| 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 cryptoint usage |
| 9 |
// 20240508 djb: include vec{128,256}_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_4x |
| 13 |
// 20221230 djb: add linker lines |
| 14 |
|
| 15 |
// linker define fft_tr |
| 16 |
// linker use vec128_mul_asm |
| 17 |
// linker use vec256_mul_asm |
| 18 |
// linker use vec256_ama_asm |
| 19 |
// linker use transpose_64x256_sp_asm |
| 20 |
// linker use fft_scalars_4x fft_consts |
| 21 |
|
| 22 |
#include "vec128_gf.h" |
| 23 |
#include "vec256_gf.h" |
| 24 |
#include "fft_tr.h" |
| 25 |
#include "fft_scalars_4x.h" |
| 26 |
#include "fft_consts.h" |
| 27 |
|
| 28 |
#include "transpose.h" |
| 29 |
|
| 30 |
#include <stdint.h> |
| 31 |
#include "crypto_int64.h" |
| 32 |
|
| 33 |
static void radix_conversions_tr(vec256 *in) |
| 34 |
{ |
| 35 |
int i, j, k; |
| 36 |
vec256 t; |
| 37 |
uint64_t v[4]; |
| 38 |
|
| 39 |
const vec256 mask[6][2] = |
| 40 |
{ |
| 41 |
{vec256_set4x(0x2222222222222222, 0x2222222222222222, 0x2222222222222222, 0x2222222222222222), |
| 42 |
vec256_set4x(0x4444444444444444, 0x4444444444444444, 0x4444444444444444, 0x4444444444444444)}, |
| 43 |
{vec256_set4x(0x0C0C0C0C0C0C0C0C, 0x0C0C0C0C0C0C0C0C, 0x0C0C0C0C0C0C0C0C, 0x0C0C0C0C0C0C0C0C), |
| 44 |
vec256_set4x(0x3030303030303030, 0x3030303030303030, 0x3030303030303030, 0x3030303030303030)}, |
| 45 |
{vec256_set4x(0x00F000F000F000F0, 0x00F000F000F000F0, 0x00F000F000F000F0, 0x00F000F000F000F0), |
| 46 |
vec256_set4x(0x0F000F000F000F00, 0x0F000F000F000F00, 0x0F000F000F000F00, 0x0F000F000F000F00)}, |
| 47 |
{vec256_set4x(0x0000FF000000FF00, 0x0000FF000000FF00, 0x0000FF000000FF00, 0x0000FF000000FF00), |
| 48 |
vec256_set4x(0x00FF000000FF0000, 0x00FF000000FF0000, 0x00FF000000FF0000, 0x00FF000000FF0000)}, |
| 49 |
{vec256_set4x(0x00000000FFFF0000, 0x00000000FFFF0000, 0x00000000FFFF0000, 0x00000000FFFF0000), |
| 50 |
vec256_set4x(0x0000FFFF00000000, 0x0000FFFF00000000, 0x0000FFFF00000000, 0x0000FFFF00000000)}, |
| 51 |
{vec256_set4x(0xFFFFFFFF00000000, 0xFFFFFFFF00000000, 0xFFFFFFFF00000000, 0xFFFFFFFF00000000), |
| 52 |
vec256_set4x(0x00000000FFFFFFFF, 0x00000000FFFFFFFF, 0x00000000FFFFFFFF, 0x00000000FFFFFFFF)} |
| 53 |
}; |
| 54 |
|
| 55 |
// |
| 56 |
|
| 57 |
for (j = 6; j >= 0; j--) |
| 58 |
{ |
| 59 |
if (j < 6) |
| 60 |
{ |
| 61 |
vec256_mul(in, in, fft_scalars_4x[j]); // scaling |
| 62 |
} |
| 63 |
|
| 64 |
for (k = j; k <= 4; k++) |
| 65 |
for (i = 0; i < GFBITS; i++) |
| 66 |
{ |
| 67 |
t = vec256_and(in[i], mask[k][0]); |
| 68 |
t = vec256_sll_4x(t, 1 << k); |
| 69 |
in[i] = vec256_xor(in[i], t); |
| 70 |
|
| 71 |
t = vec256_and(in[i], mask[k][1]); |
| 72 |
t = vec256_sll_4x(t, 1 << k); |
| 73 |
in[i] = vec256_xor(in[i], t); |
| 74 |
} |
| 75 |
|
| 76 |
if (j <= 5) |
| 77 |
for (i = 0; i < GFBITS; i++) |
| 78 |
{ |
| 79 |
v[0] = vec256_extract(in[i], 0); |
| 80 |
v[1] = vec256_extract(in[i], 1); |
| 81 |
v[2] = vec256_extract(in[i], 2); |
| 82 |
v[3] = vec256_extract(in[i], 3); |
| 83 |
|
| 84 |
v[1] ^= v[0] >> 32; |
| 85 |
v[1] ^= v[1] << 32; |
| 86 |
v[3] ^= v[2] >> 32; |
| 87 |
v[3] ^= v[3] << 32; |
| 88 |
|
| 89 |
in[i] = vec256_set4x(v[0], v[1], v[2], v[3]); |
| 90 |
} |
| 91 |
|
| 92 |
for (i = 0; i < GFBITS; i++) |
| 93 |
{ |
| 94 |
v[0] = vec256_extract(in[i], 0); |
| 95 |
v[1] = vec256_extract(in[i], 1); |
| 96 |
v[2] = vec256_extract(in[i], 2); |
| 97 |
v[3] = vec256_extract(in[i], 3); |
| 98 |
|
| 99 |
v[2] ^= v[1]; |
| 100 |
v[3] ^= v[2]; |
| 101 |
|
| 102 |
in[i] = vec256_set4x(v[0], v[1], v[2], v[3]); |
| 103 |
} |
| 104 |
|
| 105 |
} |
| 106 |
} |
| 107 |
|
| 108 |
static void butterflies_tr(vec256 *out, vec256 in[][ GFBITS ]) |
| 109 |
{ |
| 110 |
int i, j, k, s, b; |
| 111 |
|
| 112 |
vec256 t0[ GFBITS ]; |
| 113 |
vec256 t1[ GFBITS ]; |
| 114 |
vec256 t; |
| 115 |
|
| 116 |
vec128 out128[ GFBITS ][ 2 ]; |
| 117 |
vec128 tmp[ GFBITS ]; |
| 118 |
|
| 119 |
union { |
| 120 |
vec128 v[6][ GFBITS+1 ]; |
| 121 |
vec256 V[6][ (GFBITS+1)/2 ]; |
| 122 |
} pre; |
| 123 |
|
| 124 |
union { |
| 125 |
vec128 v[64][ 2 ]; |
| 126 |
vec256 V[64]; |
| 127 |
} buf; |
| 128 |
|
| 129 |
uint64_t v[4]; |
| 130 |
uint64_t consts_ptr = 33; |
| 131 |
|
| 132 |
const unsigned char reversal[] = |
| 133 |
{ |
| 134 |
0, 32, 16, 48, 8, 40, 24, 56, |
| 135 |
4, 36, 20, 52, 12, 44, 28, 60, |
| 136 |
2, 34, 18, 50, 10, 42, 26, 58, |
| 137 |
6, 38, 22, 54, 14, 46, 30, 62, |
| 138 |
1, 33, 17, 49, 9, 41, 25, 57, |
| 139 |
5, 37, 21, 53, 13, 45, 29, 61, |
| 140 |
3, 35, 19, 51, 11, 43, 27, 59, |
| 141 |
7, 39, 23, 55, 15, 47, 31, 63 |
| 142 |
}; |
| 143 |
|
| 144 |
const uint16_t beta[6] = {5246, 5306, 6039, 6685, 4905, 6755}; |
| 145 |
|
| 146 |
// butterflies |
| 147 |
|
| 148 |
for (i = 4; i >= 0; i--) |
| 149 |
{ |
| 150 |
s = 1 << i; |
| 151 |
consts_ptr -= s; |
| 152 |
|
| 153 |
for (j = 0; j < 32; j += 2*s) |
| 154 |
for (k = j; k < j+s; k++) |
| 155 |
{ |
| 156 |
vec256_ama_asm(in[k], in[k+s], fft_consts[ consts_ptr + (k-j) ]); |
| 157 |
} |
| 158 |
|
| 159 |
} |
| 160 |
|
| 161 |
for (k = 0; k < 32; k+=2) |
| 162 |
{ |
| 163 |
for (b = 0; b < GFBITS; b++) t0[b] = vec256_unpack_low(in[k][b], in[k+1][b]); |
| 164 |
for (b = 0; b < GFBITS; b++) t1[b] = vec256_unpack_high(in[k][b], in[k+1][b]); |
| 165 |
|
| 166 |
vec256_ama_asm(t0, t1, fft_consts[1]); |
| 167 |
|
| 168 |
for (b = 0; b < GFBITS; b++) in[k][b] = vec256_unpack_low(t0[b], t1[b]); |
| 169 |
for (b = 0; b < GFBITS; b++) in[k+1][b] = vec256_unpack_high(t0[b], t1[b]); |
| 170 |
|
| 171 |
for (b = 0; b < GFBITS; b++) t0[b] = vec256_unpack_low_2x(in[k][b], in[k+1][b]); |
| 172 |
for (b = 0; b < GFBITS; b++) t1[b] = vec256_unpack_high_2x(in[k][b], in[k+1][b]); |
| 173 |
|
| 174 |
vec256_ama_asm(t0, t1, fft_consts[0]); |
| 175 |
|
| 176 |
for (b = 0; b < GFBITS; b++) in[k+0][b] = vec256_unpack_low_2x(t0[b], t1[b]); |
| 177 |
for (b = 0; b < GFBITS; b++) in[k+1][b] = vec256_unpack_high_2x(t0[b], t1[b]); |
| 178 |
} |
