[CK_TILE] Fix the Wrong Output Generated by Gemm Examples on GFX11/12 (#2713)

* Introduce macro CK_TILE_USE_WMMA

Signed-off-by: Tianyuan Wu <Tianyuan.Wu@amd.com>

* Make CK_TILE_USE_WMMA global for all examples

Signed-off-by: Tianyuan Wu <Tianyuan.Wu@amd.com>

* Remove CK_TILE_USE_WMMA from config.hpp

Signed-off-by: Tianyuan Wu <Tianyuan.Wu@amd.com>

---------

Signed-off-by: Tianyuan Wu <Tianyuan.Wu@amd.com>
This commit is contained in:
Tianyuan Wu
2025-08-26 03:55:35 +08:00
committed by GitHub
parent 1d4a3341f0
commit e9605ed36d
4 changed files with 18 additions and 11 deletions

2
example/ck_tile/03_gemm/gemm_utils.hpp Executable file → Normal file
View File

@@ -172,6 +172,7 @@ struct GemmConfigComputeV3_2 : public GemmConfigBase
static constexpr int kBlockPerCu = 2;
};
#if CK_TILE_USE_WMMA
template <typename PrecType>
struct GemmConfigComputeV3_WMMA : public GemmConfigBase
{
@@ -192,6 +193,7 @@ struct GemmConfigComputeV3_WMMA : public GemmConfigBase
static constexpr int kBlockPerCu = 2;
};
#endif
template <typename PrecType>
struct GemmConfigComputeV4 : public GemmConfigBase