Commit 6daf1d8

authored

[Paged attention] Support kernel with VNNI layout (#3587)

* [paged attention] add kernel with vnni

1 parent d4174c8 commit 6daf1d8Copy full SHA for 6daf1d8

2 files changed

+660

-1

lines changed

csrc/cpu
- aten/kernels
  - PagedAttentionKrnl.cpp
- vec/vec512/perf_kernel
  - add_softmax.h

+660

-1

lines changed

Comments

(0)