| 179 |
|
| 180 |
|
| 181 |
// boradcast |
| 182 |
|
| 183 |
for (i = 0; i < GFBITS; i+=2) |
| 184 |
{ |
| 185 |
// transpose |
| 186 |
|
| 187 |
for (k = 0; k < 32; k++) |
| 188 |
{ |
| 189 |
if (i != GFBITS-1) { |
| 190 |
buf.v[ reversal[2*k+0] ][1] = vec256_extract2x(in[ k ][i+1], 0); |
| 191 |
buf.v[ reversal[2*k+1] ][1] = vec256_extract2x(in[ k ][i+1], 1); |
| 192 |
} |
| 193 |
|
| 194 |
buf.v[ reversal[2*k+0] ][0] = vec256_extract2x(in[ k ][i+0], 0); |
| 195 |
buf.v[ reversal[2*k+1] ][0] = vec256_extract2x(in[ k ][i+0], 1); |
| 196 |
} |
| 197 |
|
| 198 |
transpose_64x256_sp(buf.V); |
| 199 |
|
| 200 |
// |
| 201 |
|
| 202 |
#define XOR vec256_xor |
| 203 |
|
| 204 |
pre.V[0][i/2] = buf.V[32]; buf.V[33] = XOR(buf.V[33], buf.V[32]); |
| 205 |
pre.V[1][i/2] = buf.V[33]; buf.V[35] = XOR(buf.V[35], buf.V[33]); |
| 206 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[35]); buf.V[34] = XOR(buf.V[34], buf.V[35]); |
| 207 |
pre.V[2][i/2] = buf.V[34]; buf.V[38] = XOR(buf.V[38], buf.V[34]); |
| 208 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[38]); buf.V[39] = XOR(buf.V[39], buf.V[38]); |
| 209 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[39]); buf.V[37] = XOR(buf.V[37], buf.V[39]); |
| 210 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[37]); buf.V[36] = XOR(buf.V[36], buf.V[37]); |
| 211 |
pre.V[3][i/2] = buf.V[36]; buf.V[44] = XOR(buf.V[44], buf.V[36]); |
| 212 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[44]); buf.V[45] = XOR(buf.V[45], buf.V[44]); |
| 213 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[45]); buf.V[47] = XOR(buf.V[47], buf.V[45]); |
| 214 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[47]); buf.V[46] = XOR(buf.V[46], buf.V[47]); |
| 215 |
pre.V[2][i/2] = XOR(pre.V[2][i/2], buf.V[46]); buf.V[42] = XOR(buf.V[42], buf.V[46]); |
| 216 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[42]); buf.V[43] = XOR(buf.V[43], buf.V[42]); |
| 217 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[43]); buf.V[41] = XOR(buf.V[41], buf.V[43]); |
| 218 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[41]); buf.V[40] = XOR(buf.V[40], buf.V[41]); |
| 219 |
pre.V[4][i/2] = buf.V[40]; buf.V[56] = XOR(buf.V[56], buf.V[40]); |
| 220 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[56]); buf.V[57] = XOR(buf.V[57], buf.V[56]); |
| 221 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[57]); buf.V[59] = XOR(buf.V[59], buf.V[57]); |
| 222 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[59]); buf.V[58] = XOR(buf.V[58], buf.V[59]); |
| 223 |
pre.V[2][i/2] = XOR(pre.V[2][i/2], buf.V[58]); buf.V[62] = XOR(buf.V[62], buf.V[58]); |
| 224 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[62]); buf.V[63] = XOR(buf.V[63], buf.V[62]); |
| 225 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[63]); buf.V[61] = XOR(buf.V[61], buf.V[63]); |
| 226 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[61]); buf.V[60] = XOR(buf.V[60], buf.V[61]); |
| 227 |
