| 1 |
#define update_asm CRYPTO_SHARED_NAMESPACE(update_asm) |
| 2 |
#define _update_asm _CRYPTO_SHARED_NAMESPACE(update_asm) |
| 3 |
#define vec_reduce_asm CRYPTO_SHARED_NAMESPACE(vec_reduce_asm) |
| 4 |
#define _vec_reduce_asm _CRYPTO_SHARED_NAMESPACE(vec_reduce_asm) |
| 5 |
/* |
| 6 |
This file is for implementating the inversion-free Berlekamp-Massey algorithm |
| 7 |
see https://ieeexplore.ieee.org/document/87857 |
| 8 |
|
| 9 |
For the implementation strategy, see |
| 10 |
https://eprint.iacr.org/2017/793.pdf |
| 11 |
*/ |
| 12 |
// 20240805 djb: more use of cryptoint |
| 13 |
// 20240530 djb: CRYPTO_ALIGN instead of ALIGN |
| 14 |
// 20240508 djb: include vec128_gf.h |
| 15 |
// 20240503 djb: use crypto_*_mask functions |
| 16 |
// 20221231 djb: ALIGN(32) for some arrays; tnx thom wiggers |
| 17 |
// 20221230 djb: add linker lines |
| 18 |
|
| 19 |
// linker define bm |
| 20 |
// linker use update_asm vec_reduce_asm |
| 21 |
// linker use vec_mul_sp_asm |
| 22 |
// linker use vec128_mul_asm |
| 23 |
// linker use gf_inv |
| 24 |
// linker use gf_mul2 |
| 25 |
|
| 26 |
#include "bm.h" |
| 27 |
|
| 28 |
#include "vec128_gf.h" |
| 29 |
#include "util.h" |
| 30 |
#include "vec.h" |
| 31 |
#include "gf.h" |
| 32 |
#include "crypto_uint64.h" |
| 33 |
|
| 34 |
#include <stdint.h> |
| 35 |
#include "crypto_int64.h" |
| 36 |
|
| 37 |
extern void update_asm(void *, gf, int); |
| 38 |
extern gf vec_reduce_asm(uint64_t *); |
| 39 |
|
| 40 |
static inline void vec_cmov(uint64_t out[][2], uint64_t mask) |
| 41 |
{ |
| 42 |
int i; |
| 43 |
|
| 44 |
for (i = 0; i < GFBITS; i++) |
| 45 |
out[i][0] = (out[i][0] & ~mask) | (out[i][1] & mask); |
| 46 |
} |
| 47 |
|
| 48 |
static inline void interleave(vec128 *in, int idx0, int idx1, vec128 *mask, int b) |
| 49 |
{ |
| 50 |
int s = 1 << b; |
| 51 |
|
| 52 |
vec128 x, y; |
| 53 |
|
| 54 |
x = vec128_or(vec128_and(in[idx0], mask[0]), |
| 55 |
vec128_sll_2x(vec128_and(in[idx1], mask[0]), s)); |
| 56 |
|
| 57 |
y = vec128_or(vec128_srl_2x(vec128_and(in[idx0], mask[1]), s), |
| 58 |
vec128_and(in[idx1], mask[1])); |
| 59 |
|
| 60 |
in[idx0] = x; |
| 61 |
in[idx1] = y; |
| 62 |
} |
| 63 |
|
| 64 |
/* input: in, field elements in bitsliced form */ |
| 65 |
/* output: out, field elements in non-bitsliced form */ |
| 66 |
static inline void get_coefs(gf *out, vec128 *in) |
| 67 |
{ |
| 68 |
int i, k; |
| 69 |
|
| 70 |
vec128 mask[4][2]; |
| 71 |
vec128 buf[16]; |
| 72 |
|
| 73 |
for (i = 0; i < GFBITS; i++) buf[i] = in[i]; |
| 74 |
for (i = GFBITS; i < 16; i++) buf[i] = vec128_setzero(); |
| 75 |
|
| 76 |
mask[0][0] = vec128_set1_16b(0x5555); |
| 77 |
mask[0][1] = vec128_set1_16b(0xAAAA); |
| 78 |
mask[1][0] = vec128_set1_16b(0x3333); |
| 79 |
mask[1][1] = vec128_set1_16b(0xCCCC); |
| 80 |
mask[2][0] = vec128_set1_16b(0x0F0F); |
| 81 |
mask[2][1] = vec128_set1_16b(0xF0F0); |
| 82 |
mask[3][0] = vec128_set1_16b(0x00FF); |
| 83 |
mask[3][1] = vec128_set1_16b(0xFF00); |
| 84 |
|
| 85 |
interleave(buf, 0, 8, mask[3], 3); |
| 86 |
interleave(buf, 1, 9, mask[3], 3); |
| 87 |
interleave(buf, 2, 10, mask[3], 3); |
| 88 |
interleave(buf, 3, 11, mask[3], 3); |
| 89 |
interleave(buf, 4, 12, mask[3], 3); |
| 90 |
interleave(buf, 5, 13, mask[3], 3); |
| 91 |
interleave(buf, 6, 14, mask[3], 3); |
| 92 |
interleave(buf, 7, 15, mask[3], 3); |
| 93 |
|
| 94 |
interleave(buf, 0, 4, mask[2], 2); |
| 95 |
interleave(buf, 1, 5, mask[2], 2); |
| 96 |
interleave(buf, 2, 6, mask[2], 2); |
| 97 |
interleave(buf, 3, 7, mask[2], 2); |
