Function libdivide_s32_do_vector

numpy/core/include/numpy/libdivide/libdivide.h:1339–1366

////////// SINT32

__m512i libdivide_s32_do_vector(__m512i numers, const struct libdivide_s32_t *denom) {
    uint8_t more = denom->more;
    if (!denom->magic) {
        uint32_t shift = more & LIBDIVIDE_32_SHIFT_MASK;
        uint32_t mask = (1U << shift) - 1;
        __m512i roundToZeroTweak = _mm512_set1_epi32(mask);
        // q = numer + ((numer >> 31) & roundToZeroTweak);
        __m512i q = _mm512_add_epi32(numers, _mm512_and_si512(_mm512_srai_epi32(numers, 31), roundToZeroTweak));
        q = _mm512_srai_epi32(q, shift);
        __m512i sign = _mm512_set1_epi32((int8_t)more >> 7);
        // q = (q ^ sign) - sign;
        q = _mm512_sub_epi32(_mm512_xor_si512(q, sign), sign);
        return q;
    }
    else {
        __m512i q = libdivide_mullhi_s32_vector(numers, _mm512_set1_epi32(denom->magic));
        if (more & LIBDIVIDE_ADD_MARKER) {
             // must be arithmetic shift
            __m512i sign = _mm512_set1_epi32((int8_t)more >> 7);
             // q += ((numer ^ sign) - sign);
            q = _mm512_add_epi32(q, _mm512_sub_epi32(_mm512_xor_si512(numers, sign), sign));
        }
        // q >>= shift
        q = _mm512_srai_epi32(q, more & LIBDIVIDE_32_SHIFT_MASK);
        q = _mm512_add_epi32(q, _mm512_srli_epi32(q, 31)); // q += (q < 0)
        return q;
    }
}

__m512i libdivide_s32_branchfree_do_vector(__m512i numers, const struct libdivide_s32_branchfree_t *denom) {
    int32_t magic = denom->magic;