diff options
| author | Casey Primozic <casey@cprimozic.net> | 2023-03-21 07:35:42 -0700 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-03-21 15:35:42 +0100 |
| commit | 2e664f1ff413995506c9a54f3a8d5b8c64e37a91 (patch) | |
| tree | 0162d9c81e72e85d21a5806b35dbefc6587c105d /main.cpp | |
| parent | 8cf9f34eddc124d4ab28f4d2fe8e99d574510bde (diff) | |
Add initial AVX512 support for dot product on Linux (#320)
* Update Makefile to detect AVX512 support and add compiler flags if it's available
* Based on existing AVX2 implementation, dot product on one 32-value block of 4-bit quantized ints at a time
* Perform 8 bit -> 16 bit sign extension and multiply+add on 32 values at time instead of 16
* Use built-in AVX512 horizontal reduce add to get sum at the end
* Manual unrolling on inner dot product loop to reduce loop counter overhead
Diffstat (limited to 'main.cpp')
0 files changed, 0 insertions, 0 deletions