-
Dmitry Kazakov authored
_mm256_packus_epi16() doesn't work like _mm_packus_epi16(). It doesn't concatenate two vectors, but permutates them. So we need to permute them back to make the result correct.
61ac490c
_mm256_packus_epi16() doesn't work like _mm_packus_epi16(). It doesn't concatenate two vectors, but permutates them. So we need to permute them back to make the result correct.