2 * Copyright 2017-2021 The OpenSSL Project Authors. All Rights Reserved.
3 * Copyright 2014 Cryptography Research, Inc.
5 * Licensed under the Apache License 2.0 (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 Mike Hamburg
14 #include <openssl/macros.h>
15 #include "internal/numbers.h"
18 /* We have support for 128 bit ints, so do nothing here */
19 NON_EMPTY_TRANSLATION_UNIT
22 # include "../field.h"
24 void gf_mul(gf_s * RESTRICT cs, const gf as, const gf bs)
26 const uint32_t *a = as->limb, *b = bs->limb;
27 uint32_t *c = cs->limb;
28 uint64_t accum0 = 0, accum1 = 0, accum2 = 0;
29 uint32_t mask = (1 << 28) - 1;
30 uint32_t aa[8], bb[8];
33 for (i = 0; i < 8; i++) {
34 aa[i] = a[i] + a[i + 8];
35 bb[i] = b[i] + b[i + 8];
38 for (j = 0; j < 8; j++) {
40 for (i = 0; i < j + 1; i++) {
41 accum2 += widemul(a[j - i], b[i]);
42 accum1 += widemul(aa[j - i], bb[i]);
43 accum0 += widemul(a[8 + j - i], b[8 + i]);
48 for (i = j + 1; i < 8; i++) {
49 accum0 -= widemul(a[8 + j - i], b[i]);
50 accum2 += widemul(aa[8 + j - i], bb[i]);
51 accum1 += widemul(a[16 + j - i], b[8 + i]);
55 c[j] = ((uint32_t)(accum0)) & mask;
56 c[j + 8] = ((uint32_t)(accum1)) & mask;
64 c[8] = ((uint32_t)(accum0)) & mask;
65 c[0] = ((uint32_t)(accum1)) & mask;
69 c[9] += ((uint32_t)(accum0));
70 c[1] += ((uint32_t)(accum1));
73 void gf_mulw_unsigned(gf_s * RESTRICT cs, const gf as, uint32_t b)
75 const uint32_t *a = as->limb;
76 uint32_t *c = cs->limb;
77 uint64_t accum0 = 0, accum8 = 0;
78 uint32_t mask = (1 << 28) - 1;
83 for (i = 0; i < 8; i++) {
84 accum0 += widemul(b, a[i]);
85 accum8 += widemul(b, a[i + 8]);
88 c[i + 8] = accum8 & mask;
92 accum0 += accum8 + c[8];
93 c[8] = ((uint32_t)accum0) & mask;
94 c[9] += (uint32_t)(accum0 >> 28);
97 c[0] = ((uint32_t)accum8) & mask;
98 c[1] += (uint32_t)(accum8 >> 28);
101 void gf_sqr(gf_s * RESTRICT cs, const gf as)
103 gf_mul(cs, as, as); /* Performs better with a dedicated square */