DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector > Struct Template Reference

• ck
• tensor_operation
• device
• DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle

Classes | Public Types | Public Member Functions | Static Public Member Functions | Static Public Attributes | List of all members

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector > Struct Template Reference

#include <device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp>

Inheritance diagram for ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >:

Classes
struct	ActiveWorkgroupsPerCU
struct	Argument
struct	Invoker

Public Types
using	DeviceOp = DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle
using	ADataType = OutDataType
using	BDataType = InDataType
using	EDataType = WeiDataType
using	AElementwiseOperation = OutElementwiseOperation
using	BElementwiseOperation = InElementwiseOperation
using	CDEElementwiseOperation = WeiElementwiseOperation
using	ABDataType = InDataType
using	NGCHWTransposeDescType
using	NHWGCTransposeDescType
using	GKCYXTransposeDescType
using	GKYXCTransposeDescType
using	ABCGridDescs = decltype(GetABCGridDesc<NDimSpatial>())
using	AGridDesc_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>
using	BGridDesc_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>
using	CGridDesc_M_N = remove_cvref_t<decltype(ABCGridDescs{}[I2])>
using	CElementwiseGridDesc_M_N
template<index_t NXdlPerWave_>
using	GridwiseGemmBase
using	GridwiseGemm64 = GridwiseGemmBase<math::max(NXdlPerWave64, 1)>
using	GridwiseGemm32 = GridwiseGemmBase<NXdlPerWave32>
using	Block2TileMapElementwise = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>
using	GridwiseElementwiseCast
using	GridwiseElementwiseWeightTransposeCast
using	GridwiseElementwiseTranspose
using	CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock

Public Member Functions
bool	IsSupportedArgument (const BaseArgument *p_arg) override
std::unique_ptr< BaseArgument >	MakeArgumentPointer (const void *p_in_grid, void *p_wei_grid, const void *p_out_grid, const std::array< index_t, NDimSpatial+3 > &b_g_n_c_wis_lengths, const std::array< index_t, NDimSpatial+3 > &b_g_n_c_wis_strides, const std::array< index_t, NDimSpatial+3 > &e_g_k_c_xs_lengths, const std::array< index_t, NDimSpatial+3 > &e_g_k_c_xs_strides, const std::array< index_t, NDimSpatial+3 > &a_g_n_k_wos_lengths, const std::array< index_t, NDimSpatial+3 > &a_g_n_k_wos_strides, const std::array< ck::index_t, NDimSpatial > &conv_filter_strides, const std::array< ck::index_t, NDimSpatial > &conv_filter_dilations, const std::array< ck::index_t, NDimSpatial > &input_left_pads, const std::array< ck::index_t, NDimSpatial > &input_right_pads, InElementwiseOperation in_element_op, WeiElementwiseOperation wei_element_op, OutElementwiseOperation out_element_op, const ck::index_t split_k) override
std::unique_ptr< BaseInvoker >	MakeInvokerPointer () override
std::string	GetTypeString () const override
size_t	GetWorkSpaceSize (const BaseArgument *p_arg) const override
void	SetWorkSpacePointer (BaseArgument *p_arg, void *p_workspace, const StreamConfig &=StreamConfig{}) const override
Public Member Functions inherited from ck::tensor_operation::device::BaseOperator
	BaseOperator ()=default
	BaseOperator (const BaseOperator &)=default
BaseOperator &	operator= (const BaseOperator &)=default
virtual std::string	GetInstanceString () const
virtual std::string	GetTypeIdName () const
virtual std::optional< std::string >	GetObjectName () const
virtual std::optional< std::string >	GetTemplateInfo () const
virtual std::string	GetTypeIdHashCode () const
virtual	~BaseOperator ()

