jianxind commented on pull request #7029:
URL: https://github.com/apache/arrow/pull/7029#issuecomment-623845147
> A general question: why is this limited to `sizeof(T) == 4` and `sizeof(T)
== 8`? There are 8-bit and 16-bit types as well.
_mm512_mask_expand_epi16/_mm512_mask_expand_ep
jianxind commented on pull request #7029:
URL: https://github.com/apache/arrow/pull/7029#issuecomment-621704246
> I'd gladly see a AVX2 or SSE version indeed, as many CPUs don't have
AVX512.
@pitrou @emkornfield
Yeah, I has a version of SSE, would you like me to append it to thi
jianxind commented on pull request #7029:
URL: https://github.com/apache/arrow/pull/7029#issuecomment-619501029
> Just curious if you see and impact on parquet-arrow-reader-writer
benchmarks? That is the ultimate goal of the speedup.
No impact, I checked all items for parquet-arrow-r
jianxind commented on pull request #7029:
URL: https://github.com/apache/arrow/pull/7029#issuecomment-618855696
cc @emkornfield
The AVX512 path is straightforward as the helper of
mask_compress/mask_expand API provide by AVX512. For potential path-finding of
SSE/AVX2, as you pointe