[caffe2/fakelowp] optimize ref int8 gemm (#38294)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38294
Optimize the reference int8 gemm using avx2 intrinsics
Test Plan:
Before this diff
7.72164 GF/s
After this diff
27.7731 GF/s
Reviewed By: amylittleyang
Differential Revision: D21516439
fbshipit-source-id: 2b596605eec6a338a295701a01cf2c8639204274
