| 1 |
/* |
| 2 |
This file is for transpose of the Gao-Mateer FFT |
| 3 |
Functions with names ending with _tr are (roughly) the transpose of the corresponding functions in fft.c |
| 4 |
|
| 5 |
For the implementation strategy, see |
| 6 |
https://eprint.iacr.org/2017/793.pdf |
| 7 |
*/ |
| 8 |
// 20240805 djb: more use of cryptoint |
| 9 |
// 20240508 djb: include vec128_gf.h, vec256_gf.h |
| 10 |
// 20221230 djb: split these arrays into separate .c files |
| 11 |
// 20221230 djb: rename consts array as fft_consts |
| 12 |
// 20221230 djb: rename s array as fft_scalars_2x |
| 13 |
// 20221230 djb: add linker lines |
| 14 |
|
| 15 |
// linker define fft_tr |
| 16 |
// linker use vec_mul_asm |
| 17 |
// linker use vec128_mul_asm |
| 18 |
// linker use vec256_mul_asm |
| 19 |
// linker use transpose_64x256_sp_asm |
| 20 |
// linker use fft_scalars_2x fft_consts |
| 21 |
|
| 22 |
#include "vec128_gf.h" |
| 23 |
#include "vec256_gf.h" |
| 24 |
#include "fft_tr.h" |
| 25 |
#include "fft_scalars_2x.h" |
| 26 |
#include "fft_consts.h" |
| 27 |
|
| 28 |
#include "transpose.h" |
| 29 |
#include "vec.h" |
| 30 |
|
| 31 |
#include <stdint.h> |
| 32 |
#include "crypto_int64.h" |
| 33 |
|
| 34 |
static void radix_conversions_tr(vec128 in[ GFBITS ]) |
| 35 |
{ |
| 36 |
int i, j, k; |
| 37 |
|
| 38 |
const vec128 mask[10] = |
| 39 |
{ |
| 40 |
vec128_set2x(0x2222222222222222, 0x2222222222222222), |
| 41 |
vec128_set2x(0x4444444444444444, 0x4444444444444444), |
| 42 |
vec128_set2x(0x0C0C0C0C0C0C0C0C, 0x0C0C0C0C0C0C0C0C), |
| 43 |
vec128_set2x(0x3030303030303030, 0x3030303030303030), |
| 44 |
vec128_set2x(0x00F000F000F000F0, 0x00F000F000F000F0), |
| 45 |
vec128_set2x(0x0F000F000F000F00, 0x0F000F000F000F00), |
| 46 |
vec128_set2x(0x0000FF000000FF00, 0x0000FF000000FF00), |
| 47 |
vec128_set2x(0x00FF000000FF0000, 0x00FF000000FF0000), |
| 48 |
vec128_set2x(0x00000000FFFF0000, 0x00000000FFFF0000), |
| 49 |
vec128_set2x(0x0000FFFF00000000, 0x0000FFFF00000000) |
| 50 |
}; |
| 51 |
|
| 52 |
uint64_t v0, v1; |
| 53 |
|
| 54 |
// |
| 55 |
|
| 56 |
for (j = 5; j >= 0; j--) |
| 57 |
{ |
| 58 |
|
| 59 |
if (j < 5) vec128_mul(in, in, fft_scalars_2x[j]); |
| 60 |
|
| 61 |
for (i = 0; i < GFBITS; i++) |
| 62 |
for (k = j; k <= 4; k++) |
| 63 |
{ |
| 64 |
in[i] ^= vec128_sll_2x(in[i] & mask[2*k+0], 1 << k); |
| 65 |
in[i] ^= vec128_sll_2x(in[i] & mask[2*k+1], 1 << k); |
| 66 |
} |
| 67 |
|
| 68 |
for (i = 0; i < GFBITS; i++) |
| 69 |
{ |
| 70 |
v0 = vec128_extract(in[i], 0); |
| 71 |
v1 = vec128_extract(in[i], 1); |
| 72 |
|
| 73 |
v1 ^= v0 >> 32; |
| 74 |
v1 ^= v1 << 32; |
| 75 |
|
| 76 |
in[i] = vec128_set2x(v0, v1); |
| 77 |
} |
| 78 |
} |
| 79 |
} |
| 80 |
|
