-rw-r--r-- 11879 libmceliece-20240513/crypto_kem/348864/avx/fft_tr.c raw
/*
This file is for transpose of the Gao-Mateer FFT
Functions with names ending with _tr are (roughly) the transpose of the corresponding functions in fft.c
For the implementation strategy, see
https://eprint.iacr.org/2017/793.pdf
*/
// 20240508 djb: include vec128_gf.h, vec256_gf.h
// 20221230 djb: split these arrays into separate .c files
// 20221230 djb: rename consts array as fft_consts
// 20221230 djb: rename s array as fft_scalars_2x
// 20221230 djb: add linker lines
// linker define fft_tr
// linker use vec_mul_asm
// linker use vec128_mul_asm
// linker use vec256_mul_asm
// linker use transpose_64x256_sp_asm
// linker use fft_scalars_2x fft_consts
#include "vec128_gf.h"
#include "vec256_gf.h"
#include "fft_tr.h"
#include "fft_scalars_2x.h"
#include "fft_consts.h"
#include "transpose.h"
#include "vec.h"
#include <stdint.h>
static void radix_conversions_tr(vec128 in[ GFBITS ])
{
int i, j, k;
const vec128 mask[10] =
{
vec128_set2x(0x2222222222222222, 0x2222222222222222),
vec128_set2x(0x4444444444444444, 0x4444444444444444),
vec128_set2x(0x0C0C0C0C0C0C0C0C, 0x0C0C0C0C0C0C0C0C),
vec128_set2x(0x3030303030303030, 0x3030303030303030),
vec128_set2x(0x00F000F000F000F0, 0x00F000F000F000F0),
vec128_set2x(0x0F000F000F000F00, 0x0F000F000F000F00),
vec128_set2x(0x0000FF000000FF00, 0x0000FF000000FF00),
vec128_set2x(0x00FF000000FF0000, 0x00FF000000FF0000),
vec128_set2x(0x00000000FFFF0000, 0x00000000FFFF0000),
vec128_set2x(0x0000FFFF00000000, 0x0000FFFF00000000)
};
uint64_t v0, v1;
//
for (j = 5; j >= 0; j--)
{
if (j < 5) vec128_mul(in, in, fft_scalars_2x[j]);
for (i = 0; i < GFBITS; i++)
for (k = j; k <= 4; k++)
{
in[i] ^= vec128_sll_2x(in[i] & mask[2*k+0], 1 << k);
in[i] ^= vec128_sll_2x(in[i] & mask[2*k+1], 1 << k);
}
for (i = 0; i < GFBITS; i++)
{
v0 = vec128_extract(in[i], 0);
v1 = vec128_extract(in[i], 1);
v1 ^= v0 >> 32;
v1 ^= v1 << 32;
in[i] = vec128_set2x(v0, v1);
}
}
}
static void butterflies_tr(vec128 out[ GFBITS ], vec256 in[][ GFBITS ])
{
int i, j, k, s, b;
uint64_t tmp[ GFBITS ];
uint64_t pre[6][ GFBITS ];
uint64_t out64[2][64];
vec256 p2[ 6 ];
vec256 buf[64];
vec256 x[ GFBITS ], y[ GFBITS ];
vec256 tmp256[ GFBITS ];
uint64_t consts_ptr = 17;
const unsigned char reversal[64] =
{
0, 32, 16, 48, 8, 40, 24, 56,
4, 36, 20, 52, 12, 44, 28, 60,
2, 34, 18, 50, 10, 42, 26, 58,
6, 38, 22, 54, 14, 46, 30, 62,
1, 33, 17, 49, 9, 41, 25, 57,
5, 37, 21, 53, 13, 45, 29, 61,
3, 35, 19, 51, 11, 43, 27, 59,
7, 39, 23, 55, 15, 47, 31, 63
};
const uint16_t beta[6] = {8, 1300, 3408, 1354, 2341, 1154};
// butterflies
for (i = 3; i >= 0; i--)
{
