Argument Struct Reference

Argument Struct Reference#

Composable Kernel: ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument Struct Reference
Public Member Functions | Public Attributes | List of all members
ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument Struct Reference

#include <device_batched_gemm_xdl.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument:
ck::tensor_operation::device::BaseArgument

Public Member Functions

 Argument (const ADataType *p_a_grid_, const BDataType *p_b_grid_, CDataType *p_c_grid_, index_t M_, index_t N_, index_t K_, index_t StrideA_, index_t StrideB_, index_t StrideC_, index_t BatchStrideA, index_t BatchStrideB, index_t BatchStrideC, index_t Batch_)
Public Member Functions inherited from ck::tensor_operation::device::BaseArgument
 BaseArgument ()=default
 BaseArgument (const BaseArgument &)=default
BaseArgument & operator= (const BaseArgument &)=default
virtual ~BaseArgument ()

Public Attributes

const ADataType * p_a_grid
const BDataType * p_b_grid
CDataType * p_c_grid
index_t Batch
ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch
Public Attributes inherited from ck::tensor_operation::device::BaseArgument
void * p_workspace_ = nullptr

Constructor & Destructor Documentation

◆ Argument()

template<typename ADataType, typename BDataType, typename CDataType, typename AccDataType, typename ALayout, typename BLayout, typename CLayout, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector, ck::index_t NumGemmKPrefetchStage = 1, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument::Argument ( const ADataType * p_a_grid_,
const BDataType * p_b_grid_,
CDataType * p_c_grid_,
index_t M_,
index_t N_,
index_t K_,
index_t StrideA_,
index_t StrideB_,
index_t StrideC_,
index_t BatchStrideA,
index_t BatchStrideB,
index_t BatchStrideC,
index_t Batch_ )
inline

Member Data Documentation

◆ Batch

template<typename ADataType, typename BDataType, typename CDataType, typename AccDataType, typename ALayout, typename BLayout, typename CLayout, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector, ck::index_t NumGemmKPrefetchStage = 1, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
index_t ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument::Batch

◆ compute_ptr_offset_of_batch

template<typename ADataType, typename BDataType, typename CDataType, typename AccDataType, typename ALayout, typename BLayout, typename CLayout, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector, ck::index_t NumGemmKPrefetchStage = 1, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
ComputePtrOffsetOfStridedBatch ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument::compute_ptr_offset_of_batch

◆ p_a_grid

template<typename ADataType, typename BDataType, typename CDataType, typename AccDataType, typename ALayout, typename BLayout, typename CLayout, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector, ck::index_t NumGemmKPrefetchStage = 1, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
const ADataType* ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument::p_a_grid

◆ p_b_grid

template<typename ADataType, typename BDataType, typename CDataType, typename AccDataType, typename ALayout, typename BLayout, typename CLayout, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector, ck::index_t NumGemmKPrefetchStage = 1, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
const BDataType* ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument::p_b_grid

◆ p_c_grid

template<typename ADataType, typename BDataType, typename CDataType, typename AccDataType, typename ALayout, typename BLayout, typename CLayout, typename AElementwiseOperation, typename BElementwiseOperation, typename CElementwiseOperation, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t K0PerBlock, ck::index_t K1, ck::index_t MPerXDL, ck::index_t NPerXDL, ck::index_t MXdlPerWave, ck::index_t NXdlPerWave, typename ABlockTransferThreadClusterLengths_K0_M_K1, typename ABlockTransferThreadClusterArrangeOrder, typename ABlockTransferSrcAccessOrder, ck::index_t ABlockTransferSrcVectorDim, ck::index_t ABlockTransferSrcScalarPerVector, ck::index_t ABlockTransferDstScalarPerVector_K1, bool ABlockLdsAddExtraM, typename BBlockTransferThreadClusterLengths_K0_N_K1, typename BBlockTransferThreadClusterArrangeOrder, typename BBlockTransferSrcAccessOrder, ck::index_t BBlockTransferSrcVectorDim, ck::index_t BBlockTransferSrcScalarPerVector, ck::index_t BBlockTransferDstScalarPerVector_K1, bool BBlockLdsAddExtraN, ck::index_t CThreadTransferSrcDstVectorDim, ck::index_t CThreadTransferDstScalarPerVector, ck::index_t NumGemmKPrefetchStage = 1, ck::LoopScheduler LoopSched = make_default_loop_scheduler(), ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
CDataType* ck::tensor_operation::device::DeviceBatchedGemmXdl< ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout, AElementwiseOperation, BElementwiseOperation, CElementwiseOperation, BlockSize, MPerBlock, NPerBlock, K0PerBlock, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, BBlockLdsAddExtraN, CThreadTransferSrcDstVectorDim, CThreadTransferDstScalarPerVector, NumGemmKPrefetchStage, LoopSched, PipelineVer >::Argument::p_c_grid
The documentation for this struct was generated from the following file: