1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/security/nss/lib/freebl/mpi/mpvalpha.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,181 @@
1.4 +/* This Source Code Form is subject to the terms of the Mozilla Public
1.5 + * License, v. 2.0. If a copy of the MPL was not distributed with this
1.6 + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
1.7 +
1.8 +#include "mpi-priv.h"
1.9 +#include <c_asm.h>
1.10 +
1.11 +
1.12 +#define MP_MUL_DxD(a, b, Phi, Plo) \
1.13 + { Plo = asm ("mulq %a0, %a1, %v0", a, b); \
1.14 + Phi = asm ("umulh %a0, %a1, %v0", a, b); } \
1.15 +
1.16 +/* This is empty for the loop in s_mpv_mul_d */
1.17 +#define CARRY_ADD
1.18 +
1.19 +#define ONE_MUL \
1.20 + a_i = *a++; \
1.21 + MP_MUL_DxD(a_i, b, a1b1, a0b0); \
1.22 + a0b0 += carry; \
1.23 + if (a0b0 < carry) \
1.24 + ++a1b1; \
1.25 + CARRY_ADD \
1.26 + *c++ = a0b0; \
1.27 + carry = a1b1; \
1.28 +
1.29 +#define FOUR_MUL \
1.30 + ONE_MUL \
1.31 + ONE_MUL \
1.32 + ONE_MUL \
1.33 + ONE_MUL \
1.34 +
1.35 +#define SIXTEEN_MUL \
1.36 + FOUR_MUL \
1.37 + FOUR_MUL \
1.38 + FOUR_MUL \
1.39 + FOUR_MUL \
1.40 +
1.41 +#define THIRTYTWO_MUL \
1.42 + SIXTEEN_MUL \
1.43 + SIXTEEN_MUL \
1.44 +
1.45 +#define ONETWENTYEIGHT_MUL \
1.46 + THIRTYTWO_MUL \
1.47 + THIRTYTWO_MUL \
1.48 + THIRTYTWO_MUL \
1.49 + THIRTYTWO_MUL \
1.50 +
1.51 +
1.52 +#define EXPAND_256(CALL) \
1.53 + mp_digit carry = 0; \
1.54 + mp_digit a_i; \
1.55 + mp_digit a0b0, a1b1; \
1.56 + if (a_len &255) { \
1.57 + if (a_len &1) { \
1.58 + ONE_MUL \
1.59 + } \
1.60 + if (a_len &2) { \
1.61 + ONE_MUL \
1.62 + ONE_MUL \
1.63 + } \
1.64 + if (a_len &4) { \
1.65 + FOUR_MUL \
1.66 + } \
1.67 + if (a_len &8) { \
1.68 + FOUR_MUL \
1.69 + FOUR_MUL \
1.70 + } \
1.71 + if (a_len & 16 ) { \
1.72 + SIXTEEN_MUL \
1.73 + } \
1.74 + if (a_len & 32 ) { \
1.75 + THIRTYTWO_MUL \
1.76 + } \
1.77 + if (a_len & 64 ) { \
1.78 + THIRTYTWO_MUL \
1.79 + THIRTYTWO_MUL \
1.80 + } \
1.81 + if (a_len & 128) { \
1.82 + ONETWENTYEIGHT_MUL \
1.83 + } \
1.84 + a_len = a_len & (-256); \
1.85 + } \
1.86 + if (a_len>=256 ) { \
1.87 + carry = CALL(a, a_len, b, c, carry); \
1.88 + c += a_len; \
1.89 + } \
1.90 +
1.91 +#define FUNC_NAME(NAME) \
1.92 +mp_digit NAME(const mp_digit *a, \
1.93 + mp_size a_len, \
1.94 + mp_digit b, mp_digit *c, \
1.95 + mp_digit carry) \
1.96 +
1.97 +#define DECLARE_MUL_256(FNAME) \
1.98 +FUNC_NAME(FNAME) \
1.99 +{ \
1.100 + mp_digit a_i; \
1.101 + mp_digit a0b0, a1b1; \
1.102 + while (a_len) { \
1.103 + ONETWENTYEIGHT_MUL \
1.104 + ONETWENTYEIGHT_MUL \
1.105 + a_len-= 256; \
1.106 + } \
1.107 + return carry; \
1.108 +} \
1.109 +
1.110 +/* Expanding the loop in s_mpv_mul_d appeared to slow down the
1.111 + (admittedly) small number of tests (i.e., timetest) used to
1.112 + measure performance, so this define disables that optimization. */
1.113 +#define DO_NOT_EXPAND 1
1.114 +
1.115 +/* Need forward declaration so it can be instantiated after
1.116 + the routine that uses it; this helps locality somewhat */
1.117 +#if !defined(DO_NOT_EXPAND)
1.118 +FUNC_NAME(s_mpv_mul_d_MUL256);
1.119 +#endif
1.120 +
1.121 +/* c = a * b */
1.122 +void s_mpv_mul_d(const mp_digit *a, mp_size a_len,
1.123 + mp_digit b, mp_digit *c)
1.124 +{
1.125 +#if defined(DO_NOT_EXPAND)
1.126 + mp_digit carry = 0;
1.127 + while (a_len--) {
1.128 + mp_digit a_i = *a++;
1.129 + mp_digit a0b0, a1b1;
1.130 +
1.131 + MP_MUL_DxD(a_i, b, a1b1, a0b0);
1.132 +
1.133 + a0b0 += carry;
1.134 + if (a0b0 < carry)
1.135 + ++a1b1;
1.136 + *c++ = a0b0;
1.137 + carry = a1b1;
1.138 + }
1.139 +#else
1.140 + EXPAND_256(s_mpv_mul_d_MUL256)
1.141 +#endif
1.142 + *c = carry;
1.143 +}
1.144 +
1.145 +#if !defined(DO_NOT_EXPAND)
1.146 +DECLARE_MUL_256(s_mpv_mul_d_MUL256)
1.147 +#endif
1.148 +
1.149 +#undef CARRY_ADD
1.150 +/* This is redefined for the loop in s_mpv_mul_d_add */
1.151 +#define CARRY_ADD \
1.152 + a0b0 += a_i = *c; \
1.153 + if (a0b0 < a_i) \
1.154 + ++a1b1; \
1.155 +
1.156 +/* Need forward declaration so it can be instantiated between the
1.157 + two routines that use it; this helps locality somewhat */
1.158 +FUNC_NAME(s_mpv_mul_d_add_MUL256);
1.159 +
1.160 +/* c += a * b */
1.161 +void s_mpv_mul_d_add(const mp_digit *a, mp_size a_len,
1.162 + mp_digit b, mp_digit *c)
1.163 +{
1.164 + EXPAND_256(s_mpv_mul_d_add_MUL256)
1.165 + *c = carry;
1.166 +}
1.167 +
1.168 +/* Instantiate multiply 256 routine here */
1.169 +DECLARE_MUL_256(s_mpv_mul_d_add_MUL256)
1.170 +
1.171 +/* Presently, this is only used by the Montgomery arithmetic code. */
1.172 +/* c += a * b */
1.173 +void s_mpv_mul_d_add_prop(const mp_digit *a, mp_size a_len,
1.174 + mp_digit b, mp_digit *c)
1.175 +{
1.176 + EXPAND_256(s_mpv_mul_d_add_MUL256)
1.177 + while (carry) {
1.178 + mp_digit c_i = *c;
1.179 + carry += c_i;
1.180 + *c++ = carry;
1.181 + carry = carry < c_i;
1.182 + }
1.183 +}
1.184 +