s = 1 << i;
consts_ptr -= s;
for (j = 0; j < 16; j += 2*s)
{
for (k = j; k < j+s; k++)
{
for (b = 0; b < GFBITS; b++) in[k][b] ^= in[k+s][b];
vec256_mul(tmp256, in[k], fft_consts[ consts_ptr + (k-j) ]);
for (b = 0; b < GFBITS; b++) in[k+s][b] ^= tmp256[b];
}
}
}
for (i = 0; i < 16; i+=2)
{
for (b = 0; b < GFBITS; b++) x[b] = vec256_unpack_low(in[i+0][b], in[i+1][b]);
for (b = 0; b < GFBITS; b++) y[b] = vec256_unpack_high(in[i+0][b], in[i+1][b]);
for (b = 0; b < GFBITS; b++) x[b] ^= y[b];
vec256_mul(tmp256, x, fft_consts[ 1 ]);
for (b = 0; b < GFBITS; b++) y[b] ^= tmp256[b];
for (b = 0; b < GFBITS; b++) in[i+0][b] = vec256_unpack_low_2x(x[b], y[b]);
for (b = 0; b < GFBITS; b++) in[i+1][b] = vec256_unpack_high_2x(x[b], y[b]);
}
for (i = 0; i < 16; i+=2)
{
for (b = 0; b < GFBITS; b++) in[i+0][b] ^= in[i+1][b];
vec256_mul(tmp256, in[i+0], fft_consts[ 0 ]);
for (b = 0; b < GFBITS; b++) in[i+1][b] ^= tmp256[b];
}
// transpose
for (i = 0; i < GFBITS; i+=4)
{
for (j = 0; j < 64; j+=8)
{
buf[ reversal[j+0] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 0),
vec256_extract(in[j/4+0][i+1], 0),
vec256_extract(in[j/4+0][i+2], 0),
vec256_extract(in[j/4+0][i+3], 0));
buf[ reversal[j+1] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 0),
vec256_extract(in[j/4+1][i+1], 0),
vec256_extract(in[j/4+1][i+2], 0),
vec256_extract(in[j/4+1][i+3], 0));
buf[ reversal[j+2] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 1),
vec256_extract(in[j/4+0][i+1], 1),
vec256_extract(in[j/4+0][i+2], 1),
vec256_extract(in[j/4+0][i+3], 1));
buf[ reversal[j+3] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 1),
vec256_extract(in[j/4+1][i+1], 1),
vec256_extract(in[j/4+1][i+2], 1),
vec256_extract(in[j/4+1][i+3], 1));
buf[ reversal[j+4] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 2),
vec256_extract(in[j/4+0][i+1], 2),
vec256_extract(in[j/4+0][i+2], 2),
vec256_extract(in[j/4+0][i+3], 2));
buf[ reversal[j+5] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 2),
vec256_extract(in[j/4+1][i+1], 2),
vec256_extract(in[j/4+1][i+2], 2),
vec256_extract(in[j/4+1][i+3], 2));
buf[ reversal[j+6] ] = vec256_set4x(vec256_extract(in[j/4+0][i+0], 3),
vec256_extract(in[j/4+0][i+1], 3),
vec256_extract(in[j/4+0][i+2], 3),
vec256_extract(in[j/4+0][i+3], 3));
buf[ reversal[j+7] ] = vec256_set4x(vec256_extract(in[j/4+1][i+0], 3),
vec256_extract(in[j/4+1][i+1], 3),
vec256_extract(in[j/4+1][i+2], 3),
vec256_extract(in[j/4+1][i+3], 3));
}
transpose_64x256_sp(buf);
p2[0] = buf[32]; buf[33] = vec256_xor(buf[33], buf[32]);
p2[1] = buf[33]; buf[35] = vec256_xor(buf[35], buf[33]);
p2[0] = vec256_xor(p2[0], buf[35]); buf[34] = vec256_xor(buf[34], buf[35]);