Static Public Member Functions
template<ck::index_t NDim, typename ck::enable_if< NDim==2, bool >::type = false>
static auto	GetABCGridDesc ()
template<ck::index_t NDim, typename ck::enable_if< NDim==3, bool >::type = false>
static auto	GetABCGridDesc ()
template<ck::index_t NDim, typename ck::enable_if< NDim==2, bool >::type = false>
static auto	GetElementwiseCGridDesc ()
template<ck::index_t NDim, typename ck::enable_if< NDim==3, bool >::type = false>
static auto	GetElementwiseCGridDesc ()
static constexpr bool	IsValidCompilationParameter ()
static bool	IsSupportedArgument (const Argument &arg)
static auto	MakeArgument (const InDataType *p_in_grid, WeiDataType *p_wei_grid, const OutDataType *p_out_grid, const std::array< index_t, NDimSpatial+3 > &b_g_n_c_wis_lengths, const std::array< index_t, NDimSpatial+3 > &b_g_n_c_wis_strides, const std::array< index_t, NDimSpatial+3 > &e_g_k_c_xs_lengths, const std::array< index_t, NDimSpatial+3 > &e_g_k_c_xs_strides, const std::array< index_t, NDimSpatial+3 > &a_g_n_k_wos_lengths, const std::array< index_t, NDimSpatial+3 > &a_g_n_k_wos_strides, const std::array< ck::index_t, NDimSpatial > &conv_filter_strides, const std::array< ck::index_t, NDimSpatial > &conv_filter_dilations, const std::array< ck::index_t, NDimSpatial > &input_left_pads, const std::array< ck::index_t, NDimSpatial > &input_right_pads, InElementwiseOperation in_element_op, WeiElementwiseOperation wei_element_op, OutElementwiseOperation out_element_op, const ck::index_t split_k)
static auto	MakeInvoker ()

Static Public Attributes
static GET_NXDL_PER_WAVE_IMPL constexpr auto	NXdlPerWave64 = GetNXdlPerWave<true>()
static constexpr auto	NXdlPerWave32 = GetNXdlPerWave<false>()
static constexpr auto	I0 = Number<0>{}
static constexpr auto	I1 = Number<1>{}
static constexpr auto	I2 = Number<2>{}
static constexpr auto	I3 = Number<3>{}
static constexpr auto	I4 = Number<4>{}
static constexpr auto	I5 = Number<5>{}
static constexpr auto	K1Number = Number<K1>{}
static constexpr auto	conv_to_gemm_transformer_v2
static constexpr auto	conv_to_gemm_transformer_v1
static constexpr index_t	ClusterLengthMPerBlock
static constexpr index_t	ClusterLengthNPerBlock
static constexpr auto	conv_ngchw_to_nhwgc_transformer
static constexpr GemmSpecialization	GemmSpec = GemmSpecialization::Default

Member Typedef Documentation

◆ ABCGridDescs

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::ABCGridDescs = decltype(GetABCGridDesc<NDimSpatial>())

◆ ABDataType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::ABDataType = InDataType

◆ ADataType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::ADataType = OutDataType

◆ AElementwiseOperation

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::AElementwiseOperation = OutElementwiseOperation

◆ AGridDesc_K0_M_K1

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::AGridDesc_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>

◆ BDataType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::BDataType = InDataType

◆ BElementwiseOperation

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::BElementwiseOperation = InElementwiseOperation

◆ BGridDesc_K0_N_K1

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::BGridDesc_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>

◆ Block2TileMapElementwise

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::Block2TileMapElementwise = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>

◆ CDEElementwiseOperation

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::CDEElementwiseOperation = WeiElementwiseOperation

◆ CElementwiseGridDesc_M_N

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::CElementwiseGridDesc_M_N

Initial value:

remove_cvref_t<decltype(GetElementwiseCGridDesc<NDimSpatial>())>

ck::remove_cvref_t

remove_cv_t< remove_reference_t< T > > remove_cvref_t

Definition type.hpp:297

◆ CGridDesc_M_N

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::CGridDesc_M_N = remove_cvref_t<decltype(ABCGridDescs{}[I2])>

◆ CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock

Initial value:

decltype(GridwiseGemm64::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(

CGridDesc_M_N{}, 1, 1))

ck::GridwiseGemm_xdl_cshuffle_conv_v3< tensor_layout::gemm::RowMajor, tensor_layout::gemm::ColumnMajor, tensor_layout::gemm::RowMajor, ADataType, BDataType, AccDataType, AccDataType, AccDataType, AElementwiseOperation, BElementwiseOperation, CDEElementwiseOperation, GemmSpec, BlockSize, MPerBlock, NPerBlock, KPerBlock, K1, K1, MPerXDL, NPerXDL, MXdlPerWave, NXdlPerWave_, ABlockTransferThreadClusterLengths_K0_M_K1, ABlockTransferThreadClusterArrangeOrder, ABlockTransferSrcAccessOrder, ABlockTransferSrcVectorDim, ABlockTransferSrcScalarPerVector, ABlockTransferDstScalarPerVector_K1, false, ABlockLdsAddExtraM, BBlockTransferThreadClusterLengths_K0_N_K1, BBlockTransferThreadClusterArrangeOrder, BBlockTransferSrcAccessOrder, BBlockTransferSrcVectorDim, BBlockTransferSrcScalarPerVector, BBlockTransferDstScalarPerVector_K1, false, BBlockLdsAddExtraN, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, ComputeTypeA, ComputeTypeB >::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock

__host__ static __device__ constexpr auto MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc &c_grid_desc_m_n, index_t MBlock, index_t NBlock)

Definition gridwise_gemm_xdl_cshuffle_conv_v3.hpp:644

ck::tensor_operation::device::DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K::CGridDesc_M_N

remove_cvref_t< decltype(ABCGridDescs{}[I2])> CGridDesc_M_N

Definition device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp:285

◆ DeviceOp

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::DeviceOp = DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle

◆ EDataType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::EDataType = WeiDataType

◆ GKCYXTransposeDescType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GKCYXTransposeDescType

Initial value:

remove_cvref_t<decltype(conv_ngchw_to_nhwgc_transformer

.template MakeGKCYXTransposeDesc<NDimSpatial>({}, {}))>

ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1::conv_ngchw_to_nhwgc_transformer

static constexpr auto conv_ngchw_to_nhwgc_transformer

Definition device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp:583

◆ GKYXCTransposeDescType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GKYXCTransposeDescType

Initial value:

remove_cvref_t<decltype(conv_ngchw_to_nhwgc_transformer

.template MakeGKYXCTransposeDesc<NDimSpatial>({}, {}))>

◆ GridwiseElementwiseCast

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GridwiseElementwiseCast

Initial value:

GridwiseElementwise<Tuple<CElementwiseGridDesc_M_N>,

Tuple<CElementwiseGridDesc_M_N>,

Tuple<const AccDataType*>,

Tuple<EDataType*>,

Block2TileMapElementwise,

CDEElementwiseOperation,

BlockSize,

MPerBlock,

NPerBlock,

MPerBlock / ClusterLengthMPerBlock,

NPerBlock / ClusterLengthNPerBlock,

Sequence<0, 1>,

Sequence<CBlockTransferScalarPerVector_NWaveNPerXdl>,

Sequence<CBlockTransferScalarPerVector_NWaveNPerXdl>,

I1,

I1>

ck::GridwiseElementwise

Definition gridwise_elementwise_2d.hpp:278

ck::Sequence

Definition utility/sequence.hpp:43

ck::Tuple

Definition utility/tuple.hpp:117

ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Explicit_Xdl::I1

static constexpr auto I1

Definition device_grouped_conv_bwd_weight_explicit_xdl.hpp:52

ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Explicit_Xdl::Block2TileMapElementwise

BlockToCTileMap_M00_N0_M01Adapt< 1, ElemsPerBlock > Block2TileMapElementwise

Definition device_grouped_conv_bwd_weight_explicit_xdl.hpp:80

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle::CDEElementwiseOperation

WeiElementwiseOperation CDEElementwiseOperation

Definition device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp:238

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle::ClusterLengthMPerBlock

static constexpr index_t ClusterLengthMPerBlock

Definition device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp:269

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle::ClusterLengthNPerBlock

static constexpr index_t ClusterLengthNPerBlock

Definition device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp:271

◆ GridwiseElementwiseTranspose

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GridwiseElementwiseTranspose

Initial value:

