1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/security/nss/lib/freebl/mpi/mpi_sparc.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,224 @@
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 +/* Multiplication performance enhancements for sparc v8+vis CPUs. */
1.9 +
1.10 +#include "mpi-priv.h"
1.11 +#include <stddef.h>
1.12 +#include <sys/systeminfo.h>
1.13 +#include <strings.h>
1.14 +
1.15 +/* In the functions below, */
1.16 +/* vector y must be 8-byte aligned, and n must be even */
1.17 +/* returns carry out of high order word of result */
1.18 +/* maximum n is 256 */
1.19 +
1.20 +/* vector x += vector y * scaler a; where y is of length n words. */
1.21 +extern mp_digit mul_add_inp(mp_digit *x, const mp_digit *y, int n, mp_digit a);
1.22 +
1.23 +/* vector z = vector x + vector y * scaler a; where y is of length n words. */
1.24 +extern mp_digit mul_add(mp_digit *z, const mp_digit *x, const mp_digit *y,
1.25 + int n, mp_digit a);
1.26 +
1.27 +/* v8 versions of these functions run on any Sparc v8 CPU. */
1.28 +
1.29 +/* This trick works on Sparc V8 CPUs with the Workshop compilers. */
1.30 +#define MP_MUL_DxD(a, b, Phi, Plo) \
1.31 + { unsigned long long product = (unsigned long long)a * b; \
1.32 + Plo = (mp_digit)product; \
1.33 + Phi = (mp_digit)(product >> MP_DIGIT_BIT); }
1.34 +
1.35 +/* c = a * b */
1.36 +static void
1.37 +v8_mpv_mul_d(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.38 +{
1.39 +#if !defined(MP_NO_MP_WORD)
1.40 + mp_digit d = 0;
1.41 +
1.42 + /* Inner product: Digits of a */
1.43 + while (a_len--) {
1.44 + mp_word w = ((mp_word)b * *a++) + d;
1.45 + *c++ = ACCUM(w);
1.46 + d = CARRYOUT(w);
1.47 + }
1.48 + *c = d;
1.49 +#else
1.50 + mp_digit carry = 0;
1.51 + while (a_len--) {
1.52 + mp_digit a_i = *a++;
1.53 + mp_digit a0b0, a1b1;
1.54 +
1.55 + MP_MUL_DxD(a_i, b, a1b1, a0b0);
1.56 +
1.57 + a0b0 += carry;
1.58 + if (a0b0 < carry)
1.59 + ++a1b1;
1.60 + *c++ = a0b0;
1.61 + carry = a1b1;
1.62 + }
1.63 + *c = carry;
1.64 +#endif
1.65 +}
1.66 +
1.67 +/* c += a * b */
1.68 +static void
1.69 +v8_mpv_mul_d_add(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.70 +{
1.71 +#if !defined(MP_NO_MP_WORD)
1.72 + mp_digit d = 0;
1.73 +
1.74 + /* Inner product: Digits of a */
1.75 + while (a_len--) {
1.76 + mp_word w = ((mp_word)b * *a++) + *c + d;
1.77 + *c++ = ACCUM(w);
1.78 + d = CARRYOUT(w);
1.79 + }
1.80 + *c = d;
1.81 +#else
1.82 + mp_digit carry = 0;
1.83 + while (a_len--) {
1.84 + mp_digit a_i = *a++;
1.85 + mp_digit a0b0, a1b1;
1.86 +
1.87 + MP_MUL_DxD(a_i, b, a1b1, a0b0);
1.88 +
1.89 + a0b0 += carry;
1.90 + if (a0b0 < carry)
1.91 + ++a1b1;
1.92 + a0b0 += a_i = *c;
1.93 + if (a0b0 < a_i)
1.94 + ++a1b1;
1.95 + *c++ = a0b0;
1.96 + carry = a1b1;
1.97 + }
1.98 + *c = carry;
1.99 +#endif
1.100 +}
1.101 +
1.102 +/* Presently, this is only used by the Montgomery arithmetic code. */
1.103 +/* c += a * b */
1.104 +static void
1.105 +v8_mpv_mul_d_add_prop(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.106 +{
1.107 +#if !defined(MP_NO_MP_WORD)
