Austin Schuh | dace2a6 | 2020-08-18 10:56:48 -0700 | [diff] [blame] | 1 | /* Alternate implementations of binvert_limb to compare speeds. */ |
| 2 | |
| 3 | /* |
| 4 | Copyright 2000, 2002 Free Software Foundation, Inc. |
| 5 | |
| 6 | This file is part of the GNU MP Library. |
| 7 | |
| 8 | The GNU MP Library is free software; you can redistribute it and/or modify |
| 9 | it under the terms of either: |
| 10 | |
| 11 | * the GNU Lesser General Public License as published by the Free |
| 12 | Software Foundation; either version 3 of the License, or (at your |
| 13 | option) any later version. |
| 14 | |
| 15 | or |
| 16 | |
| 17 | * the GNU General Public License as published by the Free Software |
| 18 | Foundation; either version 2 of the License, or (at your option) any |
| 19 | later version. |
| 20 | |
| 21 | or both in parallel, as here. |
| 22 | |
| 23 | The GNU MP Library is distributed in the hope that it will be useful, but |
| 24 | WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY |
| 25 | or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 26 | for more details. |
| 27 | |
| 28 | You should have received copies of the GNU General Public License and the |
| 29 | GNU Lesser General Public License along with the GNU MP Library. If not, |
| 30 | see https://www.gnu.org/licenses/. */ |
| 31 | |
| 32 | #include <stdio.h> |
| 33 | #include "gmp-impl.h" |
| 34 | #include "longlong.h" |
| 35 | #include "speed.h" |
| 36 | |
| 37 | |
| 38 | /* Like the standard version in gmp-impl.h, but with the expressions using a |
| 39 | "1-" form. This has the same number of steps, but "1-" is on the |
| 40 | dependent chain, whereas the "2*" in the standard version isn't. |
| 41 | Depending on the CPU this should be the same or a touch slower. */ |
| 42 | |
| 43 | #if GMP_LIMB_BITS <= 32 |
| 44 | #define binvert_limb_mul1(inv,n) \ |
| 45 | do { \ |
| 46 | mp_limb_t __n = (n); \ |
| 47 | mp_limb_t __inv; \ |
| 48 | ASSERT ((__n & 1) == 1); \ |
| 49 | __inv = binvert_limb_table[(__n&0xFF)/2]; /* 8 */ \ |
| 50 | __inv = (1 - __n * __inv) * __inv + __inv; /* 16 */ \ |
| 51 | __inv = (1 - __n * __inv) * __inv + __inv; /* 32 */ \ |
| 52 | ASSERT (__inv * __n == 1); \ |
| 53 | (inv) = __inv; \ |
| 54 | } while (0) |
| 55 | #endif |
| 56 | |
| 57 | #if GMP_LIMB_BITS > 32 && GMP_LIMB_BITS <= 64 |
| 58 | #define binvert_limb_mul1(inv,n) \ |
| 59 | do { \ |
| 60 | mp_limb_t __n = (n); \ |
| 61 | mp_limb_t __inv; \ |
| 62 | ASSERT ((__n & 1) == 1); \ |
| 63 | __inv = binvert_limb_table[(__n&0xFF)/2]; /* 8 */ \ |
| 64 | __inv = (1 - __n * __inv) * __inv + __inv; /* 16 */ \ |
| 65 | __inv = (1 - __n * __inv) * __inv + __inv; /* 32 */ \ |
| 66 | __inv = (1 - __n * __inv) * __inv + __inv; /* 64 */ \ |
| 67 | ASSERT (__inv * __n == 1); \ |
| 68 | (inv) = __inv; \ |
| 69 | } while (0) |
| 70 | #endif |
| 71 | |
| 72 | |
| 73 | /* The loop based version used in GMP 3.0 and earlier. Usually slower than |
| 74 | multiplying, due to the number of steps that must be performed. Much |
