ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 14:59:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK TILE GEMM] Fixed the regression issue with transpose C in Quant Gemm (#2819)

Browse Source 
The numerical error was introduced after merging row/col quant. And it is fixed.
This commit is contained in:
Cong Ma
2025-09-11 12:38:16 -06:00
committed by GitHub
parent 28e12c62a2
commit 2ed39f8d91
2 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
example/ck_tile/38_block_scale_gemm/gemm_quant_basic.cpp
View File

@@ -57,10 +57,9 @@ float gemm_calc_quant(const ck_tile::QuantGemmHostArgs& args, const ck_tile::str
const ck_tile::index_t K_split =
(args.K + GemmConfig::K_Tile - 1) / GemmConfig::K_Tile * GemmConfig::K_Tile;
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;
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);
const auto Run = [&](const auto has_hot_loop_, const auto tail_number_) {
constexpr bool has_hot_loop_v = has_hot_loop_.value;
@@ -128,7 +127,7 @@ float gemm_calc_quant(const ck_tile::QuantGemmHostArgs& args, const ck_tile::str
GemmConfig::M_Warp_Tile,
GemmConfig::N_Warp_Tile,
GemmConfig::K_Warp_Tile,
transposed_warp_gemm,
transpose_c,
ck_tile::memory_operation_enum::set>>;
using Kernel =
ck_tile::QuantGemmKernel<TilePartitioner, GemmPipeline, GemmEpilogue, QuantMode>;