ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 13:11:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Ck tile gemm padding dim (#1516)

Browse Source 
* Support the N dimension padding

* Finished the padding feature for different dimension of K
This commit is contained in:
Thomas Ning
2024-09-18 11:32:29 -07:00
committed by GitHub
parent e84adec3ba
commit 694c300145
4 changed files with 33 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
include/ck_tile/ops/gemm/kernel/gemm_kernel.hpp
View File

@@ -123,14 +123,26 @@ struct GemmKernel
}
}();
auto ABlockWindow = make_tile_window(
auto a_pad_view = pad_tensor_view(
a_tensor_view,
make_tuple(number<TilePartitioner::kM>{}, number<TilePartitioner::kK>{}),
sequence < 0,
GemmPipeline::kPadA ? 1 : 0 > {});
auto ABlockWindow = make_tile_window(
a_pad_view,
make_tuple(number<TilePartitioner::kM>{}, number<TilePartitioner::kK>{}),
{i_m, 0});
auto BBlockWindow = make_tile_window(
auto b_pad_view = pad_tensor_view(
b_tensor_view,
make_tuple(number<TilePartitioner::kN>{}, number<TilePartitioner::kK>{}),
sequence < 0,
GemmPipeline::kPadB ? 1 : 0 > {});
auto BBlockWindow = make_tile_window(
b_pad_view,
make_tuple(number<TilePartitioner::kN>{}, number<TilePartitioner::kK>{}),
{i_n, 0});
// allocate LDS
@@ -163,12 +175,16 @@ struct GemmKernel
}
}();
auto CBlockWindow = make_tile_window(
auto c_pad_view = pad_tensor_view(
c_tensor_view,
make_tuple(number<TilePartitioner::kM>{}, number<TilePartitioner::kN>{}),
sequence < 0,
GemmPipeline::kPadC ? 1 : 0 > {});
auto CBlockWindow_pad = make_tile_window(
c_pad_view,
make_tuple(number<TilePartitioner::kM>{}, number<TilePartitioner::kN>{}),
{i_m, i_n});
// epilogue.
EpiloguePipeline{}(CBlockWindow, acc);
EpiloguePipeline{}(CBlockWindow_pad, acc);
}
};

4
include/ck_tile/ops/gemm/pipeline/block_gemm_pipeline_agmem_bgmem_creg_v1.hpp
View File

@@ -29,6 +29,10 @@ struct BlockGemmPipelineAGmemBGmemCRegV1
static constexpr index_t AlignmentB = Problem::AlignmentB;
static constexpr index_t AlignmentC = Problem::AlignmentC;
static constexpr bool kPadA = Problem::kPadA;
static constexpr bool kPadB = Problem::kPadB;
static constexpr bool kPadC = Problem::kPadC;
CK_TILE_HOST_DEVICE static constexpr ck_tile::index_t GetStaticLdsSize()
{
return ck_tile::integer_divide_ceil(

6
include/ck_tile/ops/gemm/pipeline/block_gemm_pipeline_problem.hpp
View File

@@ -28,9 +28,9 @@ struct BlockGemmPipelineProblem
static constexpr bool kPadB = kPadB_;
static constexpr bool kPadC = kPadC_;
static constexpr index_t AlignmentA = kPadA ? VectorLoadSize / sizeof(ADataType) : 1;
static constexpr index_t AlignmentB = kPadB ? VectorLoadSize / sizeof(BDataType) : 1;
static constexpr index_t AlignmentC = kPadC ? VectorLoadSize / sizeof(CDataType) : 1;
static constexpr index_t AlignmentA = kPadA ? 1 : VectorLoadSize / sizeof(ADataType);
static constexpr index_t AlignmentB = kPadB ? 1 : VectorLoadSize / sizeof(BDataType);
static constexpr index_t AlignmentC = kPadC ? 1 : VectorLoadSize / sizeof(CDataType);
};
} // namespace ck_tile