ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 05:01:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] Enhance RMSNorm Accuracy: New Pipeline Pass for Selectable Implementation (#2409)

Browse Source 
* Add Rmsnorm2dFwdPipelineModelSensitiveT5Pass

* Update rmsnorm2d_fwd_pipeline_model_sensitive_pass

1.  Add BlockReduce2dTreeCrossWarpSync

* Add Rmsnorm2dFusedModelSensitiveEnum

* Update patch

1. Reverse generate.py
2. Remove comment in generate.py
3. Update tree cross warp reduce

* Refactor RMSNorm model enum and introduce T5-like option

* Update the n stage for cross warp reduce

* Add new cmdline option in RMSNorm for new pipeline testing

---------

Co-authored-by: Clement Lin <clement.lin@amd.com>
Co-authored-by: ClementLinCF <162283536+ClementLinCF@users.noreply.github.com>
This commit is contained in:
MHYangAMD
2025-07-16 14:05:26 +08:00
committed by GitHub
parent 6b09f0823e
commit 3499fe67ff
13 changed files with 730 additions and 184 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
include/ck_tile/ops/rmsnorm2d.hpp
View File

@@ -5,6 +5,7 @@
#include "ck_tile/ops/rmsnorm2d/kernel/rmsnorm2d_fwd_kernel.hpp"
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_default_policy.hpp"
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_model_sensitive_pass.hpp"
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_one_pass.hpp"
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_problem.hpp"
#include "ck_tile/ops/rmsnorm2d/pipeline/rmsnorm2d_fwd_pipeline_two_pass.hpp"