[libc-commits] [libc] [libc] Fix buggy AVX2 `memcmp` (PR #77081)

Nafi Rouf via libc-commits libc-commits at lists.llvm.org
Sat Jan 6 03:21:23 PST 2024


https://github.com/nafi3000 commented:

FWIW, I believe `__m512i bytewise_reverse(__m512i value)` is also buggy at the moment!
https://godbolt.org/z/e8sfvP7h4

https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#text=_mm512_shuffle_epi8&ig_expand=6009
`index[5:0] := b[i+3:i] + (j & 0x30)`
It is only considering 4 lower bits from each byte of `b` and then adding 4th and 5th bits from `j` which is basically the 128-bit lane index. So, it is implemented the same way as `_mm256_shuffle_epi8`. :(


https://github.com/llvm/llvm-project/pull/77081


More information about the libc-commits mailing list