2 * Copyright 2013-2016 The OpenSSL Project Authors. All Rights Reserved.
3 * Copyright (c) 2012, Intel Corporation. All Rights Reserved.
5 * Licensed under the OpenSSL license (the "License"). You may not use
6 * this file except in compliance with the License. You can obtain a copy
7 * in the file LICENSE in the source distribution or at
8 * https://www.openssl.org/source/license.html
10 * Originally written by Shay Gueron (1, 2), and Vlad Krasnov (1)
11 * (1) Intel Corporation, Israel Development Center, Haifa, Israel
12 * (2) University of Haifa, Israel
15 #include <openssl/opensslconf.h>
19 NON_EMPTY_TRANSLATION_UNIT
23 * See crypto/bn/asm/rsaz-avx2.pl for further details.
25 void rsaz_1024_norm2red_avx2(void *red, const void *norm);
26 void rsaz_1024_mul_avx2(void *ret, const void *a, const void *b,
27 const void *n, BN_ULONG k);
28 void rsaz_1024_sqr_avx2(void *ret, const void *a, const void *n, BN_ULONG k,
30 void rsaz_1024_scatter5_avx2(void *tbl, const void *val, int i);
31 void rsaz_1024_gather5_avx2(void *val, const void *tbl, int i);
32 void rsaz_1024_red2norm_avx2(void *norm, const void *red);
35 # define ALIGN64 __attribute__((aligned(64)))
36 #elif defined(_MSC_VER)
37 # define ALIGN64 __declspec(align(64))
38 #elif defined(__SUNPRO_C)
40 # pragma align 64(one,two80)
42 /* not fatal, might hurt performance a little */
46 ALIGN64 static const BN_ULONG one[40] = {
47 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
48 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
51 ALIGN64 static const BN_ULONG two80[40] = {
52 0, 0, 1 << 22, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
53 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
56 void RSAZ_1024_mod_exp_avx2(BN_ULONG result_norm[16],
57 const BN_ULONG base_norm[16],
58 const BN_ULONG exponent[16],
59 const BN_ULONG m_norm[16], const BN_ULONG RR[16],
62 unsigned char storage[320 * 3 + 32 * 9 * 16 + 64]; /* 5.5KB */
63 unsigned char *p_str = storage + (64 - ((size_t)storage % 64));
64 unsigned char *a_inv, *m, *result;
65 unsigned char *table_s = p_str + 320 * 3;
66 unsigned char *R2 = table_s; /* borrow */
70 if ((((size_t)p_str & 4095) + 320) >> 12) {
73 m = p_str + 320 * 2; /* should not cross page */
75 m = p_str; /* should not cross page */
77 a_inv = p_str + 320 * 2;
80 rsaz_1024_norm2red_avx2(m, m_norm);
81 rsaz_1024_norm2red_avx2(a_inv, base_norm);
82 rsaz_1024_norm2red_avx2(R2, RR);
84 rsaz_1024_mul_avx2(R2, R2, R2, m, k0);
85 rsaz_1024_mul_avx2(R2, R2, two80, m, k0);
88 rsaz_1024_mul_avx2(result, R2, one, m, k0);
89 /* table[1] = a_inv^1 */
90 rsaz_1024_mul_avx2(a_inv, a_inv, R2, m, k0);
92 rsaz_1024_scatter5_avx2(table_s, result, 0);
93 rsaz_1024_scatter5_avx2(table_s, a_inv, 1);
95 /* table[2] = a_inv^2 */
96 rsaz_1024_sqr_avx2(result, a_inv, m, k0, 1);
97 rsaz_1024_scatter5_avx2(table_s, result, 2);
99 /* this is almost 2x smaller and less than 1% slower */
100 for (index = 3; index < 32; index++) {
101 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
102 rsaz_1024_scatter5_avx2(table_s, result, index);
105 /* table[4] = a_inv^4 */
106 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
107 rsaz_1024_scatter5_avx2(table_s, result, 4);
108 /* table[8] = a_inv^8 */
109 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
110 rsaz_1024_scatter5_avx2(table_s, result, 8);
111 /* table[16] = a_inv^16 */
112 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
113 rsaz_1024_scatter5_avx2(table_s, result, 16);
114 /* table[17] = a_inv^17 */
115 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
116 rsaz_1024_scatter5_avx2(table_s, result, 17);
119 rsaz_1024_gather5_avx2(result, table_s, 2);
120 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
121 rsaz_1024_scatter5_avx2(table_s, result, 3);
123 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
124 rsaz_1024_scatter5_avx2(table_s, result, 6);
126 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
127 rsaz_1024_scatter5_avx2(table_s, result, 12);
129 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
130 rsaz_1024_scatter5_avx2(table_s, result, 24);
132 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
133 rsaz_1024_scatter5_avx2(table_s, result, 25);
136 rsaz_1024_gather5_avx2(result, table_s, 4);
137 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
138 rsaz_1024_scatter5_avx2(table_s, result, 5);
140 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
141 rsaz_1024_scatter5_avx2(table_s, result, 10);
143 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
144 rsaz_1024_scatter5_avx2(table_s, result, 20);
146 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
147 rsaz_1024_scatter5_avx2(table_s, result, 21);
150 rsaz_1024_gather5_avx2(result, table_s, 6);
151 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
152 rsaz_1024_scatter5_avx2(table_s, result, 7);
