ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 00:40:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

update pipeline v1: add atomic IGLP schedule

Browse Source
This commit is contained in:
AMD-dteng
2025-07-29 22:48:00 +08:00
committed by Feng Shijie
parent f9e48148d2
commit a587701117
4 changed files with 429 additions and 811 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
example/ck_tile/18_flatmm/flatmm_basic.cpp
View File

@@ -155,7 +155,7 @@ float flatmm_calc(const ck_tile::ScaleFlatmmHostArgs<ScaleM, ScaleN>& args,
using GemmPipelineProblem =
ck_tile::GemmPipelineProblem<ADataType, BDataType, AccDataType, CodegenFlatmmShape, Traits>;
using BaseGemmPipeline = ck_tile::BaseFlatmmPipelineAGmemBGmemCRegV0<GemmPipelineProblem>;
using BaseGemmPipeline = ck_tile::BaseFlatmmPipelineAGmemBGmemCRegV1<GemmPipelineProblem>;
const ck_tile::index_t k_grain = args.k_batch * FlatmmConfig::K_Tile;
const ck_tile::index_t K_split = (args.K + k_grain - 1) / k_grain * FlatmmConfig::K_Tile;
@@ -182,7 +182,7 @@ float flatmm_calc(const ck_tile::ScaleFlatmmHostArgs<ScaleM, ScaleN>& args,
tail_number_v>;
using CodegenFlatmmPipeline =
ck_tile::FlatmmPipelineAGmemBGmemCRegV0<CodegenPipelineProblem>;
ck_tile::FlatmmPipelineAGmemBGmemCRegV1<CodegenPipelineProblem>;
using GemmEpilogue = ck_tile::CShuffleEpilogue<
ck_tile::CShuffleEpilogueProblem<ADataType,