The Tor Browser / file comparison
comparison: security/nss/lib/freebl/mpi/mpvalpha.c
security/nss/lib/freebl/mpi/mpvalpha.c
- changeset 0
- 6474c204b198
equal
deleted
inserted
replaced
| -1:000000000000 | 0:7e42f761168f |
|---|---|
| 1 /* This Source Code Form is subject to the terms of the Mozilla Public | |
| 2 * License, v. 2.0. If a copy of the MPL was not distributed with this | |
| 3 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ | |
| 4 | |
| 5 #include "mpi-priv.h" | |
| 6 #include <c_asm.h> | |
| 7 | |
| 8 | |
| 9 #define MP_MUL_DxD(a, b, Phi, Plo) \ | |
| 10 { Plo = asm ("mulq %a0, %a1, %v0", a, b); \ | |
| 11 Phi = asm ("umulh %a0, %a1, %v0", a, b); } \ | |
| 12 | |
| 13 /* This is empty for the loop in s_mpv_mul_d */ | |
| 14 #define CARRY_ADD | |
| 15 | |
| 16 #define ONE_MUL \ | |
| 17 a_i = *a++; \ | |
| 18 MP_MUL_DxD(a_i, b, a1b1, a0b0); \ | |
| 19 a0b0 += carry; \ | |
| 20 if (a0b0 < carry) \ | |
| 21 ++a1b1; \ | |
| 22 CARRY_ADD \ | |
| 23 *c++ = a0b0; \ | |
| 24 carry = a1b1; \ | |
| 25 | |
| 26 #define FOUR_MUL \ | |
| 27 ONE_MUL \ | |
| 28 ONE_MUL \ | |
| 29 ONE_MUL \ | |
| 30 ONE_MUL \ | |
| 31 | |
| 32 #define SIXTEEN_MUL \ | |
| 33 FOUR_MUL \ | |
| 34 FOUR_MUL \ | |
| 35 FOUR_MUL \ | |
| 36 FOUR_MUL \ | |
| 37 | |
| 38 #define THIRTYTWO_MUL \ | |
| 39 SIXTEEN_MUL \ | |
| 40 SIXTEEN_MUL \ | |
| 41 | |
| 42 #define ONETWENTYEIGHT_MUL \ | |
| 43 THIRTYTWO_MUL \ | |
| 44 THIRTYTWO_MUL \ | |
| 45 THIRTYTWO_MUL \ | |
| 46 THIRTYTWO_MUL \ | |
| 47 | |
| 48 | |
| 49 #define EXPAND_256(CALL) \ | |
| 50 mp_digit carry = 0; \ | |
| 51 mp_digit a_i; \ | |
| 52 mp_digit a0b0, a1b1; \ | |
| 53 if (a_len &255) { \ | |
| 54 if (a_len &1) { \ | |
| 55 ONE_MUL \ | |
| 56 } \ | |
| 57 if (a_len &2) { \ | |
| 58 ONE_MUL \ | |
| 59 ONE_MUL \ | |
| 60 } \ | |
| 61 if (a_len &4) { \ | |
| 62 FOUR_MUL \ | |
| 63 } \ | |
| 64 if (a_len &8) { \ | |
| 65 FOUR_MUL \ | |
| 66 FOUR_MUL \ | |
| 67 } \ | |
| 68 if (a_len & 16 ) { \ | |
| 69 SIXTEEN_MUL \ | |
| 70 } \ | |
| 71 if (a_len & 32 ) { \ | |
| 72 THIRTYTWO_MUL \ | |
| 73 } \ | |
| 74 if (a_len & 64 ) { \ | |
| 75 THIRTYTWO_MUL \ | |
| 76 THIRTYTWO_MUL \ | |
| 77 } \ | |
| 78 if (a_len & 128) { \ | |
| 79 ONETWENTYEIGHT_MUL \ | |
| 80 } \ | |
| 81 a_len = a_len & (-256); \ | |
| 82 } \ | |
| 83 if (a_len>=256 ) { \ | |
| 84 carry = CALL(a, a_len, b, c, carry); \ | |
| 85 c += a_len; \ | |
| 86 } \ | |
| 87 | |
| 88 #define FUNC_NAME(NAME) \ | |
| 89 mp_digit NAME(const mp_digit *a, \ | |
| 90 mp_size a_len, \ | |
| 91 mp_digit b, mp_digit *c, \ | |
| 92 mp_digit carry) \ | |
| 93 | |
| 94 #define DECLARE_MUL_256(FNAME) \ | |
| 95 FUNC_NAME(FNAME) \ | |