154 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
155 rsaz_1024_scatter5_avx2(table_s, result, 14);
157 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
158 rsaz_1024_scatter5_avx2(table_s, result, 28);
160 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
161 rsaz_1024_scatter5_avx2(table_s, result, 29);
164 rsaz_1024_gather5_avx2(result, table_s, 8);
165 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
166 rsaz_1024_scatter5_avx2(table_s, result, 9);
168 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
169 rsaz_1024_scatter5_avx2(table_s, result, 18);
171 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
172 rsaz_1024_scatter5_avx2(table_s, result, 19);
175 rsaz_1024_gather5_avx2(result, table_s, 10);
176 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
177 rsaz_1024_scatter5_avx2(table_s, result, 11);
179 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
180 rsaz_1024_scatter5_avx2(table_s, result, 22);
182 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
183 rsaz_1024_scatter5_avx2(table_s, result, 23);
186 rsaz_1024_gather5_avx2(result, table_s, 12);
187 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
188 rsaz_1024_scatter5_avx2(table_s, result, 13);
190 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
191 rsaz_1024_scatter5_avx2(table_s, result, 26);
193 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
194 rsaz_1024_scatter5_avx2(table_s, result, 27);
197 rsaz_1024_gather5_avx2(result, table_s, 14);
198 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
199 rsaz_1024_scatter5_avx2(table_s, result, 15);
201 rsaz_1024_sqr_avx2(result, result, m, k0, 1);
202 rsaz_1024_scatter5_avx2(table_s, result, 30);
204 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
205 rsaz_1024_scatter5_avx2(table_s, result, 31);
208 /* load first window */
209 p_str = (unsigned char *)exponent;
210 wvalue = p_str[127] >> 3;
211 rsaz_1024_gather5_avx2(result, table_s, wvalue);
215 while (index > -1) { /* loop for the remaining 127 windows */
217 rsaz_1024_sqr_avx2(result, result, m, k0, 5);
219 wvalue = (p_str[(index / 8) + 1] << 8) | p_str[index / 8];
220 wvalue = (wvalue >> (index % 8)) & 31;
223 rsaz_1024_gather5_avx2(a_inv, table_s, wvalue); /* borrow a_inv */
224 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
227 /* square four times */
228 rsaz_1024_sqr_avx2(result, result, m, k0, 4);
230 wvalue = p_str[0] & 15;
232 rsaz_1024_gather5_avx2(a_inv, table_s, wvalue); /* borrow a_inv */
233 rsaz_1024_mul_avx2(result, result, a_inv, m, k0);
235 /* from Montgomery */
236 rsaz_1024_mul_avx2(result, result, one, m, k0);
238 rsaz_1024_red2norm_avx2(result_norm, result);
240 OPENSSL_cleanse(storage, sizeof(storage));
244 * See crypto/bn/rsaz-x86_64.pl for further details.
246 void rsaz_512_mul(void *ret, const void *a, const void *b, const void *n,
248 void rsaz_512_mul_scatter4(void *ret, const void *a, const void *n,
249 BN_ULONG k, const void *tbl, unsigned int power);
250 void rsaz_512_mul_gather4(void *ret, const void *a, const void *tbl,
251 const void *n, BN_ULONG k, unsigned int power);
252 void rsaz_512_mul_by_one(void *ret, const void *a, const void *n, BN_ULONG k);
253 void rsaz_512_sqr(void *ret, const void *a, const void *n, BN_ULONG k,
255 void rsaz_512_scatter4(void *tbl, const BN_ULONG *val, int power);
256 void rsaz_512_gather4(BN_ULONG *val, const void *tbl, int power);
258 void RSAZ_512_mod_exp(BN_ULONG result[8],
259 const BN_ULONG base[8], const BN_ULONG exponent[8],
260 const BN_ULONG m[8], BN_ULONG k0, const BN_ULONG RR[8])
262 unsigned char storage[16 * 8 * 8 + 64 * 2 + 64]; /* 1.2KB */
263 unsigned char *table = storage + (64 - ((size_t)storage % 64));
264 BN_ULONG *a_inv = (BN_ULONG *)(table + 16 * 8 * 8);
265 BN_ULONG *temp = (BN_ULONG *)(table + 16 * 8 * 8 + 8 * 8);
266 unsigned char *p_str = (unsigned char *)exponent;
270 /* table[0] = 1_inv */
279 rsaz_512_scatter4(table, temp, 0);
281 /* table [1] = a_inv^1 */
282 rsaz_512_mul(a_inv, base, RR, m, k0);
283 rsaz_512_scatter4(table, a_inv, 1);
285 /* table [2] = a_inv^2 */
286 rsaz_512_sqr(temp, a_inv, m, k0, 1);
287 rsaz_512_scatter4(table, temp, 2);
289 for (index = 3; index < 16; index++)
290 rsaz_512_mul_scatter4(temp, a_inv, m, k0, table, index);
292 /* load first window */
295 rsaz_512_gather4(temp, table, wvalue >> 4);
296 rsaz_512_sqr(temp, temp, m, k0, 4);
297 rsaz_512_mul_gather4(temp, temp, table, m, k0, wvalue & 0xf);
299 for (index = 62; index >= 0; index--) {
300 wvalue = p_str[index];
302 rsaz_512_sqr(temp, temp, m, k0, 4);
303 rsaz_512_mul_gather4(temp, temp, table, m, k0, wvalue >> 4);
305 rsaz_512_sqr(temp, temp, m, k0, 4);
306 rsaz_512_mul_gather4(temp, temp, table, m, k0, wvalue & 0x0f);
309 /* from Montgomery */
310 rsaz_512_mul_by_one(result, temp, m, k0);
312 OPENSSL_cleanse(storage, sizeof(storage));