ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-19 22:39:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support transposed C tile in Aquant (#2679)

Browse Source 
The performance of Aquant has increased after enabling transposed C.

Do not need to exchange AQ elements among lanes after enabling
transposed C as one thread only holds data from one row.
This commit is contained in:
Cong Ma
2025-08-28 14:28:09 -06:00
committed by GitHub
parent 0758883fa4
commit 428090f749
10 changed files with 276 additions and 154 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
example/ck_tile/38_block_scale_gemm/gemm_aquant_basic.cpp
View File

@@ -71,7 +71,7 @@ float gemm_calc_aquant(const ck_tile::AQuantGemmHostArgs& args, const ck_tile::s
const ck_tile::index_t num_loop = TilePartitioner::GetLoopNum(K_split);
const bool has_hot_loop = BaseGemmPipeline::BlockHasHotloop(num_loop);
const ck_tile::TailNumber tail_num = BaseGemmPipeline::GetBlockLoopTailNum(num_loop);
constexpr bool transposed_warp_gemm = false;
constexpr bool transposed_warp_gemm = true;
const auto Run = [&](const auto has_hot_loop_, const auto tail_number_) {
constexpr bool has_hot_loop_v = has_hot_loop_.value;
@@ -85,6 +85,7 @@ float gemm_calc_aquant(const ck_tile::AQuantGemmHostArgs& args, const ck_tile::s
CodegenGemmShape,
CodegenGemmTraits,
QuantGroupSize,
transposed_warp_gemm,
ComputeDataType,
ck_tile::GemmPipelineScheduler::Intrawave,
has_hot_loop_v,

1
example/ck_tile/38_block_scale_gemm/gemm_aquant_preshuffle.cpp
View File

@@ -85,6 +85,7 @@ float gemm_calc_aquant(const ck_tile::AQuantGemmHostArgs& args, const ck_tile::s
CodegenGemmShape,
CodegenGemmTraits,
QuantGroupSize,
transposed_warp_gemm,
ComputeDataType,
ck_tile::GemmPipelineScheduler::Intrawave,
has_hot_loop_v,