1/*
2 * Copyright 2017-2021 The OpenSSL Project Authors. All Rights Reserved.
3 * Copyright 2014 Cryptography Research, Inc.
4 *
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
9 *
10 * Originally written by Mike Hamburg
11 */
12
13#include "e_os.h"
14#include <openssl/macros.h>
15#include "internal/numbers.h"
16
17#ifdef UINT128_MAX
18/* We have support for 128 bit ints, so do nothing here */
19NON_EMPTY_TRANSLATION_UNIT
20#else
21
22# include "../field.h"
23
24void gf_mul(gf_s * RESTRICT cs, const gf as, const gf bs)
25{
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];
31    int i, j;
32
33    for (i = 0; i < 8; i++) {
34        aa[i] = a[i] + a[i + 8];
35        bb[i] = b[i] + b[i + 8];
36    }
37
38    for (j = 0; j < 8; j++) {
39        accum2 = 0;
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]);
44        }
45        accum1 -= accum2;
46        accum0 += accum2;
47        accum2 = 0;
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]);
52        }
53        accum1 += accum2;
54        accum0 += accum2;
55        c[j] = ((uint32_t)(accum0)) & mask;
56        c[j + 8] = ((uint32_t)(accum1)) & mask;
57        accum0 >>= 28;
58        accum1 >>= 28;
59    }
60
61    accum0 += accum1;
62    accum0 += c[8];
63    accum1 += c[0];
64    c[8] = ((uint32_t)(accum0)) & mask;
65    c[0] = ((uint32_t)(accum1)) & mask;
66
67    accum0 >>= 28;
68    accum1 >>= 28;
69    c[9] += ((uint32_t)(accum0));
70    c[1] += ((uint32_t)(accum1));
71}
72
73void gf_mulw_unsigned(gf_s * RESTRICT cs, const gf as, uint32_t b)
74{
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;
79    int i;
80
81    assert(b <= mask);
82
83    for (i = 0; i < 8; i++) {
84        accum0 += widemul(b, a[i]);
85        accum8 += widemul(b, a[i + 8]);
86        c[i] = accum0 & mask;
87        accum0 >>= 28;
88        c[i + 8] = accum8 & mask;
89        accum8 >>= 28;
90    }
91
92    accum0 += accum8 + c[8];
93    c[8] = ((uint32_t)accum0) & mask;
94    c[9] += (uint32_t)(accum0 >> 28);
95
96    accum8 += c[0];
97    c[0] = ((uint32_t)accum8) & mask;
98    c[1] += (uint32_t)(accum8 >> 28);
99}
100
101void gf_sqr(gf_s * RESTRICT cs, const gf as)
102{
103    gf_mul(cs, as, as);         /* Performs better with a dedicated square */
104}
105#endif
106