Congma/ck tile/aquant mem pipeline (#3346)

* [CK TILE GEMM QUANT] Fix the bug in HotLoopTail of memory pipeline
This commit is contained in:
Cong Ma
2025-12-05 09:35:27 -07:00
committed by GitHub
parent 7541d9b5b0
commit ed080f5a56
3 changed files with 127 additions and 43 deletions

View File

@@ -69,7 +69,12 @@ float gemm_calc_quant(const ck_tile::QuantGemmHostArgs& args, const ck_tile::str
using BaseGemmPipeline = std::conditional_t<
GemmConfig::PreshuffleB == true,
ck_tile::BaseWeightPreshufflePipelineAGmemBGmemCRegV2<GemmPipelineProblem>,
ck_tile::BaseGemmPipelineAgBgCrCompV3<GemmPipelineProblem>>;
std::conditional_t<
QuantMode == ck_tile::QuantType::AQuantGrouped && GemmConfig::PreshuffleQuant == true,
ck_tile::BaseGemmPipelineAgBgCrCompV3<GemmPipelineProblem>,
std::conditional_t<QuantMode == ck_tile::QuantType::AQuantGrouped,
ck_tile::BaseGemmPipelineAgBgCrMem<GemmPipelineProblem>,
ck_tile::BaseGemmPipelineAgBgCrCompV3<GemmPipelineProblem>>>>;
const ck_tile::index_t K_split =
(args.K + GemmConfig::K_Tile - 1) / GemmConfig::K_Tile * GemmConfig::K_Tile;
@@ -128,7 +133,9 @@ float gemm_calc_quant(const ck_tile::QuantGemmHostArgs& args, const ck_tile::str
ck_tile::GemmPipelineAgBgCrCompV3<PipelineProblem>,
std::conditional_t<
QuantMode == ck_tile::QuantType::AQuantGrouped,
ck_tile::AQuantGemmPipelineAgBgCrCompV3<PipelineProblem>,
std::conditional_t<GemmConfig::PreshuffleQuant == true,
ck_tile::AQuantGemmPipelineAgBgCrCompV3<PipelineProblem>,
ck_tile::AQuantGemmPipelineAgBgCrMem<PipelineProblem>>,
std::conditional_t<GemmConfig::PreshuffleB == true,
ck_tile::WPQuantBPipelineAgBgCrV2<PipelineProblem>,
ck_tile::BQuantGemmPipelineAgBgCrCompV3<PipelineProblem>>>>;