Skip to content

GitLab

"...composable_kernel_rocm.git" did not exist on "f18cfec43c60abb91e5e38a1205a36b05004123c"
Commit fc62babb authored by Jing Zhang's avatar Jing Zhang
Browse files

seperate gfx12 blockwise_gemm

parent f3111877
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 354 additions and 49 deletions
include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
View file @ fc62babb
...@@ -94,13 +94,16 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout, ...@@ -94,13 +94,16 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
(MWaves == 1 && is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value) ? false : true; (MWaves == 1 && is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value) ? false : true;
// If true, LDS is used unconditionally // If true, LDS is used unconditionally
#ifdef __gfx12__
static constexpr auto AEnableLds_manu = true;
static constexpr auto BEnableLds_manu = true;
#else
static constexpr auto AEnableLds_manu = false; static constexpr auto AEnableLds_manu = false;
static constexpr auto BEnableLds_manu = false; static constexpr auto BEnableLds_manu = false;
#endif
static constexpr auto AEnableLds = static constexpr auto AEnableLds = AEnableLds_auto || AEnableLds_manu || (NumPrefetch > 1);
false; // AEnableLds_auto || AEnableLds_manu || (NumPrefetch > 1); static constexpr auto BEnableLds = BEnableLds_auto || BEnableLds_manu || (NumPrefetch > 1);
static constexpr auto BEnableLds =
true; // BEnableLds_auto || BEnableLds_manu || (NumPrefetch > 1);
static constexpr auto matrix_padder = static constexpr auto matrix_padder =
MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock}; MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock};
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment