pqc/crypto_kem/hqc-rmrs-256/avx2/gf.c
2021-03-24 21:02:48 +00:00

177 lines
5.1 KiB
C

#include "gf.h"
#include "parameters.h"
#include <stdint.h>
/**
* @file gf.c
* Galois field implementation with multiplication using the pclmulqdq instruction
*/
static uint16_t gf_reduce(uint64_t x, size_t deg_x);
/**
* Reduces polynomial x modulo primitive polynomial GF_POLY.
* @returns x mod GF_POLY
* @param[in] x Polynomial of degree less than 64
* @param[in] deg_x The degree of polynomial x
*/
static uint16_t gf_reduce(uint64_t x, size_t deg_x) {
uint16_t z1, z2, rmdr, dist;
uint64_t mod;
size_t steps, i, j;
// Deduce the number of steps of reduction
steps = CEIL_DIVIDE(deg_x - (PARAM_M - 1), PARAM_GF_POLY_M2);
// Reduce
for (i = 0; i < steps; ++i) {
mod = x >> PARAM_M;
x &= (1 << PARAM_M) - 1;
x ^= mod;
z1 = 0;
rmdr = PARAM_GF_POLY ^ 1;
for (j = PARAM_GF_POLY_WT - 2; j; --j) {
z2 = __tzcnt_u16(rmdr);
dist = (uint16_t) (z2 - z1);
mod <<= dist;
x ^= mod;
rmdr ^= 1 << z2;
z1 = z2;
}
}
return x;
}
/**
* Multiplies two elements of GF(2^GF_M).
* @returns the product a*b
* @param[in] a Element of GF(2^GF_M)
* @param[in] b Element of GF(2^GF_M)
*/
uint16_t PQCLEAN_HQCRMRS256_AVX2_gf_mul(uint16_t a, uint16_t b) {
__m128i va = _mm_cvtsi32_si128(a);
__m128i vb = _mm_cvtsi32_si128(b);
__m128i vab = _mm_clmulepi64_si128(va, vb, 0);
uint32_t ab = _mm_cvtsi128_si32(vab);
return gf_reduce(ab, 2 * (PARAM_M - 1));
}
/**
* Compute 16 products in GF(2^GF_M).
* @returns the product (a0b0,a1b1,...,a15b15) , ai,bi in GF(2^GF_M)
* @param[in] a 256-bit register where a0,..,a15 are stored as 16 bit integers
* @param[in] b 256-bit register where b0,..,b15 are stored as 16 bit integer
*
*/
__m256i PQCLEAN_HQCRMRS256_AVX2_gf_mul_vect(__m256i a, __m256i b) {
__m128i al = _mm256_extractf128_si256(a, 0);
__m128i ah = _mm256_extractf128_si256(a, 1);
__m128i bl = _mm256_extractf128_si256(b, 0);
__m128i bh = _mm256_extractf128_si256(b, 1);
__m128i abl0 = _mm_clmulepi64_si128(al & CONST128_MASKL, bl & CONST128_MASKL, 0x0);
abl0 &= CONST128_MIDDLEMASKL;
abl0 ^= (_mm_clmulepi64_si128(al & CONST128_MASKH, bl & CONST128_MASKH, 0x0) & CONST128_MIDDLEMASKH);
__m128i abh0 = _mm_clmulepi64_si128(al & CONST128_MASKL, bl & CONST128_MASKL, 0x11);
abh0 &= CONST128_MIDDLEMASKL;
abh0 ^= (_mm_clmulepi64_si128(al & CONST128_MASKH, bl & CONST128_MASKH, 0x11) & CONST128_MIDDLEMASKH);
abl0 = _mm_shuffle_epi8(abl0, CONST128_INDEXL);
