ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-13 10:37:42 +00:00
Code Issues Packages Projects Releases Wiki Activity

change to static

Browse Source
This commit is contained in:
coderfeli
2024-12-04 08:47:59 +00:00
parent 928b6d1a56
commit 99c8123f48
2 changed files with 10 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck_tile/ops/gemm/kernel/gemm_kernel.hpp
View File

@@ -168,7 +168,7 @@ struct GemmKernel
// Run GEMM cooperatively by whole wokrgroup.
auto c_block_tile =
GemmPipeline{}.template operator()(a_block_window, b_block_window, num_loop, smem_ptr_0, smem_ptr_1);
GemmPipeline::template run(a_block_window, b_block_window, num_loop, smem_ptr_0, smem_ptr_1);
CDataType* c_start = static_cast<CDataType*>(kargs.c_ptr);
auto c_tensor_view = [&]() {

18
include/ck_tile/ops/gemm/pipeline/gemm_pipeline_agmem_bgmem_creg_v1.hpp
View File

@@ -57,17 +57,17 @@ struct GemmPipelineAGmemBGmemCRegV1
}
template <typename DstBlockTile, typename SrcTileWindow>
CK_TILE_DEVICE void GlobalPrefetch(DstBlockTile& dst_block_tile,
SrcTileWindow& dram_tile_window) const
CK_TILE_DEVICE static void GlobalPrefetch(DstBlockTile& dst_block_tile,
SrcTileWindow& dram_tile_window)
{
load_tile(dst_block_tile, dram_tile_window);
move_tile_window(dram_tile_window, {0, kKPerBlock});
}
template <typename DstTileWindow, typename SrcBlockTile, typename ElementFunction>
CK_TILE_DEVICE void LocalPrefill(DstTileWindow& lds_tile_window,
CK_TILE_DEVICE static void LocalPrefill(DstTileWindow& lds_tile_window,
const SrcBlockTile& src_block_tile,
const ElementFunction& element_func) const
const ElementFunction& element_func)
{
const auto block_tile_tmp = tile_elementwise_in(element_func, src_block_tile);
store_tile(lds_tile_window, block_tile_tmp);
@@ -160,13 +160,13 @@ struct GemmPipelineAGmemBGmemCRegV1
typename BDramBlockWindowTmp,
typename AElementFunction,
typename BElementFunction>
CK_TILE_HOST_DEVICE auto operator()(const ADramBlockWindowTmp& a_dram_block_window_tmp,
CK_TILE_DEVICE static auto run(const ADramBlockWindowTmp& a_dram_block_window_tmp,
const AElementFunction& a_element_func,
const BDramBlockWindowTmp& b_dram_block_window_tmp,
const BElementFunction& b_element_func,
index_t num_loop,
void* __restrict__ p_smem_0,
void* __restrict__ p_smem_1) const
void* __restrict__ p_smem_1)
{
static_assert(
std::is_same_v<ADataType, remove_cvref_t<typename ADramBlockWindowTmp::DataType>> &&
@@ -384,13 +384,13 @@ struct GemmPipelineAGmemBGmemCRegV1
}
template <typename ADramBlockWindowTmp, typename BDramBlockWindowTmp>
CK_TILE_DEVICE auto operator()(const ADramBlockWindowTmp& a_dram_block_window_tmp,
CK_TILE_DEVICE static auto run(const ADramBlockWindowTmp& a_dram_block_window_tmp,
const BDramBlockWindowTmp& b_dram_block_window_tmp,
index_t num_loop,
void*__restrict__ p_smem_0,
void*__restrict__ p_smem_1) const
void*__restrict__ p_smem_1)
{
return operator()(
return run(
a_dram_block_window_tmp,
[](const ADataType& a) { return a; },
b_dram_block_window_tmp,