1 | /* |
2 | * Copyright 2017-2018 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 "field.h" |
14 | |
15 | void gf_mul(gf_s * RESTRICT cs, const gf as, const gf bs) |
16 | { |
17 | const uint32_t *a = as->limb, *b = bs->limb; |
18 | uint32_t *c = cs->limb; |
19 | uint64_t accum0 = 0, accum1 = 0, accum2 = 0; |
20 | uint32_t mask = (1 << 28) - 1; |
21 | uint32_t aa[8], bb[8]; |
22 | int i, j; |
23 | |
24 | for (i = 0; i < 8; i++) { |
25 | aa[i] = a[i] + a[i + 8]; |
26 | bb[i] = b[i] + b[i + 8]; |
27 | } |
28 | |
29 | for (j = 0; j < 8; j++) { |
30 | accum2 = 0; |
31 | for (i = 0; i < j + 1; i++) { |
32 | accum2 += widemul(a[j - i], b[i]); |
33 | accum1 += widemul(aa[j - i], bb[i]); |
34 | accum0 += widemul(a[8 + j - i], b[8 + i]); |
35 | } |
36 | accum1 -= accum2; |
37 | accum0 += accum2; |
38 | accum2 = 0; |
39 | for (i = j + 1; i < 8; i++) { |
40 | accum0 -= widemul(a[8 + j - i], b[i]); |
41 | accum2 += widemul(aa[8 + j - i], bb[i]); |
42 | accum1 += widemul(a[16 + j - i], b[8 + i]); |
43 | } |
44 | accum1 += accum2; |
45 | accum0 += accum2; |
46 | c[j] = ((uint32_t)(accum0)) & mask; |
47 | c[j + 8] = ((uint32_t)(accum1)) & mask; |
48 | accum0 >>= 28; |
49 | accum1 >>= 28; |
50 | } |
51 | |
52 | accum0 += accum1; |
53 | accum0 += c[8]; |
54 | accum1 += c[0]; |
55 | c[8] = ((uint32_t)(accum0)) & mask; |
56 | c[0] = ((uint32_t)(accum1)) & mask; |
57 | |
58 | accum0 >>= 28; |
59 | accum1 >>= 28; |
60 | c[9] += ((uint32_t)(accum0)); |
61 | c[1] += ((uint32_t)(accum1)); |
62 | } |
63 | |
64 | void gf_mulw_unsigned(gf_s * RESTRICT cs, const gf as, uint32_t b) |
65 | { |
66 | const uint32_t *a = as->limb; |
67 | uint32_t *c = cs->limb; |
68 | uint64_t accum0 = 0, accum8 = 0; |
69 | uint32_t mask = (1 << 28) - 1; |
70 | int i; |
71 | |
72 | assert(b <= mask); |
73 | |
74 | for (i = 0; i < 8; i++) { |
75 | accum0 += widemul(b, a[i]); |
76 | accum8 += widemul(b, a[i + 8]); |
77 | c[i] = accum0 & mask; |
78 | accum0 >>= 28; |
79 | c[i + 8] = accum8 & mask; |
80 | accum8 >>= 28; |
81 | } |
82 | |
83 | accum0 += accum8 + c[8]; |
84 | c[8] = ((uint32_t)accum0) & mask; |
85 | c[9] += (uint32_t)(accum0 >> 28); |
86 | |
87 | accum8 += c[0]; |
88 | c[0] = ((uint32_t)accum8) & mask; |
89 | c[1] += (uint32_t)(accum8 >> 28); |
90 | } |
91 | |
92 | void gf_sqr(gf_s * RESTRICT cs, const gf as) |
93 | { |
94 | gf_mul(cs, as, as); /* Performs better with a dedicated square */ |
95 | } |
96 | |