| 1 |
/* |
| 2 |
This file is for implementating the inversion-free Berlekamp-Massey algorithm |
| 3 |
see https://ieeexplore.ieee.org/document/87857 |
| 4 |
|
| 5 |
For the implementation strategy, see |
| 6 |
https://eprint.iacr.org/2017/793.pdf |
| 7 |
*/ |
| 8 |
// 20240805 djb: more use of cryptoint |
| 9 |
// 20240503 djb: use crypto_*_mask functions |
| 10 |
// 20221230 djb: add linker lines |
| 11 |
|
| 12 |
// linker define bm |
| 13 |
// linker use vec_mul |
| 14 |
// linker use gf_inv |
| 15 |
// linker use gf_mul2 |
| 16 |
|
| 17 |
#include "bm.h" |
| 18 |
|
| 19 |
#include "gf.h" |
| 20 |
#include "crypto_uint64.h" |
| 21 |
#include "crypto_int64.h" |
| 22 |
|
| 23 |
static inline void vec_cmov(vec * out, vec * in, uint16_t mask) |
| 24 |
{ |
| 25 |
int i; |
| 26 |
|
| 27 |
vec m0, m1; |
| 28 |
|
| 29 |
m0 = vec_set1_16b(mask); |
| 30 |
m1 = ~m0; |
| 31 |
|
| 32 |
for (i = 0; i < GFBITS; i++) |
| 33 |
{ |
| 34 |
out[i] = (in[i] & m0) | (out[i] & m1); |
| 35 |
out[i] = (in[i] & m0) | (out[i] & m1); |
| 36 |
} |
| 37 |
} |
| 38 |
|
| 39 |
static inline void interleave(vec *in, int idx0, int idx1, vec *mask, int b) |
| 40 |
{ |
| 41 |
int s = 1 << b; |
| 42 |
|
| 43 |
vec x, y; |
| 44 |
|
| 45 |
x = (in[idx0] & mask[0]) | ((in[idx1] & mask[0]) << s); |
| 46 |
y = ((in[idx0] & mask[1]) >> s) | (in[idx1] & mask[1]); |
| 47 |
|
| 48 |
in[idx0] = x; |
| 49 |
in[idx1] = y; |
| 50 |
} |
| 51 |
|
| 52 |
/* input: in, field elements in bitsliced form */ |
| 53 |
/* output: out, field elements in non-bitsliced form */ |
| 54 |
static inline void get_coefs(gf *out, vec *in) |
| 55 |
{ |
| 56 |
int i, k; |
| 57 |
|
| 58 |
vec mask[4][2]; |
| 59 |
vec buf[16]; |
| 60 |
|
| 61 |
for (i = 0; i < 13; i++) buf[i] = in[i]; |
| 62 |
for (i = 13; i < 16; i++) buf[i] = 0; |
| 63 |
|
| 64 |
mask[0][0] = vec_set1_16b(0x5555); |
| 65 |
mask[0][1] = vec_set1_16b(0xAAAA); |
| 66 |
mask[1][0] = vec_set1_16b(0x3333); |
| 67 |
mask[1][1] = vec_set1_16b(0xCCCC); |
| 68 |
mask[2][0] = vec_set1_16b(0x0F0F); |
| 69 |
mask[2][1] = vec_set1_16b(0xF0F0); |
| 70 |
mask[3][0] = vec_set1_16b(0x00FF); |
| 71 |
mask[3][1] = vec_set1_16b(0xFF00); |
| 72 |
|
| 73 |
interleave(buf, 0, 8, mask[3], 3); |
| 74 |
interleave(buf, 1, 9, mask[3], 3); |
| 75 |
interleave(buf, 2, 10, mask[3], 3); |
| 76 |
interleave(buf, 3, 11, mask[3], 3); |
| 77 |
interleave(buf, 4, 12, mask[3], 3); |
| 78 |
interleave(buf, 5, 13, mask[3], 3); |
| 79 |