GridwiseElementwise<Tuple<NGCHWTransposeDescType>,

Tuple<NHWGCTransposeDescType>,

Tuple<const ADataType*>,

Tuple<ADataType*>,

Block2TileMapElementwise,

element_wise::PassThrough,

BlockSize,

MPerBlock,

NPerBlock,

MPerBlock / ClusterLengthMPerBlock,

NPerBlock / ClusterLengthNPerBlock,

Sequence<1, 0>,

Sequence<TransposeTransferSrcScalarPerVector>,

Sequence<TransposeTransferDstScalarPerVector>,

I1,

I0>

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle::I1

static constexpr auto I1

Definition device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp:244

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle::Block2TileMapElementwise

BlockToCTileMap_M00_N0_M01Adapt< MPerBlock, NPerBlock > Block2TileMapElementwise

Definition device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp:456

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle::I0

static constexpr auto I0

Definition device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp:243

ck::tensor_operation::element_wise::PassThrough

Definition tensor_operation/gpu/element/unary_element_wise_operation.hpp:340

◆ GridwiseElementwiseWeightTransposeCast

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GridwiseElementwiseWeightTransposeCast

Initial value:

GridwiseElementwise<Tuple<GKYXCTransposeDescType>,

Tuple<GKCYXTransposeDescType>,

Tuple<const AccDataType*>,

Tuple<EDataType*>,

Block2TileMapElementwise,

CDEElementwiseOperation,

BlockSize,

MPerBlock,

NPerBlock,

MPerBlock / ClusterLengthMPerBlock,

NPerBlock / ClusterLengthNPerBlock,

Sequence<0, 1>,

Sequence<CBlockTransferScalarPerVector_NWaveNPerXdl>,

Sequence<1>,

I1,

I0>

◆ GridwiseGemm32

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GridwiseGemm32 = GridwiseGemmBase<NXdlPerWave32>

◆ GridwiseGemm64

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GridwiseGemm64 = GridwiseGemmBase<math::max(NXdlPerWave64, 1)>

◆ GridwiseGemmBase

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

template<index_t NXdlPerWave_>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GridwiseGemmBase

◆ NGCHWTransposeDescType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::NGCHWTransposeDescType

Initial value:

remove_cvref_t<decltype(conv_ngchw_to_nhwgc_transformer

.template MakeNGCHWTransposeDesc<NDimSpatial>({}, {}))>

◆ NHWGCTransposeDescType

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

using ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::NHWGCTransposeDescType

Initial value:

remove_cvref_t<decltype(conv_ngchw_to_nhwgc_transformer

.template MakeNHWGCTransposeDesc<NDimSpatial>({}, {}))>

Member Function Documentation

◆ GetABCGridDesc() [1/2]

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

template<ck::index_t NDim, typename ck::enable_if< NDim==3, bool >::type = false>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GetABCGridDesc ( )

inlinestatic

◆ GetABCGridDesc() [2/2]

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

template<ck::index_t NDim, typename ck::enable_if< NDim==2, bool >::type = false>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GetABCGridDesc ( )

inlinestatic

◆ GetElementwiseCGridDesc() [1/2]

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

template<ck::index_t NDim, typename ck::enable_if< NDim==3, bool >::type = false>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GetElementwiseCGridDesc ( )

inlinestatic

◆ GetElementwiseCGridDesc() [2/2]

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

template<ck::index_t NDim, typename ck::enable_if< NDim==2, bool >::type = false>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GetElementwiseCGridDesc ( )

inlinestatic

◆ GetTypeString()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

std::string ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GetTypeString ( ) const

inlineoverridevirtual

Reimplemented from ck::tensor_operation::device::BaseOperator.

◆ GetWorkSpaceSize()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

size_t ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GetWorkSpaceSize ( const BaseArgument * p_arg ) const

inlineoverridevirtual

Reimplemented from ck::tensor_operation::device::BaseOperator.

◆ IsSupportedArgument() [1/2]

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

bool ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::IsSupportedArgument ( const Argument & arg )

inlinestatic

◆ IsSupportedArgument() [2/2]

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

bool ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::IsSupportedArgument ( const BaseArgument * p_arg )

inlineoverridevirtual

Reimplemented from ck::tensor_operation::device::BaseOperator.