| 96 { \ | |
| 97 mp_digit a_i; \ | |
| 98 mp_digit a0b0, a1b1; \ | |
| 99 while (a_len) { \ | |
| 100 ONETWENTYEIGHT_MUL \ | |
| 101 ONETWENTYEIGHT_MUL \ | |
| 102 a_len-= 256; \ | |
| 103 } \ | |
| 104 return carry; \ | |
| 105 } \ | |
| 106 | |
| 107 /* Expanding the loop in s_mpv_mul_d appeared to slow down the | |
| 108 (admittedly) small number of tests (i.e., timetest) used to | |
| 109 measure performance, so this define disables that optimization. */ | |
| 110 #define DO_NOT_EXPAND 1 | |
| 111 | |
| 112 /* Need forward declaration so it can be instantiated after | |
| 113 the routine that uses it; this helps locality somewhat */ | |
| 114 #if !defined(DO_NOT_EXPAND) | |
| 115 FUNC_NAME(s_mpv_mul_d_MUL256); | |
| 116 #endif | |
| 117 | |
| 118 /* c = a * b */ | |
| 119 void s_mpv_mul_d(const mp_digit *a, mp_size a_len, | |
| 120 mp_digit b, mp_digit *c) | |
| 121 { | |
| 122 #if defined(DO_NOT_EXPAND) | |
| 123 mp_digit carry = 0; | |
| 124 while (a_len--) { | |
| 125 mp_digit a_i = *a++; | |
| 126 mp_digit a0b0, a1b1; | |
| 127 | |
| 128 MP_MUL_DxD(a_i, b, a1b1, a0b0); | |
| 129 | |
| 130 a0b0 += carry; | |
| 131 if (a0b0 < carry) | |
| 132 ++a1b1; | |
| 133 *c++ = a0b0; | |
| 134 carry = a1b1; | |
| 135 } | |
| 136 #else | |
| 137 EXPAND_256(s_mpv_mul_d_MUL256) | |
| 138 #endif | |
| 139 *c = carry; | |
| 140 } | |
| 141 | |
| 142 #if !defined(DO_NOT_EXPAND) | |
| 143 DECLARE_MUL_256(s_mpv_mul_d_MUL256) | |
| 144 #endif | |
| 145 | |
| 146 #undef CARRY_ADD | |
| 147 /* This is redefined for the loop in s_mpv_mul_d_add */ | |
| 148 #define CARRY_ADD \ | |
| 149 a0b0 += a_i = *c; \ | |
| 150 if (a0b0 < a_i) \ | |
| 151 ++a1b1; \ | |
| 152 | |
| 153 /* Need forward declaration so it can be instantiated between the | |
| 154 two routines that use it; this helps locality somewhat */ | |
| 155 FUNC_NAME(s_mpv_mul_d_add_MUL256); | |
| 156 | |
| 157 /* c += a * b */ | |
| 158 void s_mpv_mul_d_add(const mp_digit *a, mp_size a_len, | |
| 159 mp_digit b, mp_digit *c) | |
| 160 { | |
| 161 EXPAND_256(s_mpv_mul_d_add_MUL256) | |
| 162 *c = carry; | |
| 163 } | |
| 164 | |
| 165 /* Instantiate multiply 256 routine here */ | |
| 166 DECLARE_MUL_256(s_mpv_mul_d_add_MUL256) | |
| 167 | |
| 168 /* Presently, this is only used by the Montgomery arithmetic code. */ | |
| 169 /* c += a * b */ | |
| 170 void s_mpv_mul_d_add_prop(const mp_digit *a, mp_size a_len, | |
| 171 mp_digit b, mp_digit *c) | |
| 172 { | |
| 173 EXPAND_256(s_mpv_mul_d_add_MUL256) | |
| 174 while (carry) { | |
| 175 mp_digit c_i = *c; | |
| 176 carry += c_i; | |
| 177 *c++ = carry; | |
| 178 carry = carry < c_i; | |
| 179 } | |
| 180 } | |
| 181 |
