ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-16 19:09:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix the Composable Kernel CI and versions incompatibility (#4640)

Browse Source 
## Motivation

This PR has 4 patches:
1. Fix the CI error of grouped gemm.
2. Fix the incompatibility of old linux version.
3. Fix the potential errors of flatmm.
4. Address the previous comments of abquant eight warps pipeline
solution.

---------

Co-authored-by: illsilin_amdeng <Illia.Silin@amd.com>
This commit is contained in:
Thomas Ning
2026-02-18 22:59:37 +08:00
committed by GitHub
parent 058be6c6e9
commit be25dd6775
12 changed files with 67 additions and 65 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
include/ck_tile/ops/gemm_quant/block/block_universal_gemm_as_aquant_bs_bquant_cr_async.hpp
View File

@@ -76,7 +76,8 @@ struct ABQuantBlockUniversalGemmAsBsCrAsync : public BlockGemmQuantBase
static constexpr index_t NIterPerWarp = NPerBlock / (NWarp * WarpGemm::kN);
static constexpr index_t KIterPerWarp = KPerBlock / (KWarp * WarpGemm::kK);
static constexpr bool PreshuffleQuant = Problem::Traits::PreshuffleQuant;
static constexpr bool APreshuffleQuant = Problem::Traits::APreshuffleQuant;
static constexpr bool BPreshuffleQuant = Problem::Traits::BPreshuffleQuant;
static constexpr index_t QScalesPerBlockRow =
integer_divide_ceil(KPerBlock / KWarp, BQuantGroupSize::kK);
@@ -158,7 +159,8 @@ struct ABQuantBlockUniversalGemmAsBsCrAsync : public BlockGemmQuantBase
using BWarpTensor = typename WarpGemm::BWarpTensor;
using CWarpTensor = typename WarpGemm::CWarpTensor;
static constexpr bool PreshuffleQuant = Traits::PreshuffleQuant;
static constexpr bool APreshuffleQuant = Traits::APreshuffleQuant;
static constexpr bool BPreshuffleQuant = Traits::BPreshuffleQuant;
static_assert(std::is_same_v<typename WarpGemm::CDataType, float>);
@@ -364,7 +366,7 @@ struct ABQuantBlockUniversalGemmAsBsCrAsync : public BlockGemmQuantBase
AQPickerCommon<AQBlockTensor, Traits, mIter, kQScale> aq_picker(
aq_block_tensor);
if constexpr(PreshuffleQuant)
if constexpr(BPreshuffleQuant)
{
constexpr index_t reg_offset = nIter;
auto pull_from_lane =