1 /* Copyright (c) 2014 Cryptography Research, Inc.
2 * Released under the MIT License. See LICENSE.txt for license information.
7 #if (defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) && !I_HATE_UNROLLED_LOOPS) \
8 || defined(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)
12 #define FOR_LIMB(_i,_start,_end,_x) do { for (_i=_start; _i<_end; _i++) _x; } while (0)
15 void gf_mul (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;
19 uint64_t accum0 = 0, accum1 = 0, accum2 = 0;
20 uint32_t mask = (1<<28) - 1;
22 uint32_t aa[8], bb[8];
26 aa[i] = a[i] + a[i+8];
27 bb[i] = b[i] + b[i+8];
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]);
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]);
52 c[j] = ((uint32_t)(accum0)) & mask;
53 c[j+8] = ((uint32_t)(accum1)) & mask;
62 c[8] = ((uint32_t)(accum0)) & mask;
63 c[0] = ((uint32_t)(accum1)) & mask;
67 c[9] += ((uint32_t)(accum0));
68 c[1] += ((uint32_t)(accum1));
71 void gf_mulw_unsigned (gf_s *__restrict__ cs, const gf as, uint32_t b) {
74 const uint32_t *a = as->limb;
75 uint32_t *c = cs->limb;
77 uint64_t accum0 = 0, accum8 = 0;
78 uint32_t mask = (1ull<<28)-1;
82 accum0 += widemul(b, a[i]);
83 accum8 += widemul(b, a[i+8]);
85 c[i] = accum0 & mask; accum0 >>= 28;
86 c[i+8] = accum8 & mask; accum8 >>= 28;
89 accum0 += accum8 + c[8];
98 void gf_sqr (gf_s *__restrict__ cs, const gf as) {
99 gf_mul(cs,as,as); /* Performs better with a dedicated square */