interleave(buf, 6, 14, mask[3], 3); |
| 80 |
interleave(buf, 7, 15, mask[3], 3); |
| 81 |
|
| 82 |
interleave(buf, 0, 4, mask[2], 2); |
| 83 |
interleave(buf, 1, 5, mask[2], 2); |
| 84 |
interleave(buf, 2, 6, mask[2], 2); |
| 85 |
interleave(buf, 3, 7, mask[2], 2); |
| 86 |
interleave(buf, 8, 12, mask[2], 2); |
| 87 |
interleave(buf, 9, 13, mask[2], 2); |
| 88 |
interleave(buf, 10, 14, mask[2], 2); |
| 89 |
interleave(buf, 11, 15, mask[2], 2); |
| 90 |
|
| 91 |
interleave(buf, 0, 2, mask[1], 1); |
| 92 |
interleave(buf, 1, 3, mask[1], 1); |
| 93 |
interleave(buf, 4, 6, mask[1], 1); |
| 94 |
interleave(buf, 5, 7, mask[1], 1); |
| 95 |
interleave(buf, 8, 10, mask[1], 1); |
| 96 |
interleave(buf, 9, 11, mask[1], 1); |
| 97 |
interleave(buf, 12, 14, mask[1], 1); |
| 98 |
interleave(buf, 13, 15, mask[1], 1); |
| 99 |
|
| 100 |
interleave(buf, 0, 1, mask[0], 0); |
| 101 |
interleave(buf, 2, 3, mask[0], 0); |
| 102 |
interleave(buf, 4, 5, mask[0], 0); |
| 103 |
interleave(buf, 6, 7, mask[0], 0); |
| 104 |
interleave(buf, 8, 9, mask[0], 0); |
| 105 |
interleave(buf, 10, 11, mask[0], 0); |
| 106 |
interleave(buf, 12, 13, mask[0], 0); |
| 107 |
interleave(buf, 14, 15, mask[0], 0); |
| 108 |
|
| 109 |
for (i = 0; i < 16; i++) |
| 110 |
for (k = 0; k < 4; k++) |
| 111 |
out[ k*16 + i ] = (buf[i] >> (k*16)) & GFMASK; |
| 112 |
} |
| 113 |
|
| 114 |
static void update(vec in[][GFBITS], const gf e) |
| 115 |
{ |
| 116 |
int i; |
| 117 |
vec tmp; |
| 118 |
|
| 119 |
for (i = 0; i < GFBITS; i++) |
| 120 |
{ |
| 121 |
tmp = crypto_int64_bitmod_01(e, i); |
| 122 |
|
| 123 |
in[0][i] = (in[0][i] >> 1) | (in[1][i] << 63); |
| 124 |
in[1][i] = (in[1][i] >> 1) | (tmp << 63); |
| 125 |
} |
| 126 |
} |
| 127 |
|
| 128 |
static inline gf vec_reduce(vec in[][GFBITS]) |
| 129 |
{ |
| 130 |
int i; |
| 131 |
vec tmp; |
| 132 |
gf ret = 0; |
| 133 |
|
| 134 |
for (i = GFBITS-1; i >= 0; i--) |
| 135 |
{ |
| 136 |
tmp = in[0][i] ^ in[1][i]; |
| 137 |
|
| 138 |
tmp ^= tmp >> 32; |
| 139 |
tmp ^= tmp >> 16; |
| 140 |
tmp ^= tmp >> 8; |
| 141 |
tmp ^= tmp >> 4; |
| 142 |
tmp ^= tmp >> 2; |
| 143 |
tmp ^= tmp >> 1; |
| 144 |
|
| 145 |
ret <<= 1; |
| 146 |
ret |= crypto_int64_bottombit_01(tmp); |
| 147 |
} |
| 148 |
|
| 149 |
return ret; |
| 150 |
} |
| 151 |
|
| 152 |
/* input: in, sequence of field elements */ |
| 153 |
/* output: out, minimal polynomial of in */ |
| 154 |
void bm(vec out[][ GFBITS ], vec in[][ GFBITS ]) |
| 155 |
{ |
