in gemmology.h [207:215]
inline xsimd::batch<int32_t, Arch> Pack0123(xsimd::batch<int32_t, Arch> sum0,
xsimd::batch<int32_t, Arch> sum1,
xsimd::batch<int32_t, Arch> sum2,
xsimd::batch<int32_t, Arch> sum3,
xsimd::kernel::requires_arch<xsimd::avx2>) {
auto pack01 = _mm256_hadd_epi32(sum0, sum1);
auto pack23 = _mm256_hadd_epi32(sum2, sum3);
return _mm256_hadd_epi32(pack01, pack23);
}