composable_kernel icon indicating copy to clipboard operation
composable_kernel copied to clipboard

[Ck tile] Use raw store to improve layernorm performance

Open rocking5566 opened this issue 1 year ago • 0 comments

Proposed changes

  1. Simpler kernel example for layernorm
  2. use store_tile_raw for Default2DEpilogueProblem to improve performance

Checklist

use following command to check performance make -j tile_layernorm2d_fwd && ./bin/tile_layernorm2d_fwd -m=128 -n=8192 -prec_i=bf16 -fadd=1

rocking5566 avatar Dec 16 '24 19:12 rocking5566