| 81 |
static void butterflies_tr(vec128 out[ GFBITS ], vec256 in[][ GFBITS ]) |
| 82 |
{ |
| 83 |
int i, j, k, s, b; |
| 84 |
|
| 85 |
uint64_t tmp[ GFBITS ]; |
| 86 |
uint64_t pre[6][ GFBITS ]; |
| 87 |
uint64_t out64[2][64]; |
| 88 |
|
| 89 |
vec256 p2[ 6 ]; |
| 90 |
vec256 buf[64]; |
| 91 |
vec256 x[ GFBITS ], y[ GFBITS ]; |
| 92 |
vec256 tmp256[ GFBITS ]; |
| 93 |
|
| 94 |
uint64_t consts_ptr = 17; |
| 95 |
|
| 96 |
const unsigned char reversal[64] = |
| 97 |
{ |
| 98 |
0, 32, 16, 48, 8, 40, 24, 56, |
| 99 |
4, 36, 20, 52, 12, 44, 28, 60, |
| 100 |
2, 34, 18, 50, 10, 42, 26, 58, |
| 101 |
6, 38, 22, 54, 14, 46, 30, 62, |
| 102 |
1, 33, 17, 49, 9, 41, 25, 57, |
| 103 |
5, 37, 21, 53, 13, 45, 29, 61, |
| 104 |
3, 35, 19, 51, 11, 43, 27, 59, |
| 105 |
7, 39, 23, 55, 15, 47, 31, 63 |
| 106 |
}; |
| 107 |
|
| 108 |
const uint16_t beta[6] = {8, 1300, 3408, 1354, 2341, 1154}; |
| 109 |
|
| 110 |
// butterflies |
| 111 |
|
| 112 |
for (i = 3; i >= 0; i--) |
| 113 |
{ |
| 114 |
s = 1 << i; |
| 115 |
consts_ptr -= s; |
| 116 |
|
| 117 |
for (j = 0; j < 16; j += 2*s) |
| 118 |
{ |
| 119 |
for (k = j; k < j+s; k++) |
| 120 |
{ |
| 121 |
for (b = 0; b < GFBITS; b++) in[k][b] ^= in[k+s][b]; |
| 122 |
vec256_mul(tmp256, in[k], fft_consts[ consts_ptr + (k-j) ]); |
| 123 |
for (b = 0; b < GFBITS; b++) in[k+s][b] ^= tmp256[b]; |
| 124 |
} |
| 125 |
} |
| 126 |
} |
| 127 |
|
| 128 |
for (i = 0; i < 16; i+=2) |
| 129 |
{ |
| 130 |
for (b = 0; b < GFBITS; b++) x[b] = vec256_unpack_low(in[i+0][b], in[i+1][b]); |
| 131 |
for (b = 0; b < GFBITS; b++) y[b] = vec256_unpack_high(in[i+0][b], in[i+1][b]); |
| 132 |
|
| 133 |
for (b = 0; b < GFBITS; b++) x[b] ^= y[b]; |
| 134 |
vec256_mul(tmp256, x, fft_consts[ 1 ]); |
| 135 |
for (b = 0; b < GFBITS; b++) y[b] ^= tmp256[b]; |
| 136 |
|
| 137 |
for (b = 0; b < GFBITS; b++) in[i+0][b] = vec256_unpack_low_2x(x[b], y[b]); |
| 138 |
for (b = 0; b < GFBITS; b++) in[i+1][b] = vec256_unpack_high_2x(x[b], y[b]); |
| 139 |
} |
| 140 |
|
| 141 |
for (i = 0; i < 16; i+=2) |
| 142 |
{ |
| 143 |
for (b = 0; b < GFBITS; b++) in[i+0][b] ^= in[i+1][b]; |
| 144 |
vec256_mul(tmp256, in[i+0], fft_consts[ 0 ]); |
| 145 |
for (b = 0; b < GFBITS; b++) in[i+1][b] ^= tmp256[b]; |
| 146 |
} |
| 147 |
|
| 148 |
// transpose |
| 149 |
|
| 150 |
for (i = 0; i < GFBITS; i+=4) |
| 151 |
{ |
| 152 |
for (j = 0; j < 64; j+=8) |
| 153 |
{ |
| 154 |
buf[ reversal[j+0] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 0), |
| 155 |
vec256_extract(in[j/4+0][i+1], 0), |
| 156 |
vec256_extract(in[j/4+0][i+2], 0), |
| 157 |
vec256_extract(in[j/4+0][i+3], 0)); |
| 158 |
buf[ reversal[j+1] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 0), |