p2[2] = buf[34]; buf[38] = vec256_xor(buf[38], buf[34]);
p2[0] = vec256_xor(p2[0], buf[38]); buf[39] = vec256_xor(buf[39], buf[38]);
p2[1] = vec256_xor(p2[1], buf[39]); buf[37] = vec256_xor(buf[37], buf[39]);
p2[0] = vec256_xor(p2[0], buf[37]); buf[36] = vec256_xor(buf[36], buf[37]);
p2[3] = buf[36]; buf[44] = vec256_xor(buf[44], buf[36]);
p2[0] = vec256_xor(p2[0], buf[44]); buf[45] = vec256_xor(buf[45], buf[44]);
p2[1] = vec256_xor(p2[1], buf[45]); buf[47] = vec256_xor(buf[47], buf[45]);
p2[0] = vec256_xor(p2[0], buf[47]); buf[46] = vec256_xor(buf[46], buf[47]);
p2[2] = vec256_xor(p2[2], buf[46]); buf[42] = vec256_xor(buf[42], buf[46]);
p2[0] = vec256_xor(p2[0], buf[42]); buf[43] = vec256_xor(buf[43], buf[42]);
p2[1] = vec256_xor(p2[1], buf[43]); buf[41] = vec256_xor(buf[41], buf[43]);
p2[0] = vec256_xor(p2[0], buf[41]); buf[40] = vec256_xor(buf[40], buf[41]);
p2[4] = buf[40]; buf[56] = vec256_xor(buf[56], buf[40]);
p2[0] = vec256_xor(p2[0], buf[56]); buf[57] = vec256_xor(buf[57], buf[56]);
p2[1] = vec256_xor(p2[1], buf[57]); buf[59] = vec256_xor(buf[59], buf[57]);
p2[0] = vec256_xor(p2[0], buf[59]); buf[58] = vec256_xor(buf[58], buf[59]);
p2[2] = vec256_xor(p2[2], buf[58]); buf[62] = vec256_xor(buf[62], buf[58]);
p2[0] = vec256_xor(p2[0], buf[62]); buf[63] = vec256_xor(buf[63], buf[62]);
p2[1] = vec256_xor(p2[1], buf[63]); buf[61] = vec256_xor(buf[61], buf[63]);
p2[0] = vec256_xor(p2[0], buf[61]); buf[60] = vec256_xor(buf[60], buf[61]);
p2[3] = vec256_xor(p2[3], buf[60]); buf[52] = vec256_xor(buf[52], buf[60]);
p2[0] = vec256_xor(p2[0], buf[52]); buf[53] = vec256_xor(buf[53], buf[52]);
p2[1] = vec256_xor(p2[1], buf[53]); buf[55] = vec256_xor(buf[55], buf[53]);
p2[0] = vec256_xor(p2[0], buf[55]); buf[54] = vec256_xor(buf[54], buf[55]);
p2[2] = vec256_xor(p2[2], buf[54]); buf[50] = vec256_xor(buf[50], buf[54]);
p2[0] = vec256_xor(p2[0], buf[50]); buf[51] = vec256_xor(buf[51], buf[50]);
p2[1] = vec256_xor(p2[1], buf[51]); buf[49] = vec256_xor(buf[49], buf[51]);
p2[0] = vec256_xor(p2[0], buf[49]); buf[48] = vec256_xor(buf[48], buf[49]);
p2[5] = buf[48]; buf[16] = vec256_xor(buf[16], buf[48]);
p2[0] = vec256_xor(p2[0], buf[16]); buf[17] = vec256_xor(buf[17], buf[16]);
p2[1] = vec256_xor(p2[1], buf[17]); buf[19] = vec256_xor(buf[19], buf[17]);
p2[0] = vec256_xor(p2[0], buf[19]); buf[18] = vec256_xor(buf[18], buf[19]);
p2[2] = vec256_xor(p2[2], buf[18]); buf[22] = vec256_xor(buf[22], buf[18]);
p2[0] = vec256_xor(p2[0], buf[22]); buf[23] = vec256_xor(buf[23], buf[22]);
p2[1] = vec256_xor(p2[1], buf[23]); buf[21] = vec256_xor(buf[21], buf[23]);
p2[0] = vec256_xor(p2[0], buf[21]); buf[20] = vec256_xor(buf[20], buf[21]);