| 156 |
int i; |
| 157 |
uint16_t N, L; |
| 158 |
uint16_t mask; |
| 159 |
uint64_t one = 1, t; |
| 160 |
|
| 161 |
vec prod[2][GFBITS]; |
| 162 |
vec interval[2][GFBITS]; |
| 163 |
vec dd[2][GFBITS], bb[2][GFBITS]; |
| 164 |
vec B[2][GFBITS], C[2][GFBITS]; |
| 165 |
vec B_tmp[2][GFBITS], C_tmp[2][GFBITS]; |
| 166 |
vec v[GFBITS]; |
| 167 |
|
| 168 |
gf d, b, c0 = 1; |
| 169 |
gf coefs[256]; |
| 170 |
|
| 171 |
// initialization |
| 172 |
|
| 173 |
get_coefs(&coefs[ 0], in[0]); |
| 174 |
get_coefs(&coefs[ 64], in[1]); |
| 175 |
get_coefs(&coefs[128], in[2]); |
| 176 |
get_coefs(&coefs[192], in[3]); |
| 177 |
|
| 178 |
C[0][0] = 0; |
| 179 |
C[1][0] = 0; |
| 180 |
B[0][0] = 0; |
| 181 |
B[1][0] = one << 63; |
| 182 |
|
| 183 |
for (i = 1; i < GFBITS; i++) |
| 184 |
C[0][i] = C[1][i] = B[0][i] = B[1][i] = 0; |
| 185 |
|
| 186 |
b = 1; |
| 187 |
L = 0; |
| 188 |
|
| 189 |
// |
| 190 |
|
| 191 |
for (i = 0; i < GFBITS; i++) |
| 192 |
interval[0][i] = interval[1][i] = 0; |
| 193 |
|
| 194 |
for (N = 0; N < 256; N++) |
| 195 |
{ |
| 196 |
vec_mul(prod[0], C[0], interval[0]); |
| 197 |
vec_mul(prod[1], C[1], interval[1]); |
| 198 |
update(interval, coefs[N]); |
| 199 |
d = vec_reduce(prod); |
| 200 |
|
| 201 |
t = gf_mul2(c0, coefs[N], b); |
| 202 |
d ^= t & 0xFFFFFFFF; |
| 203 |
|
| 204 |
mask = crypto_uint64_nonzero_mask(d) & crypto_uint64_leq_mask(L*2, N); |
| 205 |
|
| 206 |
for (i = 0; i < GFBITS; i++) |
| 207 |
{ |
| 208 |
dd[0][i] = dd[1][i] = vec_setbits(crypto_int64_bitmod_01(d, i)); |
| 209 |
bb[0][i] = bb[1][i] = vec_setbits(crypto_int64_bitmod_01(b, i)); |
| 210 |
} |
| 211 |
|
| 212 |
vec_mul(B_tmp[0], dd[0], B[0]); |
| 213 |
vec_mul(B_tmp[1], dd[1], B[1]); |
| 214 |
vec_mul(C_tmp[0], bb[0], C[0]); |
| 215 |
vec_mul(C_tmp[1], bb[1], C[1]); |
| 216 |
|
| 217 |
vec_cmov(B[0], C[0], mask); |
| 218 |
vec_cmov(B[1], C[1], mask); |
| 219 |
update(B, c0 & mask); |
| 220 |
|
| 221 |
for (i = 0; i < GFBITS; i++) |
| 222 |
{ |
| 223 |
C[0][i] = B_tmp[0][i] ^ C_tmp[0][i]; |
| 224 |
C[1][i] = B_tmp[1][i] ^ C_tmp[1][i]; |
| 225 |
} |
| 226 |
|
| 227 |
c0 = t >> 32; |
| 228 |
b = (d & mask) | (b & ~mask); |
| 229 |
L = ((N+1-L) & mask) | (L & ~mask); |
| 230 |
} |
| 231 |
|
| 232 |
c0 = gf_inv(c0); |
| 233 |
|
| 234 |
for (i = 0; i < GFBITS; i++) |
| 235 |
v[i] = vec_setbits(crypto_int64_bitmod_01(c0, i)); |
| 236 |
|
| 237 |
vec_mul(out[0], C[0], v); |
| 238 |
vec_mul(out[1], C[1], v); |
| 239 |
} |
| 240 |
|