| 159 |
vec256_extract(in[j/4+1][i+1], 0), |
| 160 |
vec256_extract(in[j/4+1][i+2], 0), |
| 161 |
vec256_extract(in[j/4+1][i+3], 0)); |
| 162 |
buf[ reversal[j+2] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 1), |
| 163 |
vec256_extract(in[j/4+0][i+1], 1), |
| 164 |
vec256_extract(in[j/4+0][i+2], 1), |
| 165 |
vec256_extract(in[j/4+0][i+3], 1)); |
| 166 |
buf[ reversal[j+3] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 1), |
| 167 |
vec256_extract(in[j/4+1][i+1], 1), |
| 168 |
vec256_extract(in[j/4+1][i+2], 1), |
| 169 |
vec256_extract(in[j/4+1][i+3], 1)); |
| 170 |
buf[ reversal[j+4] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 2), |
| 171 |
vec256_extract(in[j/4+0][i+1], 2), |
| 172 |
vec256_extract(in[j/4+0][i+2], 2), |
| 173 |
vec256_extract(in[j/4+0][i+3], 2)); |
| 174 |
buf[ reversal[j+5] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 2), |
| 175 |
vec256_extract(in[j/4+1][i+1], 2), |
| 176 |
vec256_extract(in[j/4+1][i+2], 2), |
| 177 |
vec256_extract(in[j/4+1][i+3], 2)); |
| 178 |
buf[ reversal[j+6] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 3), |
| 179 |
vec256_extract(in[j/4+0][i+1], 3), |
| 180 |
vec256_extract(in[j/4+0][i+2], 3), |
| 181 |
vec256_extract(in[j/4+0][i+3], 3)); |
| 182 |
buf[ reversal[j+7] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 3), |
| 183 |
vec256_extract(in[j/4+1][i+1], 3), |
| 184 |
vec256_extract(in[j/4+1][i+2], 3), |
| 185 |
vec256_extract(in[j/4+1][i+3], 3)); |
| 186 |
} |
| 187 |
|
| 188 |
transpose_64x256_sp(buf); |
| 189 |
|
| 190 |
p2[0] = buf[32]; buf[33] = vec256_xor(buf[33], buf[32]); |
| 191 |
p2[1] = buf[33]; buf[35] = vec256_xor(buf[35], buf[33]); |
| 192 |
p2[0] = vec256_xor(p2[0], buf[35]); buf[34] = vec256_xor(buf[34], buf[35]); |
| 193 |
p2[2] = buf[34]; buf[38] = vec256_xor(buf[38], buf[34]); |
| 194 |
p2[0] = vec256_xor(p2[0], buf[38]); buf[39] = vec256_xor(buf[39], buf[38]); |
| 195 |
p2[1] = vec256_xor(p2[1], buf[39]); buf[37] = vec256_xor(buf[37], buf[39]); |
| 196 |
p2[0] = vec256_xor(p2[0], buf[37]); buf[36] = vec256_xor(buf[36], buf[37]); |
| 197 |
p2[3] = buf[36]; buf[44] = vec256_xor(buf[44], buf[36]); |
| 198 |
p2[0] = vec256_xor(p2[0], buf[44]); buf[45] = vec256_xor(buf[45], buf[44]); |
| 199 |
p2[1] = vec256_xor(p2[1], buf[45]); buf[47] = vec256_xor(buf[47], buf[45]); |
| 200 |
p2[0] = vec256_xor(p2[0], buf[47]); buf[46] = vec256_xor(buf[46], buf[47]); |
| 201 |
p2[2] = vec256_xor(p2[2], buf[46]); buf[42] = vec256_xor(buf[42], buf[46]); |
| 202 |
p2[0] = vec256_xor(p2[0], buf[42]); buf[43] = vec256_xor(buf[43], buf[42]); |
| 203 |
p2[1] = vec256_xor(p2[1], buf[43]); buf[41] = vec256_xor(buf[41], buf[43]); |
| 204 |
