Function libdivide_s64_do_vector

numpy/core/include/numpy/libdivide/libdivide.h:1391–1419

////////// SINT64

__m512i libdivide_s64_do_vector(__m512i numers, const struct libdivide_s64_t *denom) {
    uint8_t more = denom->more;
    int64_t magic = denom->magic;
    if (magic == 0) { // shift path
        uint32_t shift = more & LIBDIVIDE_64_SHIFT_MASK;
        uint64_t mask = (1ULL << shift) - 1;
        __m512i roundToZeroTweak = _mm512_set1_epi64(mask);
        // q = numer + ((numer >> 63) & roundToZeroTweak);
        __m512i q = _mm512_add_epi64(numers, _mm512_and_si512(libdivide_s64_signbits(numers), roundToZeroTweak));
        q = libdivide_s64_shift_right_vector(q, shift);
        __m512i sign = _mm512_set1_epi32((int8_t)more >> 7);
         // q = (q ^ sign) - sign;
        q = _mm512_sub_epi64(_mm512_xor_si512(q, sign), sign);
        return q;
    }
    else {
        __m512i q = libdivide_mullhi_s64_vector(numers, _mm512_set1_epi64(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_epi64(q, _mm512_sub_epi64(_mm512_xor_si512(numers, sign), sign));
        }
        // q >>= denom->mult_path.shift
        q = libdivide_s64_shift_right_vector(q, more & LIBDIVIDE_64_SHIFT_MASK);
        q = _mm512_add_epi64(q, _mm512_srli_epi64(q, 63)); // q += (q < 0)
        return q;
    }
}

__m512i libdivide_s64_branchfree_do_vector(__m512i numers, const struct libdivide_s64_branchfree_t *denom) {
    int64_t magic = denom->magic;