◆ IsValidCompilationParameter()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

constexpr bool ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::IsValidCompilationParameter ( )

inlinestaticconstexpr

◆ MakeArgument()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::MakeArgument ( const InDataType * p_in_grid, WeiDataType * p_wei_grid, const OutDataType * p_out_grid, const std::array< index_t, NDimSpatial+3 > & b_g_n_c_wis_lengths, const std::array< index_t, NDimSpatial+3 > & b_g_n_c_wis_strides, const std::array< index_t, NDimSpatial+3 > & e_g_k_c_xs_lengths, const std::array< index_t, NDimSpatial+3 > & e_g_k_c_xs_strides, const std::array< index_t, NDimSpatial+3 > & a_g_n_k_wos_lengths, const std::array< index_t, NDimSpatial+3 > & a_g_n_k_wos_strides, const std::array< ck::index_t, NDimSpatial > & conv_filter_strides, const std::array< ck::index_t, NDimSpatial > & conv_filter_dilations, const std::array< ck::index_t, NDimSpatial > & input_left_pads, const std::array< ck::index_t, NDimSpatial > & input_right_pads, InElementwiseOperation in_element_op, WeiElementwiseOperation wei_element_op, OutElementwiseOperation out_element_op, const ck::index_t split_k )

inlinestatic

◆ MakeArgumentPointer()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

std::unique_ptr< BaseArgument > ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::MakeArgumentPointer ( const void * p_in_grid, void * p_wei_grid, const void * p_out_grid, const std::array< index_t, NDimSpatial+3 > & b_g_n_c_wis_lengths, const std::array< index_t, NDimSpatial+3 > & b_g_n_c_wis_strides, const std::array< index_t, NDimSpatial+3 > & e_g_k_c_xs_lengths, const std::array< index_t, NDimSpatial+3 > & e_g_k_c_xs_strides, const std::array< index_t, NDimSpatial+3 > & a_g_n_k_wos_lengths, const std::array< index_t, NDimSpatial+3 > & a_g_n_k_wos_strides, const std::array< ck::index_t, NDimSpatial > & conv_filter_strides, const std::array< ck::index_t, NDimSpatial > & conv_filter_dilations, const std::array< ck::index_t, NDimSpatial > & input_left_pads, const std::array< ck::index_t, NDimSpatial > & input_right_pads, InElementwiseOperation in_element_op, WeiElementwiseOperation wei_element_op, OutElementwiseOperation out_element_op, const ck::index_t split_k )

inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceGroupedConvBwdWeight< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, InDataType, InDataType >.

◆ MakeInvoker()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::MakeInvoker ( )

inlinestatic

◆ MakeInvokerPointer()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

std::unique_ptr< BaseInvoker > ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::MakeInvokerPointer ( )

inlineoverridevirtual

Implements ck::tensor_operation::device::DeviceGroupedConvBwdWeight< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, InDataType, InDataType >.

◆ SetWorkSpacePointer()

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

void ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::SetWorkSpacePointer ( BaseArgument * p_arg, void * p_workspace, const StreamConfig & = StreamConfig{} ) const

inlineoverridevirtual

Reimplemented from ck::tensor_operation::device::BaseOperator.

Member Data Documentation

◆ ClusterLengthMPerBlock

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

index_t ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::ClusterLengthMPerBlock

staticconstexpr

Initial value:

=

CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock::At(1)

◆ ClusterLengthNPerBlock

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

index_t ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::ClusterLengthNPerBlock

staticconstexpr

Initial value:

=

CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock::At(3)

◆ conv_ngchw_to_nhwgc_transformer

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::conv_ngchw_to_nhwgc_transformer

staticconstexpr

Initial value:

=

TransformConvNGCHWToNHWGC<InLayout,

WeiLayout,

OutLayout,

NDimSpatial,

MPerBlock / ClusterLengthMPerBlock,

NPerBlock / ClusterLengthNPerBlock>{}

ck::tensor_operation::TransformConvNGCHWToNHWGC

Definition transform_conv_ngchw_to_nhwgc.hpp:31

ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1::ClusterLengthNPerBlock

static constexpr index_t ClusterLengthNPerBlock

Definition device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp:580

ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1::ClusterLengthMPerBlock

static constexpr index_t ClusterLengthMPerBlock

Definition device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp:578

◆ conv_to_gemm_transformer_v1

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::conv_to_gemm_transformer_v1

staticconstexpr

Initial value:

