.long 0
.long 1072693248
.long 0
.long 1072693248
.long 0
.long 1072693248
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC05: // { 1.0 1.0 1.0 -1.0 }
#elif defined(OS_MAC)
.align 5
LC05: // { 1.0 1.0 1.0 -1.0 }
#endif
.long 0
.long -1074790400
.long 0
.long 1072693248
.long 0
.long 1072693248
.long 0
.long 1072693248
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC06: // { 1.0 1.0 -1.0 -1.0 }
#elif defined(OS_MAC)
.align 5
LC06: // { 1.0 1.0 -1.0 -1.0 }
#endif
.long 0
.long -1074790400
.long 0
.long -1074790400
.long 0
.long 1072693248
.long 0
.long 1072693248
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC07: // { 1.0 -1.0 -1.0 -1.0 }
#elif defined(OS_MAC)
.align 5
LC07: // { 1.0 -1.0 -1.0 -1.0 }
#endif
.long 0
.long -1074790400
.long 0
.long -1074790400
.long 0
.long -1074790400
.long 0
.long 1072693248
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC08: // { -1.0 -1.0 -1.0 1.0 }
#elif defined(OS_MAC)
.align 5
LC08: // { -1.0 -1.0 -1.0 1.0 }
#endif
.long 0
.long 1072693248
.long 0
.long -1074790400
.long 0
.long -1074790400
.long 0
.long -1074790400
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC09: // { -1.0 -1.0 1.0 1.0 }
#elif defined(OS_MAC)
.align 5
LC09: // { -1.0 -1.0 1.0 1.0 }
#endif
.long 0
.long 1072693248
.long 0
.long 1072693248
.long 0
.long -1074790400
.long 0
.long -1074790400
#if defined(OS_LINUX) | defined(OS_WINDOWS)
.align 32
.LC10: // { -1.0 1.0 1.0 1.0 }
#elif defined(OS_MAC)
.align 5
LC10: // { -1.0 1.0 1.0 1.0 }
#endif
.long 0
.long 1072693248
.long 0
.long 1072693248
.long 0
.long 1072693248
.long 0
.long -1074790400
#if defined(OS_LINUX)
.section .note.GNU-stack,"",@progbits
#elif defined(OS_MAC)
.subsections_via_symbols
#endif