pqc/crypto_sign/dilithium5/avx2/rounding.c

#include "consts.h"
#include "params.h"
#include "rejsample.h"
#include "rounding.h"
#include <immintrin.h>
#include <stdint.h>
#include <string.h>

#define _mm256_blendv_epi32(a,b,mask) \
    _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(a), \
                                         _mm256_castsi256_ps(b), \
                                         _mm256_castsi256_ps(mask)))

/*************************************************
* Name:        power2round
*
* Description: For finite field elements a, compute a0, a1 such that
*              a mod^+ Q = a1*2^D + a0 with -2^{D-1} < a0 <= 2^{D-1}.
*              Assumes a to be positive standard representative.
*
* Arguments:   - __m256i *a1: output array of length N/8 with high bits
*              - __m256i *a0: output array of length N/8 with low bits a0
*              - const __m256i *a: input array of length N/8
*
**************************************************/
void PQCLEAN_DILITHIUM5_AVX2_power2round_avx(__m256i *a1, __m256i *a0, const __m256i *a) {
    unsigned int i;
    __m256i f, f0, f1;
    const __m256i mask = _mm256_set1_epi32(-(1 << D));
    const __m256i half = _mm256_set1_epi32((1 << (D - 1)) - 1);

    for (i = 0; i < N / 8; ++i) {
        f = _mm256_load_si256(&a[i]);
        f1 = _mm256_add_epi32(f, half);
        f0 = _mm256_and_si256(f1, mask);
        f1 = _mm256_srli_epi32(f1, D);
        f0 = _mm256_sub_epi32(f, f0);
        _mm256_store_si256(&a1[i], f1);
        _mm256_store_si256(&a0[i], f0);
    }
}

/*************************************************
* Name:        decompose
*
* Description: For finite field element a, compute high and low parts a0, a1 such
*              that a mod^+ Q = a1*ALPHA + a0 with -ALPHA/2 < a0 <= ALPHA/2 except
*              if a1 = (Q-1)/ALPHA where we set a1 = 0 and
*              -ALPHA/2 <= a0 = a mod Q - Q < 0. Assumes a to be positive standard
*              representative.
*
* Arguments:   - __m256i *a1: output array of length N/8 with high parts
*              - __m256i *a0: output array of length N/8 with low parts a0
*              - const __m256i *a: input array of length N/8
*
**************************************************/
void PQCLEAN_DILITHIUM5_AVX2_decompose_avx(__m256i *a1, __m256i *a0, const __m256i *a) {
    unsigned int i;
    __m256i f, f0, f1;
    const __m256i q = _mm256_load_si256(&PQCLEAN_DILITHIUM5_AVX2_qdata.vec[_8XQ / 8]);
    const __m256i hq = _mm256_srli_epi32(q, 1);
    const __m256i v = _mm256_set1_epi32(1025);
    const __m256i alpha = _mm256_set1_epi32(2 * GAMMA2);
    const __m256i off = _mm256_set1_epi32(127);
    const __m256i shift = _mm256_set1_epi32(512);
    const __m256i mask = _mm256_set1_epi32(15);

    for (i = 0; i < N / 8; i++) {
        f = _mm256_load_si256(&a[i]);
        f1 = _mm256_add_epi32(f, off);
        f1 = _mm256_srli_epi32(f1, 7);
        f1 = _mm256_mulhi_epu16(f1, v);
        f1 = _mm256_mulhrs_epi16(f1, shift);
        f1 = _mm256_and_si256(f1, mask);
        f0 = _mm256_mullo_epi32(f1, alpha);
        f0 = _mm256_sub_epi32(f, f0);
        f = _mm256_cmpgt_epi32(f0, hq);
        f = _mm256_and_si256(f, q);
        f0 = _mm256_sub_epi32(f0, f);
        _mm256_store_si256(&a1[i], f1);
        _mm256_store_si256(&a0[i], f0);
    }
}


/*************************************************
* Name:        make_hint
*
* Description: Compute indices of polynomial coefficients whose low bits
*              overflow into the high bits.
*
* Arguments:   - uint8_t *hint: hint array
*              - const __m256i *a0: low bits of input elements
*              - const __m256i *a1: high bits of input elements
*
* Returns number of overflowing low bits
**************************************************/
unsigned int PQCLEAN_DILITHIUM5_AVX2_make_hint_avx(uint8_t hint[N], const __m256i *restrict a0, const __m256i *restrict a1) {
    unsigned int i, n = 0;
    __m256i f0, f1, g0, g1;
    uint32_t bad;
    uint64_t idx;
    const __m256i low = _mm256_set1_epi32(-GAMMA2);
    const __m256i high = _mm256_set1_epi32(GAMMA2);

    for (i = 0; i < N / 8; ++i) {
        f0 = _mm256_load_si256(&a0[i]);
        f1 = _mm256_load_si256(&a1[i]);
        g0 = _mm256_abs_epi32(f0);
        g0 = _mm256_cmpgt_epi32(g0, high);
        g1 = _mm256_cmpeq_epi32(f0, low);
        g1 = _mm256_sign_epi32(g1, f1);
        g0 = _mm256_or_si256(g0, g1);

        bad = _mm256_movemask_ps((__m256)g0);
        memcpy(&idx, PQCLEAN_DILITHIUM5_AVX2_idxlut[bad], 8);
        idx += (uint64_t)0x0808080808080808 * i;
        memcpy(&hint[n], &idx, 8);
        n += _mm_popcnt_u32(bad);
    }

    return n;
}

/*************************************************
* Name:        use_hint
*
* Description: Correct high parts according to hint.
*
* Arguments:   - __m256i *b: output array of length N/8 with corrected high parts
*              - const __m256i *a: input array of length N/8
*              - const __m256i *a: input array of length N/8 with hint bits
*
**************************************************/
void PQCLEAN_DILITHIUM5_AVX2_use_hint_avx(__m256i *b, const __m256i *a, const __m256i *restrict hint) {
    unsigned int i;
    __m256i a0[N / 8];
    __m256i f, g, h, t;
    const __m256i zero = _mm256_setzero_si256();
    const __m256i mask = _mm256_set1_epi32(15);

    PQCLEAN_DILITHIUM5_AVX2_decompose_avx(b, a0, a);
    for (i = 0; i < N / 8; i++) {
        f = _mm256_load_si256(&a0[i]);
        g = _mm256_load_si256(&b[i]);
        h = _mm256_load_si256(&hint[i]);
        t = _mm256_blendv_epi32(zero, h, f);
        t = _mm256_slli_epi32(t, 1);
        h = _mm256_sub_epi32(h, t);
        g = _mm256_add_epi32(g, h);
        g = _mm256_and_si256(g, mask);
        _mm256_store_si256(&b[i], g);
    }
}