| 75 | slower when the processor has a good multiply. */ |
| 76 | |
| 77 | #define binvert_limb_loop(inv,n) \ |
| 78 | do { \ |
| 79 | mp_limb_t __v = (n); \ |
| 80 | mp_limb_t __v_orig = __v; \ |
| 81 | mp_limb_t __make_zero = 1; \ |
| 82 | mp_limb_t __two_i = 1; \ |
| 83 | mp_limb_t __v_inv = 0; \ |
| 84 | \ |
| 85 | ASSERT ((__v & 1) == 1); \ |
| 86 | \ |
| 87 | do \ |
| 88 | { \ |
| 89 | while ((__two_i & __make_zero) == 0) \ |
| 90 | __two_i <<= 1, __v <<= 1; \ |
| 91 | __v_inv += __two_i; \ |
| 92 | __make_zero -= __v; \ |
| 93 | } \ |
| 94 | while (__make_zero); \ |
| 95 | \ |
| 96 | ASSERT (__v_orig * __v_inv == 1); \ |
| 97 | (inv) = __v_inv; \ |
| 98 | } while (0) |
| 99 | |
| 100 | |
| 101 | /* Another loop based version with conditionals, but doing a fixed number of |
| 102 | steps. */ |
| 103 | |
| 104 | #define binvert_limb_cond(inv,n) \ |
| 105 | do { \ |
| 106 | mp_limb_t __n = (n); \ |
| 107 | mp_limb_t __rem = (1 - __n) >> 1; \ |
| 108 | mp_limb_t __inv = GMP_LIMB_HIGHBIT; \ |
| 109 | int __count; \ |
| 110 | \ |
| 111 | ASSERT ((__n & 1) == 1); \ |
| 112 | \ |
| 113 | __count = GMP_LIMB_BITS-1; \ |
| 114 | do \ |
| 115 | { \ |
| 116 | __inv >>= 1; \ |
| 117 | if (__rem & 1) \ |
| 118 | { \ |
| 119 | __inv |= GMP_LIMB_HIGHBIT; \ |
| 120 | __rem -= __n; \ |
| 121 | } \ |
| 122 | __rem >>= 1; \ |
| 123 | } \ |
| 124 | while (-- __count); \ |
| 125 | \ |
| 126 | ASSERT (__inv * __n == 1); \ |
| 127 | (inv) = __inv; \ |
| 128 | } while (0) |
| 129 | |
| 130 | |
| 131 | /* Another loop based bitwise version, but purely arithmetic, no |
| 132 | conditionals. */ |
| 133 | |
| 134 | #define binvert_limb_arith(inv,n) \ |
| 135 | do { \ |
| 136 | mp_limb_t __n = (n); \ |
| 137 | mp_limb_t __rem = (1 - __n) >> 1; \ |
| 138 | mp_limb_t __inv = GMP_LIMB_HIGHBIT; \ |
| 139 | mp_limb_t __lowbit; \ |
| 140 | int __count; \ |
| 141 | \ |
| 142 | ASSERT ((__n & 1) == 1); \ |
| 143 | \ |
| 144 | __count = GMP_LIMB_BITS-1; \ |
| 145 | do \ |
| 146 | { \ |
| 147 | __lowbit = __rem & 1; \ |
| 148 | __inv = (__inv >> 1) | (__lowbit << (GMP_LIMB_BITS-1)); \ |
| 149 | __rem = (__rem - (__n & -__lowbit)) >> 1; \ |
| 150 | } \ |
| 151 | while (-- __count); \ |
| 152 | \ |
| 153 | ASSERT (__inv * __n == 1); \ |
| 154 | (inv) = __inv; \ |
| 155 | } while (0) |
| 156 | |
| 157 | |
| 158 | double |
| 159 | speed_binvert_limb_mul1 (struct speed_params *s) |
| 160 | { |
| 161 | SPEED_ROUTINE_MODLIMB_INVERT (binvert_limb_mul1); |
| 162 | } |
| 163 | double |
| 164 | speed_binvert_limb_loop (struct speed_params *s) |
| 165 | { |
| 166 | SPEED_ROUTINE_MODLIMB_INVERT (binvert_limb_loop); |
| 167 | } |
| 168 | double |
| 169 | speed_binvert_limb_cond (struct speed_params *s) |
| 170 | { |
| 171 | SPEED_ROUTINE_MODLIMB_INVERT (binvert_limb_cond); |
| 172 | } |
| 173 | double |
| 174 | speed_binvert_limb_arith (struct speed_params *s) |
| 175 | { |
| 176 | SPEED_ROUTINE_MODLIMB_INVERT (binvert_limb_arith); |
| 177 | } |