pre.V[3][i/2] = XOR(pre.V[3][i/2], buf.V[60]); buf.V[52] = XOR(buf.V[52], buf.V[60]); |
| 228 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[52]); buf.V[53] = XOR(buf.V[53], buf.V[52]); |
| 229 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[53]); buf.V[55] = XOR(buf.V[55], buf.V[53]); |
| 230 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[55]); buf.V[54] = XOR(buf.V[54], buf.V[55]); |
| 231 |
pre.V[2][i/2] = XOR(pre.V[2][i/2], buf.V[54]); buf.V[50] = XOR(buf.V[50], buf.V[54]); |
| 232 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[50]); buf.V[51] = XOR(buf.V[51], buf.V[50]); |
| 233 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[51]); buf.V[49] = XOR(buf.V[49], buf.V[51]); |
| 234 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[49]); buf.V[48] = XOR(buf.V[48], buf.V[49]); |
| 235 |
pre.V[5][i/2] = buf.V[48]; buf.V[16] = XOR(buf.V[16], buf.V[48]); |
| 236 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[16]); buf.V[17] = XOR(buf.V[17], buf.V[16]); |
| 237 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[17]); buf.V[19] = XOR(buf.V[19], buf.V[17]); |
| 238 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[19]); buf.V[18] = XOR(buf.V[18], buf.V[19]); |
| 239 |
pre.V[2][i/2] = XOR(pre.V[2][i/2], buf.V[18]); buf.V[22] = XOR(buf.V[22], buf.V[18]); |
| 240 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[22]); buf.V[23] = XOR(buf.V[23], buf.V[22]); |
| 241 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[23]); buf.V[21] = XOR(buf.V[21], buf.V[23]); |
| 242 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[21]); buf.V[20] = XOR(buf.V[20], buf.V[21]); |
| 243 |
pre.V[3][i/2] = XOR(pre.V[3][i/2], buf.V[20]); buf.V[28] = XOR(buf.V[28], buf.V[20]); |
| 244 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[28]); buf.V[29] = XOR(buf.V[29], buf.V[28]); |
| 245 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[29]); buf.V[31] = XOR(buf.V[31], buf.V[29]); |
| 246 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[31]); buf.V[30] = XOR(buf.V[30], buf.V[31]); |
| 247 |
pre.V[2][i/2] = XOR(pre.V[2][i/2], buf.V[30]); buf.V[26] = XOR(buf.V[26], buf.V[30]); |
| 248 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[26]); buf.V[27] = XOR(buf.V[27], buf.V[26]); |
| 249 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[27]); buf.V[25] = XOR(buf.V[25], buf.V[27]); |
| 250 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[25]); buf.V[24] = XOR(buf.V[24], buf.V[25]); |
| 251 |
pre.V[4][i/2] = XOR(pre.V[4][i/2], buf.V[24]); buf.V[8] = XOR(buf.V[8], buf.V[24]); |
| 252 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[8]); buf.V[9] = XOR(buf.V[9], buf.V[8]); |
| 253 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[9]); buf.V[11] = XOR(buf.V[11], buf.V[9]); |
| 254 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[11]); buf.V[10] = XOR(buf.V[10], buf.V[11]); |
| 255 |
pre.V[2][i/2] = XOR(pre.V[2][i/2], buf.V[10]); buf.V[14] = XOR(buf.V[14], buf.V[10]); |