p2[3] = vec256_xor(p2[3], buf[20]); buf[28] = vec256_xor(buf[28], buf[20]);
p2[0] = vec256_xor(p2[0], buf[28]); buf[29] = vec256_xor(buf[29], buf[28]);
p2[1] = vec256_xor(p2[1], buf[29]); buf[31] = vec256_xor(buf[31], buf[29]);
p2[0] = vec256_xor(p2[0], buf[31]); buf[30] = vec256_xor(buf[30], buf[31]);
p2[2] = vec256_xor(p2[2], buf[30]); buf[26] = vec256_xor(buf[26], buf[30]);
p2[0] = vec256_xor(p2[0], buf[26]); buf[27] = vec256_xor(buf[27], buf[26]);
p2[1] = vec256_xor(p2[1], buf[27]); buf[25] = vec256_xor(buf[25], buf[27]);
p2[0] = vec256_xor(p2[0], buf[25]); buf[24] = vec256_xor(buf[24], buf[25]);
p2[4] = vec256_xor(p2[4], buf[24]); buf[8] = vec256_xor(buf[8], buf[24]);
p2[0] = vec256_xor(p2[0], buf[8]); buf[9] = vec256_xor(buf[9], buf[8]);
p2[1] = vec256_xor(p2[1], buf[9]); buf[11] = vec256_xor(buf[11], buf[9]);
p2[0] = vec256_xor(p2[0], buf[11]); buf[10] = vec256_xor(buf[10], buf[11]);
p2[2] = vec256_xor(p2[2], buf[10]); buf[14] = vec256_xor(buf[14], buf[10]);
p2[0] = vec256_xor(p2[0], buf[14]); buf[15] = vec256_xor(buf[15], buf[14]);
p2[1] = vec256_xor(p2[1], buf[15]); buf[13] = vec256_xor(buf[13], buf[15]);
p2[0] = vec256_xor(p2[0], buf[13]); buf[12] = vec256_xor(buf[12], buf[13]);
p2[3] = vec256_xor(p2[3], buf[12]); buf[4] = vec256_xor(buf[4], buf[12]);
p2[0] = vec256_xor(p2[0], buf[4]); buf[5] = vec256_xor(buf[5], buf[4]);
p2[1] = vec256_xor(p2[1], buf[5]); buf[7] = vec256_xor(buf[7], buf[5]);
p2[0] = vec256_xor(p2[0], buf[7]); buf[6] = vec256_xor(buf[6], buf[7]);
p2[2] = vec256_xor(p2[2], buf[6]); buf[2] = vec256_xor(buf[2], buf[6]);
p2[0] = vec256_xor(p2[0], buf[2]); buf[3] = vec256_xor(buf[3], buf[2]);
p2[1] = vec256_xor(p2[1], buf[3]); buf[1] = vec256_xor(buf[1], buf[3]);
p2[0] = vec256_xor(p2[0], buf[1]); buf[0] = vec256_xor(buf[0], buf[1]);
for (j = 0; j < 6; j++)
{
pre[j][i+0] = vec256_extract(p2[j], 0);
pre[j][i+1] = vec256_extract(p2[j], 1);
pre[j][i+2] = vec256_extract(p2[j], 2);
pre[j][i+3] = vec256_extract(p2[j], 3);
}
out64[0][i+0] = vec256_extract(buf[0], 0);
out64[0][i+1] = vec256_extract(buf[0], 1);
out64[0][i+2] = vec256_extract(buf[0], 2);
out64[0][i+3] = vec256_extract(buf[0], 3);
}
//
for (j = 0; j < GFBITS; j++) { tmp[j] = (beta[0] >> j) & 1; tmp[j] = -tmp[j]; }
vec_mul(out64[1], pre[0], tmp);
for (i = 1; i < 6; i++)
{
for (j = 0; j < GFBITS; j++) { tmp[j] = (beta[i] >> j) & 1; tmp[j] = -tmp[j]; }
vec_mul(tmp, pre[i], tmp);
vec_add(out64[1], out64[1], tmp);
}
for (i = 0; i < GFBITS; i++)
out[i] = vec128_set2x(out64[0][i], out64[1][i]);
}
void fft_tr(vec128 out[GFBITS], vec256 in[][ GFBITS ])
{
butterflies_tr(out, in);
radix_conversions_tr(out);
}