1.108 + mp_digit d = 0;
1.109 +
1.110 + /* Inner product: Digits of a */
1.111 + while (a_len--) {
1.112 + mp_word w = ((mp_word)b * *a++) + *c + d;
1.113 + *c++ = ACCUM(w);
1.114 + d = CARRYOUT(w);
1.115 + }
1.116 +
1.117 + while (d) {
1.118 + mp_word w = (mp_word)*c + d;
1.119 + *c++ = ACCUM(w);
1.120 + d = CARRYOUT(w);
1.121 + }
1.122 +#else
1.123 + mp_digit carry = 0;
1.124 + while (a_len--) {
1.125 + mp_digit a_i = *a++;
1.126 + mp_digit a0b0, a1b1;
1.127 +
1.128 + MP_MUL_DxD(a_i, b, a1b1, a0b0);
1.129 +
1.130 + a0b0 += carry;
1.131 + if (a0b0 < carry)
1.132 + ++a1b1;
1.133 +
1.134 + a0b0 += a_i = *c;
1.135 + if (a0b0 < a_i)
1.136 + ++a1b1;
1.137 +
1.138 + *c++ = a0b0;
1.139 + carry = a1b1;
1.140 + }
1.141 + while (carry) {
1.142 + mp_digit c_i = *c;
1.143 + carry += c_i;
1.144 + *c++ = carry;
1.145 + carry = carry < c_i;
1.146 + }
1.147 +#endif
1.148 +}
1.149 +
1.150 +/* These functions run only on v8plus+vis or v9+vis CPUs. */
1.151 +
1.152 +/* c = a * b */
1.153 +void
1.154 +s_mpv_mul_d(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.155 +{
1.156 + mp_digit d;
1.157 + mp_digit x[258];
1.158 + if (a_len <= 256) {
1.159 + if (a == c || ((ptrdiff_t)a & 0x7) != 0 || (a_len & 1) != 0) {
1.160 + mp_digit * px;
1.161 + px = (((ptrdiff_t)x & 0x7) != 0) ? x + 1 : x;
1.162 + memcpy(px, a, a_len * sizeof(*a));
1.163 + a = px;
1.164 + if (a_len & 1) {
1.165 + px[a_len] = 0;
1.166 + }
1.167 + }
1.168 + s_mp_setz(c, a_len + 1);
1.169 + d = mul_add_inp(c, a, a_len, b);
1.170 + c[a_len] = d;
1.171 + } else {
1.172 + v8_mpv_mul_d(a, a_len, b, c);
1.173 + }
1.174 +}
1.175 +
1.176 +/* c += a * b, where a is a_len words long. */
1.177 +void
1.178 +s_mpv_mul_d_add(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.179 +{
1.180 + mp_digit d;
1.181 + mp_digit x[258];
1.182 + if (a_len <= 256) {
1.183 + if (((ptrdiff_t)a & 0x7) != 0 || (a_len & 1) != 0) {
1.184 + mp_digit * px;
1.185 + px = (((ptrdiff_t)x & 0x7) != 0) ? x + 1 : x;
1.186 + memcpy(px, a, a_len * sizeof(*a));
1.187 + a = px;
1.188 + if (a_len & 1) {
1.189 + px[a_len] = 0;
1.190 + }
1.191 + }
1.192 + d = mul_add_inp(c, a, a_len, b);
1.193 + c[a_len] = d;
1.194 + } else {
1.195 + v8_mpv_mul_d_add(a, a_len, b, c);
1.196 + }
1.197 +}
1.198 +
1.199 +/* c += a * b, where a is y words long. */
1.200 +void
1.201 +s_mpv_mul_d_add_prop(const mp_digit *a, mp_size a_len, mp_digit b, mp_digit *c)
1.202 +{
1.203 + mp_digit d;
1.204 + mp_digit x[258];
1.205 + if (a_len <= 256) {
1.206 + if (((ptrdiff_t)a & 0x7) != 0 || (a_len & 1) != 0) {
1.207 + mp_digit * px;
1.208 + px = (((ptrdiff_t)x & 0x7) != 0) ? x + 1 : x;
1.209 + memcpy(px, a, a_len * sizeof(*a));
1.210 + a = px;
1.211 + if (a_len & 1) {
1.212 + px[a_len] = 0;
1.213 + }
1.214 + }
1.215 + d = mul_add_inp(c, a, a_len, b);
1.216 + if (d) {
1.217 + c += a_len;
1.218 + do {
1.219 + mp_digit sum = d + *c;
1.220 + *c++ = sum;
1.221 + d = sum < d;
1.222 + } while (d);
1.223 + }
1.224 + } else {
1.225 + v8_mpv_mul_d_add_prop(a, a_len, b, c);
1.226 + }
1.227 +}