=

TransformConvBwdWeightToGemm<NDimSpatial,

MPerBlock,

NPerBlock,

K1Number,

KPerBlock / K1Number,

ConvBackwardWeightSpecialization>{}

ck::tensor_operation::TransformConvBwdWeightToGemm

Definition tensor_operation/operator_transform/transform_conv_bwd_weight_to_gemm.hpp:24

ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle::K1Number

static constexpr auto K1Number

Definition device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp:250

◆ conv_to_gemm_transformer_v2

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::conv_to_gemm_transformer_v2

staticconstexpr

Initial value:

=

TransformConvBwdWeightToGemmV2<NDimSpatial,

MPerBlock,

NPerBlock,

K1Number,

KPerBlock / K1Number,

NumGroupsToMerge,

ConvBackwardWeightSpecialization>{}

ck::tensor_operation::TransformConvBwdWeightToGemmV2

Transform conv bwd weight to gemm v2.

Definition transform_conv_bwd_weight_to_gemm_v2.hpp:33

◆ GemmSpec

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

GemmSpecialization ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::GemmSpec = GemmSpecialization::Default

staticconstexpr

◆ I0

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::I0 = Number<0>{}

staticconstexpr

◆ I1

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::I1 = Number<1>{}

staticconstexpr

◆ I2

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::I2 = Number<2>{}

staticconstexpr

◆ I3

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::I3 = Number<3>{}

staticconstexpr

◆ I4

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::I4 = Number<4>{}

staticconstexpr

◆ I5

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::I5 = Number<5>{}

staticconstexpr

◆ K1Number

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::K1Number = Number<K1>{}

staticconstexpr

◆ NXdlPerWave32

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::NXdlPerWave32 = GetNXdlPerWave<false>()

staticconstexpr

◆ NXdlPerWave64

template<ck::index_t NDimSpatial, typename InLayout, typename WeiLayout, typename OutLayout, typename InDataType, typename WeiDataType, typename OutDataType, typename AccDataType, typename InElementwiseOperation, typename WeiElementwiseOperation, typename OutElementwiseOperation, ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization, ck::index_t BlockSize, ck::index_t MPerBlock, ck::index_t NPerBlock, ck::index_t KPerBlock, 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, index_t CShuffleMXdlPerWavePerShuffle, index_t CShuffleNXdlPerWavePerShuffle, typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, index_t CBlockTransferScalarPerVector_NWaveNPerXdl, BlockGemmPipelineScheduler BlkGemmPipeSched = BlockGemmPipelineScheduler::Intrawave, BlockGemmPipelineVersion BlkGemmPipelineVer = BlockGemmPipelineVersion::v1, index_t NumGroupsToMerge = 1, typename ComputeTypeA = InDataType, typename ComputeTypeB = ComputeTypeA, index_t TransposeTransferSrcScalarPerVector = 1, index_t TransposeTransferDstScalarPerVector = 1>

GET_NXDL_PER_WAVE_IMPL constexpr auto ck::tensor_operation::device::DeviceGroupedConvBwdWeightTwoStage_Xdl_CShuffle< NDimSpatial, InLayout, WeiLayout, OutLayout, InDataType, WeiDataType, OutDataType, AccDataType, InElementwiseOperation, WeiElementwiseOperation, OutElementwiseOperation, ConvBackwardWeightSpecialization, BlockSize, MPerBlock, NPerBlock, KPerBlock, 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, CShuffleMXdlPerWavePerShuffle, CShuffleNXdlPerWavePerShuffle, CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock, CBlockTransferScalarPerVector_NWaveNPerXdl, BlkGemmPipeSched, BlkGemmPipelineVer, NumGroupsToMerge, ComputeTypeA, ComputeTypeB, TransposeTransferSrcScalarPerVector, TransposeTransferDstScalarPerVector >::NXdlPerWave64 = GetNXdlPerWave<true>()

staticconstexpr

---

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

• device_grouped_conv_bwd_weight_two_stage_xdl_cshuffle.hpp