p2[0] = vec256_xor(p2[0], buf[41]); buf[40] = vec256_xor(buf[40], buf[41]); |
| 205 |
p2[4] = buf[40]; buf[56] = vec256_xor(buf[56], buf[40]); |
| 206 |
p2[0] = vec256_xor(p2[0], buf[56]); buf[57] = vec256_xor(buf[57], buf[56]); |
| 207 |
p2[1] = vec256_xor(p2[1], buf[57]); buf[59] = vec256_xor(buf[59], buf[57]); |
| 208 |
p2[0] = vec256_xor(p2[0], buf[59]); buf[58] = vec256_xor(buf[58], buf[59]); |
| 209 |
p2[2] = vec256_xor(p2[2], buf[58]); buf[62] = vec256_xor(buf[62], buf[58]); |
| 210 |
p2[0] = vec256_xor(p2[0], buf[62]); buf[63] = vec256_xor(buf[63], buf[62]); |
| 211 |
p2[1] = vec256_xor(p2[1], buf[63]); buf[61] = vec256_xor(buf[61], buf[63]); |
| 212 |
p2[0] = vec256_xor(p2[0], buf[61]); buf[60] = vec256_xor(buf[60], buf[61]); |
| 213 |
p2[3] = vec256_xor(p2[3], buf[60]); buf[52] = vec256_xor(buf[52], buf[60]); |
| 214 |
p2[0] = vec256_xor(p2[0], buf[52]); buf[53] = vec256_xor(buf[53], buf[52]); |
| 215 |
p2[1] = vec256_xor(p2[1], buf[53]); buf[55] = vec256_xor(buf[55], buf[53]); |
| 216 |
p2[0] = vec256_xor(p2[0], buf[55]); buf[54] = vec256_xor(buf[54], buf[55]); |
| 217 |
p2[2] = vec256_xor(p2[2], buf[54]); buf[50] = vec256_xor(buf[50], buf[54]); |
| 218 |
p2[0] = vec256_xor(p2[0], buf[50]); buf[51] = vec256_xor(buf[51], buf[50]); |
| 219 |
p2[1] = vec256_xor(p2[1], buf[51]); buf[49] = vec256_xor(buf[49], buf[51]); |
| 220 |
p2[0] = vec256_xor(p2[0], buf[49]); buf[48] = vec256_xor(buf[48], buf[49]); |
| 221 |
p2[5] = buf[48]; buf[16] = vec256_xor(buf[16], buf[48]); |
| 222 |
p2[0] = vec256_xor(p2[0], buf[16]); buf[17] = vec256_xor(buf[17], buf[16]); |
| 223 |
p2[1] = vec256_xor(p2[1], buf[17]); buf[19] = vec256_xor(buf[19], buf[17]); |
| 224 |
p2[0] = vec256_xor(p2[0], buf[19]); buf[18] = vec256_xor(buf[18], buf[19]); |
| 225 |
p2[2] = vec256_xor(p2[2], buf[18]); buf[22] = vec256_xor(buf[22], buf[18]); |
| 226 |
p2[0] = vec256_xor(p2[0], buf[22]); buf[23] = vec256_xor(buf[23], buf[22]); |
| 227 |
p2[1] = vec256_xor(p2[1], buf[23]); buf[21] = vec256_xor(buf[21], buf[23]); |
| 228 |
p2[0] = vec256_xor(p2[0], buf[21]); buf[20] = vec256_xor(buf[20], buf[21]); |
| 229 |
p2[3] = vec256_xor(p2[3], buf[20]); buf[28] = vec256_xor(buf[28], buf[20]); |
| 230 |
p2[0] = vec256_xor(p2[0], buf[28]); buf[29] = vec256_xor(buf[29], buf[28]); |
| 231 |
p2[1] = vec256_xor(p2[1], buf[29]); buf[31] = vec256_xor(buf[31], buf[29]); |
| 232 |
p2[0] = vec256_xor(p2[0], buf[31]); buf[30] = vec256_xor(buf[30], buf[31]); |
| 233 |
p2[2] = vec256_xor(p2[2], buf[30]); buf[26] = vec256_xor(buf[26], buf[30]); |
| 234 |
p2[0] = vec256_xor(p2[0], buf[26]); buf[27] = vec256_xor(buf[27], buf[26]); |
| 235 |
p2[1] = vec256_xor(p2[1], buf[27]); buf[25] = vec256_xor(buf[25], buf[27]); |