| 98 |
interleave(buf, 8, 12, mask[2], 2); |
| 99 |
interleave(buf, 9, 13, mask[2], 2); |
| 100 |
interleave(buf, 10, 14, mask[2], 2); |
| 101 |
interleave(buf, 11, 15, mask[2], 2); |
| 102 |
|
| 103 |
interleave(buf, 0, 2, mask[1], 1); |
| 104 |
interleave(buf, 1, 3, mask[1], 1); |
| 105 |
interleave(buf, 4, 6, mask[1], 1); |
| 106 |
interleave(buf, 5, 7, mask[1], 1); |
| 107 |
interleave(buf, 8, 10, mask[1], 1); |
| 108 |
interleave(buf, 9, 11, mask[1], 1); |
| 109 |
interleave(buf, 12, 14, mask[1], 1); |
| 110 |
interleave(buf, 13, 15, mask[1], 1); |
| 111 |
|
| 112 |
interleave(buf, 0, 1, mask[0], 0); |
| 113 |
interleave(buf, 2, 3, mask[0], 0); |
| 114 |
interleave(buf, 4, 5, mask[0], 0); |
| 115 |
interleave(buf, 6, 7, mask[0], 0); |
| 116 |
interleave(buf, 8, 9, mask[0], 0); |
| 117 |
interleave(buf, 10, 11, mask[0], 0); |
| 118 |
interleave(buf, 12, 13, mask[0], 0); |
| 119 |
interleave(buf, 14, 15, mask[0], 0); |
| 120 |
|
| 121 |
for (i = 0; i < 16; i++) |
| 122 |
for (k = 0; k < 4; k++) { |
| 123 |
out[ (4*0 + k)*16 + i ] = (vec128_extract(buf[i], 0) >> (k*16)) & GFMASK; |
| 124 |
out[ (4*1 + k)*16 + i ] = (vec128_extract(buf[i], 1) >> (k*16)) & GFMASK; |
| 125 |
} |
| 126 |
} |
| 127 |
|
| 128 |
/* input: in, field elements in bitsliced form */ |
| 129 |
/* output: out, field elements in non-bitsliced form */ |
| 130 |
void bm(uint64_t out[ GFBITS ], vec128 in[ GFBITS ]) |
| 131 |
{ |
| 132 |
uint16_t i; |
| 133 |
uint16_t N, L; |
| 134 |
|
| 135 |
uint64_t prod[ GFBITS ]; |
| 136 |
uint64_t in_tmp[ GFBITS ]; |
| 137 |
|
| 138 |
CRYPTO_ALIGN(32) uint64_t db[ GFBITS ][ 2 ]; |
| 139 |
CRYPTO_ALIGN(32) uint64_t BC_tmp[ GFBITS ][ 2 ]; |
| 140 |
CRYPTO_ALIGN(32) uint64_t BC[ GFBITS ][ 2 ]; |
| 141 |
|
| 142 |
uint64_t mask, t; |
| 143 |
|
| 144 |
gf d, b, c0=1; |
| 145 |
|
| 146 |
gf coefs[SYS_T * 2]; |
| 147 |
|
| 148 |
// init |
| 149 |
|
| 150 |
BC[0][1] = 0; |
| 151 |
BC[0][0] = 1; BC[0][0] <<= 63; |
| 152 |
|
| 153 |
for (i = 1; i < GFBITS; i++) |
| 154 |
BC[i][0] = BC[i][1] = 0; |
| 155 |
|
| 156 |
b = 1; |
| 157 |
L = 0; |
| 158 |
|
| 159 |
// |
| 160 |
|
| 161 |
get_coefs(coefs, in); |
| 162 |
|
| 163 |
for (i = 0; i < GFBITS; i++) |
| 164 |
in_tmp[i] = 0; |
| 165 |
|
| 166 |
for (N = 0; N < SYS_T * 2; N++) |
| 167 |
{ |
| 168 |
// computing d |
| 169 |
|
| 170 |
vec_mul_sp(prod, in_tmp, &BC[0][0]); |
| 171 |
|
| 172 |
update_asm(in_tmp, coefs[N], 8); |
| 173 |
|
| 174 |
d = vec_reduce_asm(prod); |
| 175 |
|
| 176 |
t = gf_mul2(c0, coefs[N], b); |
| 177 |
|
| 178 |
d ^= t & 0xFFFFFFFF; |
| 179 |
|
| 180 |
// 3 cases |
| 181 |
|
| 182 |
mask = crypto_uint64_nonzero_mask(d) & crypto_uint64_leq_mask(L*2, N); |
| 183 |
|
| 184 |
for (i = 0; i < GFBITS; i++) |
| 185 |
{ |
| 186 |
db[i][0] = crypto_int64_bitmod_mask(d, i); |
| 187 |
db[i][1] = crypto_int64_bitmod_mask(b, i); |
| 188 |
} |
| 189 |
|
| 190 |
vec128_mul((vec128*) BC_tmp, (vec128*) db, (vec128*) BC); |
| 191 |
|
| 192 |
vec_cmov(BC, mask); |
| 193 |
|
| 194 |
update_asm(BC, mask & c0, 16); |
| 195 |
|
| 196 |
for (i = 0; i < GFBITS; i++) |
| 197 |
BC[i][1] = BC_tmp[i][0] ^ BC_tmp[i][1]; |
| 198 |
|
| 199 |
c0 = t >> 32; |
| 200 |
b = (d & mask) | (b & ~mask); |
| 201 |
L = ((N+1-L) & mask) | (L & ~mask); |
| 202 |
|
| 203 |
} |
| 204 |
|
| 205 |
c0 = gf_inv(c0); |
| 206 |
|
| 207 |
for (i = 0; i < GFBITS; i++) { out[i] = crypto_int64_bitmod_mask(c0, i); } |
| 208 |
|
| 209 |
vec_mul_sp(out, out, &BC[0][0]); |
| 210 |
} |
| 211 |
|