1 /* Copyright (c) 2014 Cryptography Research, Inc.
2 * Released under the MIT License. See LICENSE.txt for license information.
3 */
4
5 #include "f_field.h"
6
7 #if (defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) && !I_HATE_UNROLLED_LOOPS) \
8 || defined(CRYPTONITE_DECAF_FORCE_UNROLL)
9 #define REPEAT8(_x) _x _x _x _x _x _x _x _x
10 #define FOR_LIMB(_i,_start,_end,_x) do { _i=_start; REPEAT8( if (_i<_end) { _x; } _i++;) } while (0)
11 #else
12 #define FOR_LIMB(_i,_start,_end,_x) do { for (_i=_start; _i<_end; _i++) _x; } while (0)
13 #endif
14
cryptonite_gf_mul(cryptonite_gf_s * __restrict__ cs,const gf as,const gf bs)15 void cryptonite_gf_mul (cryptonite_gf_s *__restrict__ cs, const gf as, const gf bs) {
16 const uint32_t *a = as->limb, *b = bs->limb;
17 uint32_t *c = cs->limb;
18
19 uint64_t accum0 = 0, accum1 = 0, accum2 = 0;
20 uint32_t mask = (1<<28) - 1;
21
22 uint32_t aa[8], bb[8];
23
24 int i,j;
25 for (i=0; i<8; i++) {
26 aa[i] = a[i] + a[i+8];
27 bb[i] = b[i] + b[i+8];
28 }
29
30 FOR_LIMB(j,0,8,{
31 accum2 = 0;
32
33 FOR_LIMB (i,0,j+1,{
34 accum2 += widemul(a[j-i],b[i]);
35 accum1 += widemul(aa[j-i],bb[i]);
36 accum0 += widemul(a[8+j-i], b[8+i]);
37 });
38
39 accum1 -= accum2;
40 accum0 += accum2;
41 accum2 = 0;
42
43 FOR_LIMB (i,j+1,8,{
44 accum0 -= widemul(a[8+j-i], b[i]);
45 accum2 += widemul(aa[8+j-i], bb[i]);
46 accum1 += widemul(a[16+j-i], b[8+i]);
47 });
48
49 accum1 += accum2;
50 accum0 += accum2;
51
52 c[j] = ((uint32_t)(accum0)) & mask;
53 c[j+8] = ((uint32_t)(accum1)) & mask;
54
55 accum0 >>= 28;
56 accum1 >>= 28;
57 });
58
59 accum0 += accum1;
60 accum0 += c[8];
61 accum1 += c[0];
62 c[8] = ((uint32_t)(accum0)) & mask;
63 c[0] = ((uint32_t)(accum1)) & mask;
64
65 accum0 >>= 28;
66 accum1 >>= 28;
67 c[9] += ((uint32_t)(accum0));
68 c[1] += ((uint32_t)(accum1));
69 }
70
cryptonite_gf_mulw_unsigned(cryptonite_gf_s * __restrict__ cs,const gf as,uint32_t b)71 void cryptonite_gf_mulw_unsigned (cryptonite_gf_s *__restrict__ cs, const gf as, uint32_t b) {
72 assert(b<1<<28);
73
74 const uint32_t *a = as->limb;
75 uint32_t *c = cs->limb;
76
77 uint64_t accum0 = 0, accum8 = 0;
78 uint32_t mask = (1ull<<28)-1;
79
80 int i;
81 FOR_LIMB(i,0,8,{
82 accum0 += widemul(b, a[i]);
83 accum8 += widemul(b, a[i+8]);
84
85 c[i] = accum0 & mask; accum0 >>= 28;
86 c[i+8] = accum8 & mask; accum8 >>= 28;
87 });
88
89 accum0 += accum8 + c[8];
90 c[8] = accum0 & mask;
91 c[9] += accum0 >> 28;
92
93 accum8 += c[0];
94 c[0] = accum8 & mask;
95 c[1] += accum8 >> 28;
96 }
97
cryptonite_gf_sqr(cryptonite_gf_s * __restrict__ cs,const gf as)98 void cryptonite_gf_sqr (cryptonite_gf_s *__restrict__ cs, const gf as) {
99 cryptonite_gf_mul(cs,as,as); /* Performs better with a dedicated square */
100 }
101
102