BatchedGemmKernelArgs Struct Reference

• ck_tile
• BatchedGemmKernel
• BatchedGemmKernelArgs

Public Attributes | List of all members

ck_tile::BatchedGemmKernel< TilePartitioner_, GemmPipeline_, EpiloguePipeline_ >::BatchedGemmKernelArgs Struct Reference

ALayout and ADataType are expected to be scalars, not a tuple. More...

#include <batched_gemm_kernel.hpp>

Inheritance diagram for ck_tile::BatchedGemmKernel< TilePartitioner_, GemmPipeline_, EpiloguePipeline_ >::BatchedGemmKernelArgs:

Public Attributes
index_t	batch_stride_A
index_t	batch_stride_B
index_t	batch_stride_E
index_t	batch_count
Public Attributes inherited from ck_tile::UniversalGemmKernelArgs< NumATensor, NumBTensor, NumDTensor >
const std::array< const void *, NumATensor >	as_ptr
	The As input tensor's pointer to device memory.
const std::array< const void *, NumBTensor >	bs_ptr
	The Bs input tensor's pointer to device memory.
const std::array< const void *, NumDTensor >	ds_ptr
	The Ds input tensor's pointer to device memory.
void *	e_ptr
	The E output tensor's pointer to device memory.
index_t	M
	GEMM's M dimension size.
index_t	N
	GEMM's N dimension size.
index_t	K
	GEMM's K dimension size.
std::array< index_t, NumATensor >	stride_As
	The distance between consecutive elements of non-contiguous dimension (in memory) of As tensor.
std::array< index_t, NumBTensor >	stride_Bs
	The distance between consecutive elements of non-contiguous dimension (in memory) of Bs tensor.
std::array< index_t, NumDTensor >	stride_Ds
	The distance between consecutive elements of non-contiguous dimension (in memory) of Ds tensor.
index_t	stride_E
	The distance between consecutive elements of non-contiguous dimension (in memory) of E tensor.
index_t	k_batch

Detailed Description

template<typename TilePartitioner_, typename GemmPipeline_, typename EpiloguePipeline_>
struct ck_tile::BatchedGemmKernel< TilePartitioner_, GemmPipeline_, EpiloguePipeline_ >::BatchedGemmKernelArgs

ALayout and ADataType are expected to be scalars, not a tuple.

BLayout and BDataType are expected to be scalars, not a tuple.

C/CLayout and C/EDataType are expected to be scalars, not a tuple.

Member Data Documentation

◆ batch_count

template<typename TilePartitioner_, typename GemmPipeline_, typename EpiloguePipeline_>

index_t ck_tile::BatchedGemmKernel< TilePartitioner_, GemmPipeline_, EpiloguePipeline_ >::BatchedGemmKernelArgs::batch_count

◆ batch_stride_A

template<typename TilePartitioner_, typename GemmPipeline_, typename EpiloguePipeline_>

index_t ck_tile::BatchedGemmKernel< TilePartitioner_, GemmPipeline_, EpiloguePipeline_ >::BatchedGemmKernelArgs::batch_stride_A

◆ batch_stride_B

template<typename TilePartitioner_, typename GemmPipeline_, typename EpiloguePipeline_>

index_t ck_tile::BatchedGemmKernel< TilePartitioner_, GemmPipeline_, EpiloguePipeline_ >::BatchedGemmKernelArgs::batch_stride_B

◆ batch_stride_E

template<typename TilePartitioner_, typename GemmPipeline_, typename EpiloguePipeline_>

index_t ck_tile::BatchedGemmKernel< TilePartitioner_, GemmPipeline_, EpiloguePipeline_ >::BatchedGemmKernelArgs::batch_stride_E

---

The documentation for this struct was generated from the following file:

• batched_gemm_kernel.hpp