| 236 |
p2[0] = vec256_xor(p2[0], buf[25]); buf[24] = vec256_xor(buf[24], buf[25]); |
| 237 |
p2[4] = vec256_xor(p2[4], buf[24]); buf[8] = vec256_xor(buf[8], buf[24]); |
| 238 |
p2[0] = vec256_xor(p2[0], buf[8]); buf[9] = vec256_xor(buf[9], buf[8]); |
| 239 |
p2[1] = vec256_xor(p2[1], buf[9]); buf[11] = vec256_xor(buf[11], buf[9]); |
| 240 |
p2[0] = vec256_xor(p2[0], buf[11]); buf[10] = vec256_xor(buf[10], buf[11]); |
| 241 |
p2[2] = vec256_xor(p2[2], buf[10]); buf[14] = vec256_xor(buf[14], buf[10]); |
| 242 |
p2[0] = vec256_xor(p2[0], buf[14]); buf[15] = vec256_xor(buf[15], buf[14]); |
| 243 |
p2[1] = vec256_xor(p2[1], buf[15]); buf[13] = vec256_xor(buf[13], buf[15]); |
| 244 |
p2[0] = vec256_xor(p2[0], buf[13]); buf[12] = vec256_xor(buf[12], buf[13]); |
| 245 |
p2[3] = vec256_xor(p2[3], buf[12]); buf[4] = vec256_xor(buf[4], buf[12]); |
| 246 |
p2[0] = vec256_xor(p2[0], buf[4]); buf[5] = vec256_xor(buf[5], buf[4]); |
| 247 |
p2[1] = vec256_xor(p2[1], buf[5]); buf[7] = vec256_xor(buf[7], buf[5]); |
| 248 |
p2[0] = vec256_xor(p2[0], buf[7]); buf[6] = vec256_xor(buf[6], buf[7]); |
| 249 |
p2[2] = vec256_xor(p2[2], buf[6]); buf[2] = vec256_xor(buf[2], buf[6]); |
| 250 |
p2[0] = vec256_xor(p2[0], buf[2]); buf[3] = vec256_xor(buf[3], buf[2]); |
| 251 |
p2[1] = vec256_xor(p2[1], buf[3]); buf[1] = vec256_xor(buf[1], buf[3]); |
| 252 |
|
| 253 |
p2[0] = vec256_xor(p2[0], buf[1]); buf[0] = vec256_xor(buf[0], buf[1]); |
| 254 |
|
| 255 |
for (j = 0; j < 6; j++) |
| 256 |
{ |
| 257 |
pre[j][i+0] = vec256_extract(p2[j], 0); |
| 258 |
pre[j][i+1] = vec256_extract(p2[j], 1); |
| 259 |
pre[j][i+2] = vec256_extract(p2[j], 2); |
| 260 |
pre[j][i+3] = vec256_extract(p2[j], 3); |
| 261 |
} |
| 262 |
|
| 263 |
out64[0][i+0] = vec256_extract(buf[0], 0); |
| 264 |
out64[0][i+1] = vec256_extract(buf[0], 1); |
| 265 |
out64[0][i+2] = vec256_extract(buf[0], 2); |
| 266 |
out64[0][i+3] = vec256_extract(buf[0], 3); |
| 267 |
} |
| 268 |
|
| 269 |
// |
| 270 |
|
| 271 |
for (j = 0; j < GFBITS; j++) { tmp[j] = crypto_int64_bitmod_mask(beta[0], j); } |
| 272 |
|
| 273 |
vec_mul(out64[1], pre[0], tmp); |
| 274 |
|
| 275 |
for (i = 1; i < 6; i++) |
| 276 |
{ |
| 277 |
for (j = 0; j < GFBITS; j++) { tmp[j] = crypto_int64_bitmod_mask(beta[i], j); } |
| 278 |
|
| 279 |
vec_mul(tmp, pre[i], tmp); |
| 280 |
vec_add(out64[1], out64[1], tmp); |
| 281 |
} |
| 282 |
|
| 283 |
for (i = 0; i < GFBITS; i++) |
| 284 |
out[i] = vec128_set2x(out64[0][i], out64[1][i]); |
| 285 |
} |
| 286 |
|
| 287 |
void fft_tr(vec128 out[GFBITS], vec256 in[][ GFBITS ]) |
| 288 |
{ |
| 289 |
butterflies_tr(out, in); |
| 290 |
radix_conversions_tr(out); |
| 291 |
} |
| 292 |
|