abl0 ^= _mm_shuffle_epi8(abh0, CONST128_INDEXH);
__m128i abl1 = _mm_clmulepi64_si128(ah & CONST128_MASKL, bh & CONST128_MASKL, 0x0);
abl1 &= CONST128_MIDDLEMASKL;
abl1 ^= (_mm_clmulepi64_si128(ah & CONST128_MASKH, bh & CONST128_MASKH, 0x0) & CONST128_MIDDLEMASKH);
__m128i abh1 = _mm_clmulepi64_si128(ah & CONST128_MASKL, bh & CONST128_MASKL, 0x11);
abh1 &= CONST128_MIDDLEMASKL;
abh1 ^= (_mm_clmulepi64_si128(ah & CONST128_MASKH, bh & CONST128_MASKH, 0x11) & CONST128_MIDDLEMASKH);
abl1 = _mm_shuffle_epi8(abl1, CONST128_INDEXL);
abl1 ^= _mm_shuffle_epi8(abh1, CONST128_INDEXH);
__m256i ret = _mm256_set_m128i(abl1, abl0);
__m256i aux = CONST256_MR0;
for (int32_t i = 0; i < 7; i++) {
ret ^= red[i] & _mm256_cmpeq_epi16((ret & aux), aux);
aux = aux << 1;
}
ret &= CONST256_LASTMASK;
return ret;
}
/**
* Squares an element of GF(2^GF_M).
* @returns a^2
* @param[in] a Element of GF(2^GF_M)
*/
uint16_t PQCLEAN_HQCRMRS256_AVX2_gf_square(uint16_t a) {
uint32_t b = a;
uint32_t s = b & 1;
for (size_t i = 1; i < PARAM_M; ++i) {
b <<= 1;
s ^= b & (1 << 2 * i);
}
return gf_reduce(s, 2 * (PARAM_M - 1));
}
/**
* Computes the inverse of an element of GF(2^8),
* using the addition chain 1 2 3 4 7 11 15 30 60 120 127 254
* @returns the inverse of a
* @param[in] a Element of GF(2^GF_M)
*/
uint16_t PQCLEAN_HQCRMRS256_AVX2_gf_inverse(uint16_t a) {
uint16_t inv = a;
uint16_t tmp1, tmp2;
inv = PQCLEAN_HQCRMRS256_AVX2_gf_square(a); /* a^2 */
tmp1 = PQCLEAN_HQCRMRS256_AVX2_gf_mul(inv, a); /* a^3 */
inv = PQCLEAN_HQCRMRS256_AVX2_gf_square(inv); /* a^4 */
tmp2 = PQCLEAN_HQCRMRS256_AVX2_gf_mul(inv, tmp1); /* a^7 */
tmp1 = PQCLEAN_HQCRMRS256_AVX2_gf_mul(inv, tmp2); /* a^11 */
inv = PQCLEAN_HQCRMRS256_AVX2_gf_mul(tmp1, inv); /* a^15 */
inv = PQCLEAN_HQCRMRS256_AVX2_gf_square(inv); /* a^30 */
inv = PQCLEAN_HQCRMRS256_AVX2_gf_square(inv); /* a^60 */
inv = PQCLEAN_HQCRMRS256_AVX2_gf_square(inv); /* a^120 */
inv = PQCLEAN_HQCRMRS256_AVX2_gf_mul(inv, tmp2); /* a^127 */
inv = PQCLEAN_HQCRMRS256_AVX2_gf_square(inv); /* a^254 */
return inv;
}
/**
* Returns i modulo 2^GF_M-1.
* i must be less than 2*(2^GF_M-1).
* Therefore, the return value is either i or i-2^GF_M+1.
* @returns i mod (2^GF_M-1)
* @param[in] i The integer whose modulo is taken
*/
uint16_t PQCLEAN_HQCRMRS256_AVX2_gf_mod(uint16_t i) {
uint16_t tmp = (uint16_t) (i - PARAM_GF_MUL_ORDER);
// mask = 0xffff if (i < GF_MUL_ORDER)
uint16_t mask = -(tmp >> 15);
return tmp + (mask & PARAM_GF_MUL_ORDER);
}