| 256 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[14]); buf.V[15] = XOR(buf.V[15], buf.V[14]); |
| 257 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[15]); buf.V[13] = XOR(buf.V[13], buf.V[15]); |
| 258 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[13]); buf.V[12] = XOR(buf.V[12], buf.V[13]); |
| 259 |
pre.V[3][i/2] = XOR(pre.V[3][i/2], buf.V[12]); buf.V[4] = XOR(buf.V[4], buf.V[12]); |
| 260 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[4]); buf.V[5] = XOR(buf.V[5], buf.V[4]); |
| 261 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[5]); buf.V[7] = XOR(buf.V[7], buf.V[5]); |
| 262 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[7]); buf.V[6] = XOR(buf.V[6], buf.V[7]); |
| 263 |
pre.V[2][i/2] = XOR(pre.V[2][i/2], buf.V[6]); buf.V[2] = XOR(buf.V[2], buf.V[6]); |
| 264 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[2]); buf.V[3] = XOR(buf.V[3], buf.V[2]); |
| 265 |
pre.V[1][i/2] = XOR(pre.V[1][i/2], buf.V[3]); buf.V[1] = XOR(buf.V[1], buf.V[3]); |
| 266 |
|
| 267 |
pre.V[0][i/2] = XOR(pre.V[0][i/2], buf.V[1]); t = XOR(buf.V[0], buf.V[1]); |
| 268 |
|
| 269 |
if (i != GFBITS-1) |
| 270 |
out128[i+1][0] = vec256_extract2x(t, 1); |
| 271 |
out128[i+0][0] = vec256_extract2x(t, 0); |
| 272 |
#undef XOR |
| 273 |
|
| 274 |
} |
| 275 |
|
| 276 |
// |
| 277 |
|
| 278 |
for (j = 0; j < GFBITS; j++) tmp[j] = vec128_setbits(crypto_int64_bitmod_01(beta[0], j)); |
| 279 |
|
| 280 |
vec128_mul(tmp, pre.v[0], tmp); |
| 281 |
|
| 282 |
for (b = 0; b < GFBITS; b++) out128[b][1] = tmp[b]; |
| 283 |
|
| 284 |
for (i = 1; i < 6; i++) |
| 285 |
{ |
| 286 |
for (j = 0; j < GFBITS; j++) tmp[j] = vec128_setbits(crypto_int64_bitmod_01(beta[i], j)); |
| 287 |
|
| 288 |
vec128_mul(tmp, pre.v[i], tmp); |
| 289 |
|
| 290 |
for (b = 0; b < GFBITS; b++) out128[b][1] = vec128_xor(out128[b][1], tmp[b]); |
| 291 |
} |
| 292 |
|
| 293 |
for (b = 0; b < GFBITS; b++) |
| 294 |
{ |
| 295 |
v[0] = vec128_extract(out128[b][0], 0); |
| 296 |
v[1] = vec128_extract(out128[b][0], 1); |
| 297 |
v[2] = vec128_extract(out128[b][1], 0); |
| 298 |
v[3] = vec128_extract(out128[b][1], 1); |
| 299 |
|
| 300 |
out[b] = vec256_set4x(v[0], v[1], v[2], v[3]); |
| 301 |
} |
| 302 |
} |
| 303 |
|
| 304 |
/* justifying the length of the output */ |
| 305 |
static void postprocess(vec256 *out) |
| 306 |
{ |
| 307 |
int i; |
| 308 |
uint64_t v[4]; |
| 309 |
|
| 310 |
for (i = 0; i < 13; i++) |
| 311 |
{ |
| 312 |
v[0] = vec256_extract(out[i], 0); |
| 313 |
v[1] = vec256_extract(out[i], 1); |
| 314 |
v[2] = vec256_extract(out[i], 2); |
| 315 |
v[3] = vec256_extract(out[i], 3); |
| 316 |
|
| 317 |
v[3] = 0; |
| 318 |
|
| 319 |
out[i] = vec256_set4x(v[0], v[1], v[2], v[3]); |
| 320 |
} |
| 321 |
} |
| 322 |
|
| 323 |
void fft_tr(vec256 *out, vec256 in[][ GFBITS ]) |
| 324 |
{ |
| 325 |
butterflies_tr(out, in); |
| 326 |
radix_conversions_tr(out); |
| 327 |
|
| 328 |
postprocess(out); |
| 329 |
} |
| 330 |
|